2022-01-05 14:49:10 +00:00
|
|
|
# beacon_chain
|
|
|
|
|
# Copyright (c) 2018-2021 Status Research & Development GmbH
|
|
|
|
|
# Licensed and distributed under either of
|
|
|
|
|
# * 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).
|
|
|
|
|
# at your option. This file may not be copied, modified, or distributed except according to those terms.
|
|
|
|
|
|
|
|
|
|
import
|
|
|
|
|
chronos,
|
|
|
|
|
chronicles,
|
2022-03-16 07:20:40 +00:00
|
|
|
../spec/beaconstate,
|
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
|
|
|
../consensus_object_pools/blockchain_dag
|
2022-01-05 14:49:10 +00:00
|
|
|
|
|
|
|
|
type
|
|
|
|
|
CacheEntry = ref object
|
2022-03-16 07:20:40 +00:00
|
|
|
state: ref ForkedHashedBeaconState
|
2022-01-05 14:49:10 +00:00
|
|
|
lastUsed: Moment
|
|
|
|
|
|
|
|
|
|
# This is ref object because we need to capture it by
|
|
|
|
|
# reference in the `scheduleEntryExpiration` function.
|
|
|
|
|
StateTtlCache* = ref object
|
|
|
|
|
entries: seq[CacheEntry]
|
|
|
|
|
ttl: Duration
|
|
|
|
|
|
|
|
|
|
const
|
|
|
|
|
slotDifferenceForCacheHit = 5 * SLOTS_PER_EPOCH
|
|
|
|
|
|
|
|
|
|
logScope:
|
|
|
|
|
topics = "state_ttl_cache"
|
|
|
|
|
|
|
|
|
|
proc init*(T: type StateTtlCache,
|
|
|
|
|
cacheSize: Natural,
|
|
|
|
|
cacheTtl: Duration): T =
|
|
|
|
|
doAssert cacheSize > 0
|
|
|
|
|
|
|
|
|
|
StateTtlCache(
|
|
|
|
|
entries: newSeq[CacheEntry](cacheSize),
|
|
|
|
|
ttl: cacheTtl)
|
|
|
|
|
|
|
|
|
|
proc scheduleEntryExpiration(cache: StateTtlCache,
|
|
|
|
|
entryIdx: int) =
|
|
|
|
|
proc removeElement(arg: pointer) =
|
|
|
|
|
if cache.entries[entryIdx] == nil:
|
|
|
|
|
return
|
|
|
|
|
let expirationTime = cache.entries[entryIdx].lastUsed + cache.ttl
|
|
|
|
|
if expirationTime > Moment.now:
|
|
|
|
|
return
|
|
|
|
|
cache.entries[entryIdx] = nil
|
|
|
|
|
debug "Cached REST state expired", index = entryIdx
|
|
|
|
|
|
|
|
|
|
discard setTimer(Moment.now + cache.ttl, removeElement)
|
|
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
proc add*(cache: StateTtlCache, state: ref ForkedHashedBeaconState) =
|
2022-01-05 14:49:10 +00:00
|
|
|
var
|
|
|
|
|
now = Moment.now
|
|
|
|
|
lruTime = now
|
|
|
|
|
index = -1
|
|
|
|
|
|
|
|
|
|
for i in 0 ..< cache.entries.len:
|
|
|
|
|
if cache.entries[i] == nil:
|
|
|
|
|
index = i
|
|
|
|
|
break
|
|
|
|
|
if cache.entries[i].lastUsed <= lruTime:
|
|
|
|
|
index = i
|
|
|
|
|
lruTime = cache.entries[i].lastUsed
|
|
|
|
|
|
|
|
|
|
doAssert index != -1
|
|
|
|
|
cache.entries[index] = CacheEntry(state: state, lastUsed: now)
|
|
|
|
|
debug "Cached REST state added", index = index
|
|
|
|
|
|
|
|
|
|
cache.scheduleEntryExpiration(index)
|
|
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
proc getClosestState*(
|
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
|
|
|
cache: StateTtlCache, dag: ChainDAGRef,
|
|
|
|
|
bsi: BlockSlotId): ref ForkedHashedBeaconState =
|
2022-01-05 14:49:10 +00:00
|
|
|
var
|
|
|
|
|
bestSlotDifference = Slot.high
|
|
|
|
|
index = -1
|
|
|
|
|
|
|
|
|
|
for i in 0 ..< cache.entries.len:
|
|
|
|
|
if cache.entries[i] == nil:
|
|
|
|
|
continue
|
|
|
|
|
|
2022-03-16 07:20:40 +00:00
|
|
|
let stateSlot = getStateField(cache.entries[i][].state[], 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
|
|
|
if stateSlot > bsi.slot:
|
2022-01-05 14:49:10 +00:00
|
|
|
# We can use only states that can be advanced forward in time.
|
|
|
|
|
continue
|
|
|
|
|
|
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
|
|
|
let slotDifference = bsi.slot - stateSlot
|
2022-01-05 14:49:10 +00:00
|
|
|
if slotDifference > slotDifferenceForCacheHit:
|
|
|
|
|
# The state is too old to be useful as a rewind starting point.
|
|
|
|
|
continue
|
|
|
|
|
|
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
|
|
|
var cur = bsi
|
2022-01-05 14:49:10 +00:00
|
|
|
for j in 0 ..< slotDifference:
|
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
|
|
|
cur = dag.parentOrSlot(cur).valueOr:
|
|
|
|
|
break
|
2022-01-05 14:49:10 +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
|
|
|
if not cache.entries[i].state[].matches_block(cur.bid.root):
|
2022-01-05 14:49:10 +00:00
|
|
|
# The cached state and the requested BlockSlot are at different branches
|
|
|
|
|
# of history.
|
|
|
|
|
continue
|
|
|
|
|
|
|
|
|
|
if slotDifference < bestSlotDifference:
|
|
|
|
|
bestSlotDifference = slotDifference.Slot
|
|
|
|
|
index = i
|
|
|
|
|
|
|
|
|
|
if index == -1:
|
|
|
|
|
return nil
|
|
|
|
|
|
|
|
|
|
cache.entries[index].lastUsed = Moment.now
|
|
|
|
|
cache.scheduleEntryExpiration(index)
|
|
|
|
|
|
|
|
|
|
return cache.entries[index].state
|
