2019-02-28 21:21:29 +00:00
|
|
|
# beacon_chain
|
2022-02-20 20:13:06 +00:00
|
|
|
# Copyright (c) 2018-2022 Status Research & Development GmbH
|
2019-02-28 21:21:29 +00:00
|
|
|
# Licensed and distributed under either of
|
2019-11-25 15:30:02 +00:00
|
|
|
# * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
|
|
|
|
|
# * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
|
2019-02-28 21:21:29 +00:00
|
|
|
# at your option. This file may not be copied, modified, or distributed except according to those terms.
|
|
|
|
|
|
2019-11-14 10:47:55 +00:00
|
|
|
{.used.}
|
|
|
|
|
|
2019-02-28 21:21:29 +00:00
|
|
|
import
|
2021-04-28 16:41:02 +00:00
|
|
|
unittest2,
|
2021-09-17 00:13:52 +00:00
|
|
|
eth/keys, taskpools,
|
2021-06-21 08:35:24 +00:00
|
|
|
../beacon_chain/spec/datatypes/base,
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
../beacon_chain/spec/[beaconstate, forks, helpers, signatures, state_transition],
|
2021-08-18 18:57:58 +00:00
|
|
|
../beacon_chain/[beacon_chain_db],
|
2021-04-16 08:49:37 +00:00
|
|
|
../beacon_chain/consensus_object_pools/[
|
2021-09-08 03:46:33 +00:00
|
|
|
attestation_pool, blockchain_dag, block_quarantine, block_clearance],
|
2021-04-28 16:41:02 +00:00
|
|
|
./testutil, ./testdbutil, ./testblockutil
|
2019-02-28 21:21:29 +00:00
|
|
|
|
2021-12-09 17:06:21 +00:00
|
|
|
func `$`(x: BlockRef): string = shortLog(x)
|
2020-10-14 20:23:04 +00:00
|
|
|
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
const
|
|
|
|
|
nilPhase0Callback = OnPhase0BlockAdded(nil)
|
|
|
|
|
nilAltairCallback = OnAltairBlockAdded(nil)
|
|
|
|
|
|
2021-03-09 14:36:17 +00:00
|
|
|
proc pruneAtFinalization(dag: ChainDAGRef) =
|
|
|
|
|
if dag.needStateCachesAndForkChoicePruning():
|
|
|
|
|
dag.pruneStateCachesDAG()
|
|
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
suite "Block pool processing" & preset():
|
2020-05-05 09:18:44 +00:00
|
|
|
setup:
|
|
|
|
|
var
|
|
|
|
|
db = makeTestDB(SLOTS_PER_EPOCH)
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor, {})
|
2021-12-06 09:49:01 +00:00
|
|
|
verifier = BatchVerifier(rng: keys.newRng(), taskpool: Taskpool.new())
|
|
|
|
|
quarantine = Quarantine.init()
|
2022-03-16 07:20:40 +00:00
|
|
|
state = newClone(dag.headState)
|
2020-07-15 10:44:18 +00:00
|
|
|
cache = StateCache()
|
2021-10-13 14:24:36 +00:00
|
|
|
info = ForkedEpochInfo()
|
2021-05-21 09:23:28 +00:00
|
|
|
att0 = makeFullAttestations(state[], dag.tail.root, 0.Slot, cache)
|
2021-11-18 12:02:43 +00:00
|
|
|
b1 = addTestBlock(state[], cache, attestations = att0).phase0Data
|
|
|
|
|
b2 = addTestBlock(state[], cache).phase0Data
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
test "basic ops":
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.getBlockRef(default Eth2Digest).isNone()
|
2020-04-10 14:06:24 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
b0 = dag.getForkedBlock(dag.tail.root)
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
bh = dag.getForkedBlock(dag.head.root)
|
|
|
|
|
bh2 = dag.getForkedBlock(dag.head.bid)
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
|
|
|
|
b0.isSome()
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
bh.isSome()
|
|
|
|
|
bh2.isSome()
|
|
|
|
|
|
|
|
|
|
dag.getBlockRef(dag.finalizedHead.blck.root).get() ==
|
|
|
|
|
dag.finalizedHead.blck
|
|
|
|
|
dag.getBlockRef(dag.head.root).get() == dag.head
|
2020-05-04 21:07:18 +00:00
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
test "Simple block add&get" & preset():
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
b1Get = dag.getForkedBlock(b1.root)
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
|
|
|
|
b1Get.isSome()
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
b1Get.get().root == b1.root
|
|
|
|
|
b1Add[].root == b1Get.get().root
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.heads.len == 1
|
|
|
|
|
dag.heads[0] == b1Add[]
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b2Add = dag.addHeadBlock(verifier, b2, nilPhase0Callback)
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
b2Get = dag.getForkedBlock(b2.root)
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
er = dag.findEpochRef(b1Add[].bid, b1Add[].slot.epoch)
|
2022-03-16 07:20:40 +00:00
|
|
|
validators = getStateField(dag.headState, validators).lenu64()
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
|
|
|
|
b2Get.isSome()
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
b2Get.get().root == b2.root
|
|
|
|
|
b2Add[].root == b2Get.get().root
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.heads.len == 1
|
|
|
|
|
dag.heads[0] == b2Add[]
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.containsForkBlock(b2.root)
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.parent(b2Add[].bid).get() == b1Add[].bid
|
|
|
|
|
# head not updated yet - getBlockIdAtSlot won't give those blocks
|
|
|
|
|
dag.getBlockIdAtSlot(b2Add[].slot).get() ==
|
|
|
|
|
BlockSlotId.init(dag.genesis, b2Add[].slot)
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
|
2022-01-05 18:38:04 +00:00
|
|
|
not er.isErr()
|
2021-06-10 07:37:02 +00:00
|
|
|
# Same epoch - same epochRef
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
er[] == dag.findEpochRef(b2Add[].bid, b2Add[].slot.epoch)[]
|
2021-06-10 07:37:02 +00:00
|
|
|
# Different epoch that was never processed
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.findEpochRef(b1Add[].bid, b1Add[].slot.epoch + 1).isErr()
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2022-01-05 18:38:04 +00:00
|
|
|
er[].validatorKey(0'u64).isSome()
|
|
|
|
|
er[].validatorKey(validators - 1).isSome()
|
|
|
|
|
er[].validatorKey(validators).isNone()
|
2021-06-10 07:37:02 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
# Skip one slot to get a gap
|
2020-05-19 15:46:29 +00:00
|
|
|
check:
|
2021-06-11 17:51:46 +00:00
|
|
|
process_slots(
|
Implement split preset/config support (#2710)
* Implement split preset/config support
This is the initial bulk refactor to introduce runtime config values in
a number of places, somewhat replacing the existing mechanism of loading
network metadata.
It still needs more work, this is the initial refactor that introduces
runtime configuration in some of the places that need it.
The PR changes the way presets and constants work, to match the spec. In
particular, a "preset" now refers to the compile-time configuration
while a "cfg" or "RuntimeConfig" is the dynamic part.
A single binary can support either mainnet or minimal, but not both.
Support for other presets has been removed completely (can be readded,
in case there's need).
There's a number of outstanding tasks:
* `SECONDS_PER_SLOT` still needs fixing
* loading custom runtime configs needs redoing
* checking constants against YAML file
* yeerongpilly support
`build/nimbus_beacon_node --network=yeerongpilly --discv5:no --log-level=DEBUG`
* load fork epoch from config
* fix fork digest sent in status
* nicer error string for request failures
* fix tools
* one more
* fixup
* fixup
* fixup
* use "standard" network definition folder in local testnet
Files are loaded from their standard locations, including genesis etc,
to conform to the format used in the `eth2-networks` repo.
* fix launch scripts, allow unknown config values
* fix base config of rest test
* cleanups
* bundle mainnet config using common loader
* fix spec links and names
* only include supported preset in binary
* drop yeerongpilly, add altair-devnet-0, support boot_enr.yaml
2021-07-12 13:01:38 +00:00
|
|
|
defaultRuntimeConfig, state[], getStateField(state[], slot) + 1, cache,
|
2022-01-17 11:19:58 +00:00
|
|
|
info, {}).isOk()
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
2021-11-18 12:02:43 +00:00
|
|
|
b4 = addTestBlock(state[], cache).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b4Add = dag.addHeadBlock(verifier, b4, nilPhase0Callback)
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
2020-07-09 09:29:32 +00:00
|
|
|
b4Add[].parent == b2Add[]
|
2019-12-19 14:13:35 +00:00
|
|
|
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(b4Add[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-04-21 06:43:39 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
check: # getBlockIdAtSlot operates on the head chain!
|
|
|
|
|
dag.getBlockIdAtSlot(b2Add[].slot).get() ==
|
|
|
|
|
BlockSlotId.init(b2Add[].bid, b2Add[].slot)
|
|
|
|
|
dag.parentOrSlot(dag.getBlockIdAtSlot(b2Add[].slot).get()).get() ==
|
|
|
|
|
BlockSlotId.init(b1Add[].bid, b2Add[].slot)
|
|
|
|
|
dag.parentOrSlot(dag.getBlockIdAtSlot(b2Add[].slot + 1).get()).get() ==
|
|
|
|
|
BlockSlotId.init(b2Add[].bid, b2Add[].slot)
|
|
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
var blocks: array[3, BlockId]
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(0), 1, blocks.toOpenArray(0, 0)) == 0
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
blocks[0..<1] == [dag.tail]
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(0), 1, blocks.toOpenArray(0, 1)) == 0
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
blocks[0..<2] == [dag.tail, b1Add[].bid]
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(0), 2, blocks.toOpenArray(0, 1)) == 0
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
blocks[0..<2] == [dag.tail, b2Add[].bid]
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(0), 3, blocks.toOpenArray(0, 1)) == 1
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
blocks[1..<2] == [dag.tail] # block 3 is missing!
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(2), 2, blocks.toOpenArray(0, 1)) == 0
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks[0..<2] == [b2Add[].bid, b4Add[].bid] # block 3 is missing!
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-10-14 20:23:04 +00:00
|
|
|
# large skip step
|
|
|
|
|
dag.getBlockRange(Slot(0), uint64.high, blocks.toOpenArray(0, 2)) == 2
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
blocks[2..2] == [dag.tail]
|
2020-10-14 20:23:04 +00:00
|
|
|
|
|
|
|
|
# large skip step
|
|
|
|
|
dag.getBlockRange(Slot(2), uint64.high, blocks.toOpenArray(0, 1)) == 1
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks[1..1] == [b2Add[].bid]
|
2020-10-14 20:23:04 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
# empty length
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(2), 2, blocks.toOpenArray(0, -1)) == 0
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
# No blocks in sight
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(5), 1, blocks.toOpenArray(0, 1)) == 2
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2020-10-14 20:23:04 +00:00
|
|
|
# No blocks in sight
|
|
|
|
|
dag.getBlockRange(Slot(uint64.high), 1, blocks.toOpenArray(0, 1)) == 2
|
|
|
|
|
|
2020-05-05 09:18:44 +00:00
|
|
|
# No blocks in sight either due to gaps
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.getBlockRange(Slot(3), 2, blocks.toOpenArray(0, 1)) == 2
|
2020-08-05 23:22:12 +00:00
|
|
|
blocks[2..<2].len == 0
|
2020-04-21 06:43:39 +00:00
|
|
|
|
2021-12-30 11:33:03 +00:00
|
|
|
# A fork forces the clearance state to a point where it cannot be advanced
|
|
|
|
|
let
|
|
|
|
|
nextEpoch = dag.head.slot.epoch + 1
|
2022-01-11 10:01:54 +00:00
|
|
|
nextEpochSlot = nextEpoch.start_slot()
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
parentBsi = dag.head.parent.atSlot(nextEpochSlot).toBlockSlotId().get()
|
|
|
|
|
stateCheckpoint = dag.stateCheckpoint(parentBsi)
|
2021-12-30 11:33:03 +00:00
|
|
|
|
|
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
parentBsi.bid == dag.head.parent.bid
|
|
|
|
|
parentBsi.slot == nextEpochSlot
|
2022-01-05 18:38:04 +00:00
|
|
|
dag.getEpochRef(dag.head.parent, nextEpoch, true).isOk()
|
2021-12-30 11:33:03 +00:00
|
|
|
|
|
|
|
|
# Getting an EpochRef should not result in states being stored
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
db.getStateRoot(stateCheckpoint.bid.root, stateCheckpoint.slot).isErr()
|
2021-12-30 11:33:03 +00:00
|
|
|
# this is required for the test to work - it's not a "public"
|
|
|
|
|
# post-condition of getEpochRef
|
2022-03-16 07:20:40 +00:00
|
|
|
getStateField(dag.epochRefState, slot) == nextEpochSlot
|
|
|
|
|
assign(state[], dag.epochRefState)
|
2021-12-30 11:33:03 +00:00
|
|
|
|
|
|
|
|
let
|
|
|
|
|
bnext = addTestBlock(state[], cache).phase0Data
|
|
|
|
|
bnextAdd = dag.addHeadBlock(verifier, bnext, nilPhase0Callback)
|
|
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
# Getting an EpochRef should not result in states being stored
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
db.getStateRoot(stateCheckpoint.bid.root, stateCheckpoint.slot).isOk()
|
2021-12-30 11:33:03 +00:00
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
test "Adding the same block twice returns a Duplicate error" & preset():
|
|
|
|
|
let
|
|
|
|
|
b10 = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
|
|
|
|
b11 = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
|
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
b11.error == BlockError.Duplicate
|
|
|
|
|
not b10[].isNil
|
|
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
test "updateHead updates head and headState" & preset():
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
2020-05-05 09:18:44 +00:00
|
|
|
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(b1Add[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
check:
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.head == b1Add[]
|
2022-03-16 07:20:40 +00:00
|
|
|
getStateField(dag.headState, slot) == b1Add[].slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
test "updateState sanity" & preset():
|
2020-05-05 09:18:44 +00:00
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
|
|
|
|
b2Add = dag.addHeadBlock(verifier, b2, nilPhase0Callback)
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
bs1 = BlockSlotId.init(b1Add[].bid, b1.message.slot)
|
|
|
|
|
bs1_3 = BlockSlotId.init(b1Add[].bid, 3.Slot)
|
|
|
|
|
bs2_3 = BlockSlotId.init(b2Add[].bid, 3.Slot)
|
2020-05-05 09:18:44 +00:00
|
|
|
|
2022-02-20 20:13:06 +00:00
|
|
|
let tmpState = assignClone(dag.headState)
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# move to specific block
|
2020-08-18 20:29:33 +00:00
|
|
|
var cache = StateCache()
|
2020-05-05 09:18:44 +00:00
|
|
|
check:
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.updateState(tmpState[], bs1, false, cache)
|
|
|
|
|
tmpState[].latest_block_root == b1Add[].root
|
|
|
|
|
getStateField(tmpState[], slot) == bs1.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# Skip slots
|
|
|
|
|
check:
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.updateState(tmpState[], bs1_3, false, cache) # skip slots
|
|
|
|
|
tmpState[].latest_block_root == b1Add[].root
|
|
|
|
|
getStateField(tmpState[], slot) == bs1_3.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# Move back slots, but not blocks
|
|
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.updateState(
|
|
|
|
|
tmpState[], dag.parent(bs1_3.bid).expect("block").atSlot(), false, cache)
|
|
|
|
|
tmpState[].latest_block_root == b1Add[].parent.root
|
|
|
|
|
getStateField(tmpState[], slot) == b1Add[].parent.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# Move to different block and slot
|
|
|
|
|
check:
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.updateState(tmpState[], bs2_3, false, cache)
|
|
|
|
|
tmpState[].latest_block_root == b2Add[].root
|
|
|
|
|
getStateField(tmpState[], slot) == bs2_3.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# Move back slot and block
|
|
|
|
|
check:
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.updateState(tmpState[], bs1, false, cache)
|
|
|
|
|
tmpState[].latest_block_root == b1Add[].root
|
|
|
|
|
getStateField(tmpState[], slot) == bs1.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
|
|
|
|
# Move back to genesis
|
|
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.updateState(
|
|
|
|
|
tmpState[], dag.parent(bs1.bid).expect("block").atSlot(), false, cache)
|
2022-03-16 07:20:40 +00:00
|
|
|
tmpState[].latest_block_root == b1Add[].parent.root
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
getStateField(tmpState[], slot) == b1Add[].parent.slot
|
2020-05-05 09:18:44 +00:00
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
when declared(GC_fullCollect): # i386 test machines seem to run low..
|
|
|
|
|
GC_fullCollect()
|
|
|
|
|
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
suite "Block pool altair processing" & preset():
|
|
|
|
|
setup:
|
|
|
|
|
var
|
|
|
|
|
cfg = defaultRuntimeConfig
|
|
|
|
|
cfg.ALTAIR_FORK_EPOCH = Epoch(1)
|
|
|
|
|
|
|
|
|
|
var
|
|
|
|
|
db = makeTestDB(SLOTS_PER_EPOCH)
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, cfg, db, validatorMonitor, {})
|
2021-12-06 09:49:01 +00:00
|
|
|
verifier = BatchVerifier(rng: keys.newRng(), taskpool: Taskpool.new())
|
|
|
|
|
quarantine = Quarantine.init()
|
2022-03-16 07:20:40 +00:00
|
|
|
state = newClone(dag.headState)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
cache = StateCache()
|
|
|
|
|
info = ForkedEpochInfo()
|
|
|
|
|
|
|
|
|
|
# Advance to altair
|
|
|
|
|
check:
|
|
|
|
|
process_slots(
|
2022-01-11 10:01:54 +00:00
|
|
|
cfg, state[], cfg.ALTAIR_FORK_EPOCH.start_slot(), cache,
|
2022-01-17 11:19:58 +00:00
|
|
|
info, {}).isOk()
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
state[].kind == BeaconStateFork.Altair
|
|
|
|
|
|
|
|
|
|
var
|
|
|
|
|
b1 = addTestBlock(state[], cache).altairData
|
|
|
|
|
att1 = makeFullAttestations(state[], b1.root, b1.message.slot, cache)
|
|
|
|
|
b2 = addTestBlock(state[], cache, attestations = att1).altairData
|
|
|
|
|
|
|
|
|
|
test "Invalid signatures" & preset():
|
|
|
|
|
let badSignature = get_slot_signature(
|
2022-06-18 04:57:37 +00:00
|
|
|
Fork(), ZERO_HASH, 42.Slot,
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
MockPrivKeys[ValidatorIndex(0)]).toValidatorSig()
|
|
|
|
|
|
|
|
|
|
check:
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
dag.addHeadBlock(verifier, b1, nilAltairCallback).isOk()
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
block: # Main signature
|
|
|
|
|
var b = b2
|
|
|
|
|
b.signature = badSignature
|
|
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
bAdd = dag.addHeadBlock(verifier, b, nilAltairCallback)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
check:
|
2021-12-06 09:49:01 +00:00
|
|
|
bAdd.error() == BlockError.Invalid
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
block: # Randao reveal
|
|
|
|
|
var b = b2
|
|
|
|
|
b.message.body.randao_reveal = badSignature
|
|
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
bAdd = dag.addHeadBlock(verifier, b, nilAltairCallback)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
check:
|
2021-12-06 09:49:01 +00:00
|
|
|
bAdd.error() == BlockError.Invalid
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
block: # Attestations
|
|
|
|
|
var b = b2
|
|
|
|
|
b.message.body.attestations[0].signature = badSignature
|
|
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
bAdd = dag.addHeadBlock(verifier, b, nilAltairCallback)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
check:
|
2021-12-06 09:49:01 +00:00
|
|
|
bAdd.error() == BlockError.Invalid
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
block: # SyncAggregate empty
|
|
|
|
|
var b = b2
|
|
|
|
|
b.message.body.sync_aggregate.sync_committee_signature = badSignature
|
|
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
bAdd = dag.addHeadBlock(verifier, b, nilAltairCallback)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
check:
|
2021-12-06 09:49:01 +00:00
|
|
|
bAdd.error() == BlockError.Invalid
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
|
|
|
|
|
block: # SyncAggregate junk
|
|
|
|
|
var b = b2
|
|
|
|
|
b.message.body.sync_aggregate.sync_committee_signature = badSignature
|
|
|
|
|
b.message.body.sync_aggregate.sync_committee_bits[0] = true
|
|
|
|
|
|
|
|
|
|
let
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
bAdd = dag.addHeadBlock(verifier, b, nilAltairCallback)
|
Speed up altair block processing 2x (#3115)
* Speed up altair block processing >2x
Like #3089, this PR drastially speeds up historical REST queries and
other long state replays.
* cache sync committee validator indices
* use ~80mb less memory for validator pubkey mappings
* batch-verify sync aggregate signature (fixes #2985)
* document sync committee hack with head block vs sync message block
* add batch signature verification failure tests
Before:
```
../env.sh nim c -d:release -r ncli_db --db:mainnet_0/db bench --start-slot:-1000
All time are ms
Average, StdDev, Min, Max, Samples, Test
Validation is turned off meaning that no BLS operations are performed
5830.675, 0.000, 5830.675, 5830.675, 1, Initialize DB
0.481, 1.878, 0.215, 59.167, 981, Load block from database
8422.566, 0.000, 8422.566, 8422.566, 1, Load state from database
6.996, 1.678, 0.042, 14.385, 969, Advance slot, non-epoch
93.217, 8.318, 84.192, 122.209, 32, Advance slot, epoch
20.513, 23.665, 11.510, 201.561, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
After:
```
7081.422, 0.000, 7081.422, 7081.422, 1, Initialize DB
0.553, 2.122, 0.175, 66.692, 981, Load block from database
5439.446, 0.000, 5439.446, 5439.446, 1, Load state from database
6.829, 1.575, 0.043, 12.156, 969, Advance slot, non-epoch
94.716, 2.749, 88.395, 100.026, 32, Advance slot, epoch
11.636, 23.766, 4.889, 205.250, 981, Apply block, no slot processing
0.000, 0.000, 0.000, 0.000, 0, Database load
0.000, 0.000, 0.000, 0.000, 0, Database store
```
* add comment
2021-11-24 12:43:50 +00:00
|
|
|
check:
|
2021-12-06 09:49:01 +00:00
|
|
|
bAdd.error() == BlockError.Invalid
|
2020-05-05 09:18:44 +00:00
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
suite "chain DAG finalization tests" & preset():
|
2020-07-01 17:00:14 +00:00
|
|
|
setup:
|
|
|
|
|
var
|
|
|
|
|
db = makeTestDB(SLOTS_PER_EPOCH)
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor, {})
|
2021-12-06 09:49:01 +00:00
|
|
|
verifier = BatchVerifier(rng: keys.newRng(), taskpool: Taskpool.new())
|
|
|
|
|
quarantine = Quarantine.init()
|
2020-07-15 10:44:18 +00:00
|
|
|
cache = StateCache()
|
2021-10-13 14:24:36 +00:00
|
|
|
info = ForkedEpochInfo()
|
2020-02-05 11:41:46 +00:00
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
test "prune heads on finalization" & preset():
|
2020-07-01 17:00:14 +00:00
|
|
|
# Create a fork that will not be taken
|
|
|
|
|
var
|
2022-03-16 07:20:40 +00:00
|
|
|
blck = makeTestBlock(dag.headState, cache).phase0Data
|
|
|
|
|
tmpState = assignClone(dag.headState)
|
2020-07-01 17:00:14 +00:00
|
|
|
check:
|
|
|
|
|
process_slots(
|
Implement split preset/config support (#2710)
* Implement split preset/config support
This is the initial bulk refactor to introduce runtime config values in
a number of places, somewhat replacing the existing mechanism of loading
network metadata.
It still needs more work, this is the initial refactor that introduces
runtime configuration in some of the places that need it.
The PR changes the way presets and constants work, to match the spec. In
particular, a "preset" now refers to the compile-time configuration
while a "cfg" or "RuntimeConfig" is the dynamic part.
A single binary can support either mainnet or minimal, but not both.
Support for other presets has been removed completely (can be readded,
in case there's need).
There's a number of outstanding tasks:
* `SECONDS_PER_SLOT` still needs fixing
* loading custom runtime configs needs redoing
* checking constants against YAML file
* yeerongpilly support
`build/nimbus_beacon_node --network=yeerongpilly --discv5:no --log-level=DEBUG`
* load fork epoch from config
* fix fork digest sent in status
* nicer error string for request failures
* fix tools
* one more
* fixup
* fixup
* fixup
* use "standard" network definition folder in local testnet
Files are loaded from their standard locations, including genesis etc,
to conform to the format used in the `eth2-networks` repo.
* fix launch scripts, allow unknown config values
* fix base config of rest test
* cleanups
* bundle mainnet config using common loader
* fix spec links and names
* only include supported preset in binary
* drop yeerongpilly, add altair-devnet-0, support boot_enr.yaml
2021-07-12 13:01:38 +00:00
|
|
|
defaultRuntimeConfig, tmpState[],
|
|
|
|
|
getStateField(tmpState[], slot) + (5 * SLOTS_PER_EPOCH).uint64,
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2020-07-01 17:00:14 +00:00
|
|
|
|
2021-11-18 12:02:43 +00:00
|
|
|
let lateBlock = addTestBlock(tmpState[], cache).phase0Data
|
2020-07-09 09:29:32 +00:00
|
|
|
block:
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let status = dag.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-07-09 09:29:32 +00:00
|
|
|
check: status.isOk()
|
|
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
assign(tmpState[], dag.headState)
|
2020-08-27 07:34:12 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
# skip slots so we can test gappy getBlockIdAtSlot
|
2022-02-26 18:16:19 +00:00
|
|
|
check process_slots(
|
|
|
|
|
defaultRuntimeConfig, tmpState[],
|
|
|
|
|
getStateField(tmpState[], slot) + 2.uint64,
|
|
|
|
|
cache, info, {}).isOk()
|
|
|
|
|
|
2020-07-01 17:00:14 +00:00
|
|
|
for i in 0 ..< (SLOTS_PER_EPOCH * 6):
|
|
|
|
|
if i == 1:
|
|
|
|
|
# There are 2 heads now because of the fork at slot 1
|
|
|
|
|
check:
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.heads.len == 2
|
2020-07-09 09:29:32 +00:00
|
|
|
|
2020-08-27 07:34:12 +00:00
|
|
|
blck = addTestBlock(
|
2021-11-18 12:02:43 +00:00
|
|
|
tmpState[], cache,
|
2020-07-09 09:29:32 +00:00
|
|
|
attestations = makeFullAttestations(
|
2021-10-18 16:37:27 +00:00
|
|
|
tmpState[], dag.head.root, getStateField(tmpState[], slot), cache, {})).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added = dag.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-07-09 09:29:32 +00:00
|
|
|
check: added.isOk()
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(added[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-07-01 17:00:14 +00:00
|
|
|
|
|
|
|
|
check:
|
2020-07-30 19:18:17 +00:00
|
|
|
dag.heads.len() == 1
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockIdAtSlot(0.Slot).get() == BlockSlotId.init(dag.genesis, 0.Slot)
|
|
|
|
|
dag.getBlockIdAtSlot(2.Slot).get() ==
|
|
|
|
|
BlockSlotId.init(dag.getBlockIdAtSlot(1.Slot).get().bid, 2.Slot)
|
2022-02-26 18:16:19 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.head.slot).get() == BlockSlotId.init(
|
|
|
|
|
dag.head.bid, dag.head.slot)
|
|
|
|
|
dag.getBlockIdAtSlot(dag.head.slot + 1).get() == BlockSlotId.init(
|
|
|
|
|
dag.head.bid, dag.head.slot + 1)
|
2020-07-01 17:00:14 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
not dag.containsForkBlock(dag.getBlockIdAtSlot(5.Slot).get().bid.root)
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.containsForkBlock(dag.finalizedHead.blck.root)
|
|
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockRef(dag.genesis.root).isNone() # Finalized - no BlockRef
|
|
|
|
|
|
|
|
|
|
dag.getBlockRef(dag.finalizedHead.blck.root).isSome()
|
|
|
|
|
|
|
|
|
|
isNil dag.finalizedHead.blck.parent
|
|
|
|
|
|
2020-08-26 15:06:40 +00:00
|
|
|
check:
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.db.immutableValidators.len() == getStateField(dag.headState, validators).len()
|
2020-08-26 15:06:40 +00:00
|
|
|
|
2022-03-24 14:37:37 +00:00
|
|
|
block:
|
|
|
|
|
var cur = dag.head.bid
|
|
|
|
|
while true:
|
|
|
|
|
let parent = dag.parent(cur)
|
|
|
|
|
if cur.slot > 0:
|
|
|
|
|
check:
|
|
|
|
|
parent.isSome and parent.get().slot < cur.slot
|
|
|
|
|
cur = parent.get()
|
|
|
|
|
else:
|
|
|
|
|
check:
|
|
|
|
|
parent.isErr()
|
|
|
|
|
break
|
|
|
|
|
check: cur.slot == 0
|
|
|
|
|
|
|
|
|
|
block:
|
|
|
|
|
var cur = dag.head.bid.atSlot()
|
|
|
|
|
while true:
|
|
|
|
|
let parent = dag.parentOrSlot(cur)
|
|
|
|
|
if cur.slot > 0:
|
|
|
|
|
check:
|
|
|
|
|
parent.isSome and (parent.get().slot < cur.slot or parent.get().bid != cur.bid)
|
|
|
|
|
cur = parent.get()
|
|
|
|
|
else:
|
|
|
|
|
check:
|
|
|
|
|
parent.isErr()
|
|
|
|
|
break
|
|
|
|
|
check: cur.slot == 0
|
|
|
|
|
|
2021-06-01 11:13:40 +00:00
|
|
|
let
|
2022-01-05 18:38:04 +00:00
|
|
|
finalER = dag.getEpochRef(
|
|
|
|
|
dag.finalizedHead.blck, dag.finalizedHead.slot.epoch, false)
|
2020-08-11 19:39:53 +00:00
|
|
|
|
2020-11-20 14:16:04 +00:00
|
|
|
# The EpochRef for the finalized block is needed for eth1 voting, so we
|
|
|
|
|
# should never drop it!
|
2021-06-01 11:13:40 +00:00
|
|
|
check:
|
2022-01-05 18:38:04 +00:00
|
|
|
not finalER.isErr()
|
2020-11-20 14:16:04 +00:00
|
|
|
|
2020-08-26 15:06:40 +00:00
|
|
|
block:
|
2021-03-17 10:17:15 +00:00
|
|
|
for er in dag.epochRefs:
|
2021-06-10 22:07:16 +00:00
|
|
|
check: er == nil or er.epoch >= dag.finalizedHead.slot.epoch
|
2021-03-17 10:17:15 +00:00
|
|
|
|
2021-06-03 13:32:00 +00:00
|
|
|
block:
|
|
|
|
|
let tmpStateData = assignClone(dag.headState)
|
|
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
# Check that cached data is available after updateState - since we
|
2021-06-03 13:32:00 +00:00
|
|
|
# just processed the head the relevant epochrefs should not have been
|
|
|
|
|
# evicted yet
|
|
|
|
|
cache = StateCache()
|
2022-03-16 07:20:40 +00:00
|
|
|
check: updateState(
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag, tmpStateData[],
|
|
|
|
|
dag.head.atSlot(dag.head.slot).toBlockSlotId().expect("not nil"),
|
|
|
|
|
false, cache)
|
2021-06-03 13:32:00 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.head.slot.epoch in cache.shuffled_active_validator_indices
|
|
|
|
|
(dag.head.slot.epoch - 1) in cache.shuffled_active_validator_indices
|
|
|
|
|
|
|
|
|
|
dag.head.slot in cache.beacon_proposer_indices
|
|
|
|
|
|
2020-07-09 09:29:32 +00:00
|
|
|
block:
|
2020-07-01 17:00:14 +00:00
|
|
|
# The late block is a block whose parent was finalized long ago and thus
|
|
|
|
|
# is no longer a viable head candidate
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let status = dag.addHeadBlock(verifier, lateBlock, nilPhase0Callback)
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
# This _should_ be Unviable, but we can't tell, from the data that we have
|
|
|
|
|
# so MissingParent is the least wrong thing to reply
|
2021-12-06 09:49:01 +00:00
|
|
|
check: status.error == BlockError.UnviableFork
|
2020-07-01 17:00:14 +00:00
|
|
|
|
2021-03-01 19:50:43 +00:00
|
|
|
block:
|
|
|
|
|
let
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
finalizedCheckpoint = dag.stateCheckpoint(dag.finalizedHead.toBlockSlotId().get())
|
|
|
|
|
headCheckpoint = dag.stateCheckpoint(dag.head.bid.atSlot())
|
2022-03-24 14:37:37 +00:00
|
|
|
prunedCheckpoint = dag.stateCheckpoint(dag.parent(dag.finalizedHead.blck.bid).get().atSlot())
|
2021-03-01 19:50:43 +00:00
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
db.getStateRoot(headCheckpoint.bid.root, headCheckpoint.slot).isSome
|
|
|
|
|
db.getStateRoot(finalizedCheckpoint.bid.root, finalizedCheckpoint.slot).isSome
|
2022-03-24 14:37:37 +00:00
|
|
|
db.getStateRoot(prunedCheckpoint.bid.root, prunedCheckpoint.slot).isNone
|
2021-03-01 19:50:43 +00:00
|
|
|
|
2020-07-01 17:00:14 +00:00
|
|
|
let
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor2 = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag2 = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor2, {})
|
2020-07-01 17:00:14 +00:00
|
|
|
|
|
|
|
|
# check that the state reloaded from database resembles what we had before
|
|
|
|
|
check:
|
2020-07-30 19:18:17 +00:00
|
|
|
dag2.tail.root == dag.tail.root
|
|
|
|
|
dag2.head.root == dag.head.root
|
|
|
|
|
dag2.finalizedHead.blck.root == dag.finalizedHead.blck.root
|
|
|
|
|
dag2.finalizedHead.slot == dag.finalizedHead.slot
|
2022-03-16 07:20:40 +00:00
|
|
|
getStateRoot(dag2.headState) == getStateRoot(dag.headState)
|
2020-06-25 09:36:03 +00:00
|
|
|
|
2022-03-21 09:20:26 +00:00
|
|
|
# No canonical block data should be pruned by the removal of the fork
|
|
|
|
|
for i in Slot(0)..dag2.head.slot:
|
|
|
|
|
let bids = dag.getBlockIdAtSlot(i).expect("found it")
|
|
|
|
|
if bids.isProposed:
|
|
|
|
|
check: dag2.getForkedBlock(bids.bid).isSome
|
|
|
|
|
|
|
|
|
|
# The unviable block should have been pruned however
|
|
|
|
|
check: dag2.getForkedBlock(lateBlock.root).isNone
|
|
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
test "orphaned epoch block" & preset():
|
2022-02-20 20:13:06 +00:00
|
|
|
let prestate = (ref ForkedHashedBeaconState)(kind: BeaconStateFork.Phase0)
|
2020-08-13 09:50:05 +00:00
|
|
|
for i in 0 ..< SLOTS_PER_EPOCH:
|
|
|
|
|
if i == SLOTS_PER_EPOCH - 1:
|
2022-03-16 07:20:40 +00:00
|
|
|
assign(prestate[], dag.headState)
|
2020-08-13 09:50:05 +00:00
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
let blck = makeTestBlock(dag.headState, cache).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added = dag.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-08-13 09:50:05 +00:00
|
|
|
check: added.isOk()
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(added[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.heads.len() == 1
|
|
|
|
|
|
2020-09-07 15:04:33 +00:00
|
|
|
# The loop creates multiple branches, which StateCache isn't suitable for
|
|
|
|
|
cache = StateCache()
|
|
|
|
|
|
2021-06-11 17:51:46 +00:00
|
|
|
doAssert process_slots(
|
Implement split preset/config support (#2710)
* Implement split preset/config support
This is the initial bulk refactor to introduce runtime config values in
a number of places, somewhat replacing the existing mechanism of loading
network metadata.
It still needs more work, this is the initial refactor that introduces
runtime configuration in some of the places that need it.
The PR changes the way presets and constants work, to match the spec. In
particular, a "preset" now refers to the compile-time configuration
while a "cfg" or "RuntimeConfig" is the dynamic part.
A single binary can support either mainnet or minimal, but not both.
Support for other presets has been removed completely (can be readded,
in case there's need).
There's a number of outstanding tasks:
* `SECONDS_PER_SLOT` still needs fixing
* loading custom runtime configs needs redoing
* checking constants against YAML file
* yeerongpilly support
`build/nimbus_beacon_node --network=yeerongpilly --discv5:no --log-level=DEBUG`
* load fork epoch from config
* fix fork digest sent in status
* nicer error string for request failures
* fix tools
* one more
* fixup
* fixup
* fixup
* use "standard" network definition folder in local testnet
Files are loaded from their standard locations, including genesis etc,
to conform to the format used in the `eth2-networks` repo.
* fix launch scripts, allow unknown config values
* fix base config of rest test
* cleanups
* bundle mainnet config using common loader
* fix spec links and names
* only include supported preset in binary
* drop yeerongpilly, add altair-devnet-0, support boot_enr.yaml
2021-07-12 13:01:38 +00:00
|
|
|
defaultRuntimeConfig, prestate[], getStateField(prestate[], slot) + 1,
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
# create another block, orphaning the head
|
2021-11-18 12:02:43 +00:00
|
|
|
let blck = makeTestBlock(prestate[], cache).phase0Data
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
# Add block, but don't update head
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added = dag.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-08-13 09:50:05 +00:00
|
|
|
check: added.isOk()
|
|
|
|
|
|
|
|
|
|
var
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor2 = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag2 = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor2, {})
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
# check that we can apply the block after the orphaning
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added2 = dag2.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-08-13 09:50:05 +00:00
|
|
|
check: added2.isOk()
|
|
|
|
|
|
2021-04-28 16:41:02 +00:00
|
|
|
test "init with gaps" & preset():
|
2020-08-27 07:34:12 +00:00
|
|
|
for blck in makeTestBlocks(
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.headState, cache, int(SLOTS_PER_EPOCH * 6 - 2),
|
2020-08-27 07:34:12 +00:00
|
|
|
true):
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added = dag.addHeadBlock(verifier, blck.phase0Data, nilPhase0Callback)
|
2020-08-13 09:50:05 +00:00
|
|
|
check: added.isOk()
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(added[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
# Advance past epoch so that the epoch transition is gapped
|
|
|
|
|
check:
|
|
|
|
|
process_slots(
|
2022-03-16 07:20:40 +00:00
|
|
|
defaultRuntimeConfig, dag.headState, Slot(SLOTS_PER_EPOCH * 6 + 2),
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2020-08-13 09:50:05 +00:00
|
|
|
|
2022-02-20 20:13:06 +00:00
|
|
|
let blck = makeTestBlock(
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.headState, cache,
|
2020-08-13 09:50:05 +00:00
|
|
|
attestations = makeFullAttestations(
|
2022-03-16 07:20:40 +00:00
|
|
|
dag.headState, dag.head.root, getStateField(dag.headState, slot),
|
2021-10-18 16:37:27 +00:00
|
|
|
cache, {})).phase0Data
|
2020-08-13 09:50:05 +00:00
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
let added = dag.addHeadBlock(verifier, blck, nilPhase0Callback)
|
2020-08-13 09:50:05 +00:00
|
|
|
check: added.isOk()
|
2020-08-31 09:00:38 +00:00
|
|
|
dag.updateHead(added[], quarantine)
|
2021-03-09 14:36:17 +00:00
|
|
|
dag.pruneAtFinalization()
|
2020-08-13 09:50:05 +00:00
|
|
|
|
2021-05-17 16:37:26 +00:00
|
|
|
block:
|
|
|
|
|
# Check that we can rewind to every block from head to finalized
|
|
|
|
|
var
|
|
|
|
|
cur = dag.head
|
|
|
|
|
tmpStateData = assignClone(dag.headState)
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
while cur != nil: # Go all the way to dag.finalizedHead
|
2021-05-17 16:37:26 +00:00
|
|
|
assign(tmpStateData[], dag.headState)
|
|
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.updateState(tmpStateData[], cur.bid.atSlot(), false, cache)
|
2022-03-11 12:08:17 +00:00
|
|
|
dag.getForkedBlock(cur.bid).get().phase0Data.message.state_root ==
|
2022-03-16 07:20:40 +00:00
|
|
|
getStateRoot(tmpStateData[])
|
|
|
|
|
getStateRoot(tmpStateData[]) == hash_tree_root(
|
|
|
|
|
tmpStateData[].phase0Data.data)
|
2021-05-17 16:37:26 +00:00
|
|
|
cur = cur.parent
|
|
|
|
|
|
2020-08-13 09:50:05 +00:00
|
|
|
let
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor2 = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag2 = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor2, {})
|
2020-08-13 09:50:05 +00:00
|
|
|
|
|
|
|
|
# check that the state reloaded from database resembles what we had before
|
|
|
|
|
check:
|
|
|
|
|
dag2.tail.root == dag.tail.root
|
|
|
|
|
dag2.head.root == dag.head.root
|
|
|
|
|
dag2.finalizedHead.blck.root == dag.finalizedHead.blck.root
|
|
|
|
|
dag2.finalizedHead.slot == dag.finalizedHead.slot
|
2022-03-16 07:20:40 +00:00
|
|
|
getStateRoot(dag2.headState) == getStateRoot(dag.headState)
|
2021-08-05 08:26:10 +00:00
|
|
|
|
|
|
|
|
suite "Old database versions" & preset():
|
|
|
|
|
setup:
|
|
|
|
|
let
|
2021-11-18 12:02:43 +00:00
|
|
|
genState = newClone(initialize_hashed_beacon_state_from_eth1(
|
2022-06-18 04:57:37 +00:00
|
|
|
defaultRuntimeConfig, ZERO_HASH, 0,
|
2021-08-05 08:26:10 +00:00
|
|
|
makeInitialDeposits(SLOTS_PER_EPOCH.uint64, flags = {skipBlsValidation}),
|
2021-11-18 12:02:43 +00:00
|
|
|
{skipBlsValidation}))
|
2021-08-05 08:26:10 +00:00
|
|
|
genBlock = get_initial_beacon_block(genState[])
|
2021-12-06 09:49:01 +00:00
|
|
|
var
|
|
|
|
|
verifier = BatchVerifier(rng: keys.newRng(), taskpool: Taskpool.new())
|
|
|
|
|
quarantine = Quarantine.init()
|
2021-08-05 08:26:10 +00:00
|
|
|
|
|
|
|
|
test "pre-1.1.0":
|
|
|
|
|
# only kvstore, no immutable validator keys
|
|
|
|
|
|
2021-08-09 11:14:28 +00:00
|
|
|
let db = BeaconChainDB.new("", inMemory = true)
|
2021-08-05 08:26:10 +00:00
|
|
|
|
|
|
|
|
# preInit a database to a v1.0.12 state
|
2021-11-18 12:02:43 +00:00
|
|
|
db.putStateRoot(
|
2022-03-16 07:20:40 +00:00
|
|
|
genState[].latest_block_root, genState[].data.slot, genState[].root)
|
2021-11-18 12:02:43 +00:00
|
|
|
db.putStateV0(genState[].root, genState[].data)
|
|
|
|
|
|
2021-08-05 08:26:10 +00:00
|
|
|
db.putBlockV0(genBlock)
|
|
|
|
|
db.putTailBlock(genBlock.root)
|
|
|
|
|
db.putHeadBlock(genBlock.root)
|
2021-11-05 07:34:34 +00:00
|
|
|
db.putGenesisBlock(genBlock.root)
|
2021-08-05 08:26:10 +00:00
|
|
|
|
|
|
|
|
var
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, defaultRuntimeConfig, db,validatorMonitor, {})
|
2022-03-16 07:20:40 +00:00
|
|
|
state = newClone(dag.headState)
|
2021-08-05 08:26:10 +00:00
|
|
|
cache = StateCache()
|
|
|
|
|
att0 = makeFullAttestations(state[], dag.tail.root, 0.Slot, cache)
|
2021-11-18 12:02:43 +00:00
|
|
|
b1 = addTestBlock(state[], cache, attestations = att0).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
2021-08-05 08:26:10 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
b1Add.isOk()
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
suite "Diverging hardforks":
|
|
|
|
|
setup:
|
|
|
|
|
var
|
|
|
|
|
phase0RuntimeConfig = defaultRuntimeConfig
|
|
|
|
|
altairRuntimeConfig = defaultRuntimeConfig
|
|
|
|
|
|
|
|
|
|
phase0RuntimeConfig.ALTAIR_FORK_EPOCH = FAR_FUTURE_EPOCH
|
|
|
|
|
altairRuntimeConfig.ALTAIR_FORK_EPOCH = 2.Epoch
|
|
|
|
|
|
|
|
|
|
var
|
|
|
|
|
db = makeTestDB(SLOTS_PER_EPOCH)
|
2021-12-20 19:20:31 +00:00
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, phase0RuntimeConfig, db, validatorMonitor, {})
|
2021-12-06 09:49:01 +00:00
|
|
|
verifier = BatchVerifier(rng: keys.newRng(), taskpool: Taskpool.new())
|
|
|
|
|
quarantine = newClone(Quarantine.init())
|
2021-09-08 03:46:33 +00:00
|
|
|
cache = StateCache()
|
2021-10-13 14:24:36 +00:00
|
|
|
info = ForkedEpochInfo()
|
2022-03-16 07:20:40 +00:00
|
|
|
tmpState = assignClone(dag.headState)
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
test "Tail block only in common":
|
|
|
|
|
check:
|
|
|
|
|
process_slots(
|
|
|
|
|
phase0RuntimeConfig, tmpState[],
|
|
|
|
|
getStateField(tmpState[], slot) + (3 * SLOTS_PER_EPOCH).uint64,
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
# Because the first block is after the Altair transition, the only block in
|
|
|
|
|
# common is the tail block
|
|
|
|
|
var
|
2021-11-18 12:02:43 +00:00
|
|
|
b1 = addTestBlock(tmpState[], cache).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
check b1Add.isOk()
|
2021-12-06 09:49:01 +00:00
|
|
|
dag.updateHead(b1Add[], quarantine[])
|
2021-09-08 03:46:33 +00:00
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
let validatorMonitorAltair = newClone(ValidatorMonitor.init())
|
|
|
|
|
|
2022-02-20 20:13:06 +00:00
|
|
|
let dagAltair = init(
|
2021-12-20 19:20:31 +00:00
|
|
|
ChainDAGRef, altairRuntimeConfig, db, validatorMonitorAltair, {})
|
2021-09-08 03:46:33 +00:00
|
|
|
discard AttestationPool.init(dagAltair, quarantine)
|
|
|
|
|
|
|
|
|
|
test "Non-tail block in common":
|
|
|
|
|
check:
|
|
|
|
|
process_slots(
|
|
|
|
|
phase0RuntimeConfig, tmpState[],
|
|
|
|
|
getStateField(tmpState[], slot) + SLOTS_PER_EPOCH.uint64,
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
# There's a block in the shared-correct phase0 hardfork, before epoch 2
|
|
|
|
|
var
|
2021-11-18 12:02:43 +00:00
|
|
|
b1 = addTestBlock(tmpState[], cache).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b1Add = dag.addHeadBlock(verifier, b1, nilPhase0Callback)
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
b1Add.isOk()
|
|
|
|
|
process_slots(
|
|
|
|
|
phase0RuntimeConfig, tmpState[],
|
|
|
|
|
getStateField(tmpState[], slot) + (3 * SLOTS_PER_EPOCH).uint64,
|
2022-01-17 11:19:58 +00:00
|
|
|
cache, info, {}).isOk()
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
var
|
2021-11-18 12:02:43 +00:00
|
|
|
b2 = addTestBlock(tmpState[], cache).phase0Data
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
b2Add = dag.addHeadBlock(verifier, b2, nilPhase0Callback)
|
2021-09-08 03:46:33 +00:00
|
|
|
|
|
|
|
|
check b2Add.isOk()
|
2021-12-06 09:49:01 +00:00
|
|
|
dag.updateHead(b2Add[], quarantine[])
|
2021-09-08 03:46:33 +00:00
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
let validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
|
2022-02-20 20:13:06 +00:00
|
|
|
let dagAltair = init(
|
2021-12-20 19:20:31 +00:00
|
|
|
ChainDAGRef, altairRuntimeConfig, db, validatorMonitor, {})
|
2021-09-08 03:46:33 +00:00
|
|
|
discard AttestationPool.init(dagAltair, quarantine)
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
suite "Backfill":
|
|
|
|
|
setup:
|
|
|
|
|
let
|
|
|
|
|
genState = (ref ForkedHashedBeaconState)(
|
|
|
|
|
kind: BeaconStateFork.Phase0,
|
|
|
|
|
phase0Data: initialize_hashed_beacon_state_from_eth1(
|
2022-06-18 04:57:37 +00:00
|
|
|
defaultRuntimeConfig, ZERO_HASH, 0,
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
makeInitialDeposits(SLOTS_PER_EPOCH.uint64, flags = {skipBlsValidation}),
|
|
|
|
|
{skipBlsValidation}))
|
|
|
|
|
tailState = assignClone(genState[])
|
|
|
|
|
|
|
|
|
|
blocks = block:
|
|
|
|
|
var blocks: seq[ForkedSignedBeaconBlock]
|
|
|
|
|
var cache: StateCache
|
2022-01-05 18:38:04 +00:00
|
|
|
for i in 0..<SLOTS_PER_EPOCH * 2:
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks.add addTestBlock(tailState[], cache)
|
|
|
|
|
blocks
|
|
|
|
|
|
|
|
|
|
let
|
|
|
|
|
db = BeaconChainDB.new("", inMemory = true)
|
|
|
|
|
|
|
|
|
|
test "backfill to genesis":
|
|
|
|
|
let
|
|
|
|
|
tailBlock = blocks[^1]
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
genBlock = get_initial_beacon_block(genState[])
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
ChainDAGRef.preInit(
|
|
|
|
|
db, genState[], tailState[], tailBlock.asTrusted())
|
|
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
let
|
|
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor, {})
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
check:
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockRef(tailBlock.root).get().bid == dag.tail
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.getBlockRef(blocks[^2].root).isNone()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockId(tailBlock.root).get() == dag.tail
|
|
|
|
|
dag.getBlockId(blocks[^2].root).isNone()
|
|
|
|
|
|
|
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot).get().bid == dag.tail
|
|
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot - 1).isNone()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockIdAtSlot(Slot(0)).get() == dag.genesis.atSlot()
|
|
|
|
|
dag.getBlockIdAtSlot(Slot(1)).isNone()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
2022-01-05 18:38:04 +00:00
|
|
|
# No epochref for pre-tail epochs
|
|
|
|
|
dag.getEpochRef(dag.tail, dag.tail.slot.epoch - 1, true).isErr()
|
|
|
|
|
|
2022-03-18 12:13:57 +00:00
|
|
|
dag.getFinalizedEpochRef() != nil
|
|
|
|
|
|
2022-02-26 18:16:19 +00:00
|
|
|
dag.backfill == tailBlock.phase0Data.message.toBeaconBlockSummary()
|
|
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
var
|
|
|
|
|
badBlock = blocks[^2].phase0Data
|
|
|
|
|
badBlock.signature = blocks[^3].phase0Data.signature
|
|
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(badBlock).error == BlockError.Invalid
|
|
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(blocks[^3].phase0Data).error == BlockError.MissingParent
|
|
|
|
|
dag.addBackfillBlock(tailBlock.phase0Data).error == BlockError.Duplicate
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.addBackfillBlock(genBlock.phase0Data.asSigned()).error == BlockError.MissingParent
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(blocks[^2].phase0Data).isOk()
|
|
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockRef(tailBlock.root).get().bid == dag.tail
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag.getBlockRef(blocks[^2].root).isNone()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockId(tailBlock.root).get() == dag.tail
|
|
|
|
|
dag.getBlockId(blocks[^2].root).get().root == blocks[^2].root
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot).get().bid == dag.tail
|
2022-03-15 08:24:55 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot - 1).get() ==
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks[^2].toBlockId().atSlot()
|
2022-03-15 08:24:55 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot - 2).isNone
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
2022-02-26 18:16:19 +00:00
|
|
|
dag.backfill == blocks[^2].phase0Data.message.toBeaconBlockSummary()
|
|
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(blocks[^3].phase0Data).isOk()
|
|
|
|
|
|
2022-03-15 08:24:55 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot - 2).get() ==
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks[^3].toBlockId().atSlot()
|
2022-03-15 08:24:55 +00:00
|
|
|
dag.getBlockIdAtSlot(dag.tail.slot - 3).isNone
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
for i in 3..<blocks.len:
|
|
|
|
|
check: dag.addBackfillBlock(blocks[blocks.len - i - 1].phase0Data).isOk()
|
|
|
|
|
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(genBlock.phase0Data.asSigned).error == BlockError.Duplicate
|
|
|
|
|
|
|
|
|
|
dag.backfill.slot == GENESIS_SLOT
|
|
|
|
|
|
2022-03-11 12:49:47 +00:00
|
|
|
dag.rebuildIndex()
|
|
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.getFinalizedEpochRef() != nil
|
|
|
|
|
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
test "reload backfill position":
|
|
|
|
|
let
|
|
|
|
|
tailBlock = blocks[^1]
|
|
|
|
|
|
|
|
|
|
ChainDAGRef.preInit(
|
|
|
|
|
db, genState[], tailState[], tailBlock.asTrusted())
|
|
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
let
|
|
|
|
|
validatorMonitor = newClone(ValidatorMonitor.init())
|
|
|
|
|
dag = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor, {})
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
check:
|
|
|
|
|
dag.addBackfillBlock(blocks[^2].phase0Data).isOk()
|
2022-02-26 18:16:19 +00:00
|
|
|
dag.backfill == blocks[^2].phase0Data.message.toBeaconBlockSummary()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
2021-12-20 19:20:31 +00:00
|
|
|
let
|
|
|
|
|
validatorMonitor2 = newClone(ValidatorMonitor.init())
|
|
|
|
|
|
|
|
|
|
dag2 = init(ChainDAGRef, defaultRuntimeConfig, db, validatorMonitor2, {})
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
|
|
|
|
check:
|
2022-03-18 12:13:57 +00:00
|
|
|
dag2.getFinalizedEpochRef() != nil
|
|
|
|
|
|
limit by-root requests to non-finalized blocks (#3293)
* limit by-root requests to non-finalized blocks
Presently, we keep a mapping from block root to `BlockRef` in memory -
this has simplified reasoning about the dag, but is not sustainable with
the chain growing.
We can distinguish between two cases where by-root access is useful:
* unfinalized blocks - this is where the beacon chain is operating
generally, by validating incoming data as interesting for future fork
choice decisions - bounded by the length of the unfinalized period
* finalized blocks - historical access in the REST API etc - no bounds,
really
In this PR, we limit the by-root block index to the first use case:
finalized chain data can more efficiently be addressed by slot number.
Future work includes:
* limiting the `BlockRef` horizon in general - each instance is 40
bytes+overhead which adds up - this needs further refactoring to deal
with the tail vs state problem
* persisting the finalized slot-to-hash index - this one also keeps
growing unbounded (albeit slowly)
Anyway, this PR easily shaves ~128mb of memory usage at the time of
writing.
* No longer honor `BeaconBlocksByRoot` requests outside of the
non-finalized period - previously, Nimbus would generously return any
block through this libp2p request - per the spec, finalized blocks
should be fetched via `BeaconBlocksByRange` instead.
* return `Opt[BlockRef]` instead of `nil` when blocks can't be found -
this becomes a lot more common now and thus deserves more attention
* `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized
blocks from now - `finalizedBlocks` covers the other `BlockRef`
instances
* in backfill, verify that the last backfilled block leads back to
genesis, or panic
* add backfill timings to log
* fix missing check that `BlockRef` block can be fetched with
`getForkedBlock` reliably
* shortcut doppelganger check when feature is not enabled
* in REST/JSON-RPC, fetch blocks without involving `BlockRef`
* fix dag.blocks ref
2022-01-21 11:33:16 +00:00
|
|
|
dag2.getBlockRef(tailBlock.root).get().root == dag.tail.root
|
|
|
|
|
dag2.getBlockRef(blocks[^2].root).isNone()
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
Prune `BlockRef` on finalization (#3513)
Up til now, the block dag has been using `BlockRef`, a structure adapted
for a full DAG, to represent all of chain history. This is a correct and
simple design, but does not exploit the linearity of the chain once
parts of it finalize.
By pruning the in-memory `BlockRef` structure at finalization, we save,
at the time of writing, a cool ~250mb (or 25%:ish) chunk of memory
landing us at a steady state of ~750mb normal memory usage for a
validating node.
Above all though, we prevent memory usage from growing proportionally
with the length of the chain, something that would not be sustainable
over time - instead, the steady state memory usage is roughly
determined by the validator set size which grows much more slowly. With
these changes, the core should remain sustainable memory-wise post-merge
all the way to withdrawals (when the validator set is expected to grow).
In-memory indices are still used for the "hot" unfinalized portion of
the chain - this ensure that consensus performance remains unchanged.
What changes is that for historical access, we use a db-based linear
slot index which is cache-and-disk-friendly, keeping the cost for
accessing historical data at a similar level as before, achieving the
savings at no percievable cost to functionality or performance.
A nice collateral benefit is the almost-instant startup since we no
longer load any large indicies at dag init.
The cost of this functionality instead can be found in the complexity of
having to deal with two ways of traversing the chain - by `BlockRef` and
by slot.
* use `BlockId` instead of `BlockRef` where finalized / historical data
may be required
* simplify clearance pre-advancement
* remove dag.finalizedBlocks (~50:ish mb)
* remove `getBlockAtSlot` - use `getBlockIdAtSlot` instead
* `parent` and `atSlot` for `BlockId` now require a `ChainDAGRef`
instance, unlike `BlockRef` traversal
* prune `BlockRef` parents on finality (~200:ish mb)
* speed up ChainDAG init by not loading finalized history index
* mess up light client server error handling - this need revisiting :)
2022-03-17 17:42:56 +00:00
|
|
|
dag2.getBlockIdAtSlot(dag.tail.slot).get().bid.root == dag.tail.root
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
|
2022-03-15 08:24:55 +00:00
|
|
|
dag2.getBlockIdAtSlot(dag.tail.slot - 1).get() ==
|
Backfill support for ChainDAG (#3171)
In the ChainDAG, 3 block pointers are kept: genesis, tail and head. This
PR adds one more block pointer: the backfill block which represents the
block that has been backfilled so far.
When doing a checkpoint sync, a random block is given as starting point
- this is the tail block, and we require that the tail block has a
corresponding state.
When backfilling, we end up with blocks without corresponding states,
hence we cannot use `tail` as a backfill pointer - there is no state.
Nonetheless, we need to keep track of where we are in the backfill
process between restarts, such that we can answer GetBeaconBlocksByRange
requests.
This PR adds the basic support for backfill handling - it needs to be
integrated with backfill sync, and the REST API needs to be adjusted to
take advantage of the new backfilled blocks when responding to certain
requests.
Future work will also enable moving the tail in either direction:
* pruning means moving the tail forward in time and removing states
* backwards means recreating past states from genesis, such that
intermediate states are recreated step by step all the way to the tail -
at that point, tail, genesis and backfill will match up.
* backfilling is done when backfill != genesis - later, this will be the
WSS checkpoint instead
2021-12-13 13:36:06 +00:00
|
|
|
blocks[^2].toBlockId().atSlot()
|
2022-03-15 08:24:55 +00:00
|
|
|
dag2.getBlockIdAtSlot(dag.tail.slot - 2).isNone
|
2022-02-26 18:16:19 +00:00
|
|
|
dag2.backfill == blocks[^2].phase0Data.message.toBeaconBlockSummary()
|
