When the BN exits after writing new `head` to database, but before
completing the `updateFinalizedBlocks` call, the database is slightly
inconsistent due to the partial write. We currently fail to start up
after that. Fix that by catching up on partial `updateFinalizedBlocks`
tasks on start up, and add a test for this edge case.
Gnosis uses `MIN_EPOCHS_FOR_BLOCK_REQUESTS` = 33024, but the computed
safe minimum (that Nimbus was using) is 2304. Relax the compatibility
check to allow `MIN_EPOCHS_FOR_BLOCK_REQUESTS` above the safe minimum
and honor `config.yaml` preferences for `MIN_EPOCHS_FOR_BLOCK_REQUESTS`.
* ShufflingRef approach to next-epoch validator duty calculation/prediction
* refactor action_tracker.updateActions to take ShufflingRef + beacon_proposers; refactor maybeUpdateActionTrackerNextEpoch to be separate and reused function; add actual fallback logic
* document one possible set of conditions
* check epoch participation flags and inactivity scores to ensure no penalties and MAX_EFFECTIVE_BALANCE to ensure rewards don't matter
* correctly (un)shuffle each proposer index
* remove debugging assertion
For symmetry with `forkyState` when using `withState`, and to avoid
problems with shadowing of `blck` when using `withBlck` in `template`,
also rename the injected `blck` to `forkyBlck`.
- https://github.com/nim-lang/Nim/issues/22698
* async batch verification
When batch verification is done, the main thread is blocked reducing
concurrency.
With this PR, the new thread signalling primitive in chronos is used to
offload the full batch verification process to a separate thread
allowing the main threads to continue async operations while the other
threads verify signatures.
Similar to previous behavior, the number of ongoing batch verifications
is capped to prevent runaway resource usage.
In addition to the asynchronous processing, 3 addition changes help
drive throughput:
* A loop is used for batch accumulation: this prevents a stampede of
small batches in eager mode where both the eager and the scheduled batch
runner would pick batches off the queue, prematurely picking "fresh"
batches off the queue
* An additional small wait is introduced for small batches - this helps
create slightly larger batches which make better used of the increased
concurrency
* Up to 2 batches are scheduled to the threadpool during high pressure,
reducing startup latency for the threads
Together, these changes increase attestation verification throughput
under load up to 30%.
* fixup
* Update submodules
* fix blst build issues (and a PIC warning)
* bump
---------
Co-authored-by: Zahary Karadjov <zahary@gmail.com>
Split up the `ShufflingRef` acceleration logic into generically usable
parts and attester shuffling specific parts. The generic parts could be
used to accelerate other purposes, e.g., REST `/states/xxx/randao` API.
To enable additional use cases, e.g., `/states/###/randao` beacon API,
`ShufflingRef` acceleration logic needs to be able to operate on parts
of the DAG that do not have `BlockRef`. Changing `commonAncestor` to
act on `BlockId` instead of `BlockRef` is a step toward that and also
simplifies the logic some more.
Post-merge blocks contain all information to directly obtain RANDAO
without having to load any additional info. Take advantage of that to
further accelerate `ShufflingRef` computation. Note that it is still
necessary to verify that `blck` / `state` share a sufficiently recent
ancestor for the purpose of computing attester shufflings.
- new: 243.71s, 239.67s, 237.32s, 238.36s, 239.57s
- old: 251.33s, 234.29s, 249.28s, 237.03s, 236.78s
Current RANDAO recovery logic is quite complex as it optimizes for the
minimum amount of database reads. Loading blocks isn't the bottleneck
though, so rather make the implementation more concise by avoiding the
complex strategy planning step. Note that this also prepares for an even
faster implementation for post-merge blocks in the future that extracts
RANDAO from `ExecutionPayload` directly if available, so even in cases
where efficiency is slightly lower, only historical data is affected.
`time nim c -r tests/test_blockchain_dag` (cached binary):
- new: 145.45s, 133.59s, 144.65s, 127.69s, 136.14s
- old: 149.15s, 150.84s, 135.77s, 137.49s, 133.89s
* early exit `commonAncestor` when comparing with `finalizedHead`
As all `BlockRef` lead to `finalizedHead` (`parent == nil`),
can shortcut in that situation and immediately return `finalizedHead`
if passed as one of the arguments.
* typo in comment
* add test from #5152
Co-authored-by: tersec <tersec@users.noreply.github.com>
* add note about test complexity
* regenerate test summary
---------
Co-authored-by: tersec <tersec@users.noreply.github.com>
We have several modules that import `nim-eth` for the sole purpose of
its `keys.newRng` function. This function is meanwhile a simple wrapper
around `nim-bearssl`'s `HmacDrbgContext.new()`, so the import doesn't
really serve a use anymore. Replace `keys.newRng` with the direct call
to reduce `nim-eth` imports.
When an uncached `ShufflingRef` is requested, we currently replay state
which can take several seconds. Acceleration is possible by:
1. Start from any state with locked-in `get_active_validator_indices`.
Any blocks / slots applied to such a state can only affect that
result for future epochs, so are viable for querying target epoch.
`compute_activation_exit_epoch(state.slot.epoch) > target.epoch`
2. Determine highest common ancestor among `state` and `target.blck`.
At the ancestor slot, same rules re `get_active_validator_indices`.
`compute_activation_exit_epoch(ancestorSlot.epoch) > target.epoch`
3. We now have a `state` that shares history with `target.blck` up
through a common ancestor slot. Any blocks / slots that the `state`
contains, which are not part of the `target.blck` history, affect
`get_active_validator_indices` at epochs _after_ `target.epoch`.
4. Select `state.randao_mixes[N]` that is closest to common ancestor.
Either direction is fine (above / below ancestor).
5. From that RANDAO mix, mix in / out all RANDAO reveals from blocks
in-between. This is just an XOR operation, so fully reversible.
`mix = mix xor SHA256(blck.message.body.randao_reveal)`
6. Compute the attester dependent slot from `target.epoch`.
`if epoch >= 2: (target.epoch - 1).start_slot - 1 else: GENESIS_SLOT`
7. Trace back from `target.blck` to the attester dependent slot.
We now have the destination for which we want to obtain RANDAO.
8. Mix in all RANDAO reveals from blocks up through the `dependentBlck`.
Same method, no special handling necessary for epoch transitions.
9. Combine `get_active_validator_indices` from `state` at `target.epoch`
with the recovered RANDAO value at `dependentBlck` to obtain the
requested shuffling, and construct the `ShufflingRef` without replay.
* more tests and simplify logic
* test with different number of deposits per branch
* Update beacon_chain/consensus_object_pools/blockchain_dag.nim
Co-authored-by: Jacek Sieka <jacek@status.im>
* `commonAncestor` tests
* lint
---------
Co-authored-by: Jacek Sieka <jacek@status.im>
Introduce (optional) pruning of historical data - a pruned node will
continue to answer queries for historical data up to
`MIN_EPOCHS_FOR_BLOCK_REQUESTS` epochs, or roughly 5 months, capping
typical database usage at around 60-70gb.
To enable pruning, add `--history=prune` to the command line - on the
first start, old data will be cleared (which may take a while) - after
that, data is pruned continuously.
When pruning an existing database, the database will not shrink -
instead, the freed space is recycled as the node continues to run - to
free up space, perform a trusted node sync with a fresh database.
When switching on archive mode in a pruned node, history is retained
from that point onwards.
History pruning is scheduled to be enabled by default in a future
release.
In this PR, `minimal` mode from #4419 is not implemented meaning
retention periods for states and blocks are always the same - depending
on user demand, a future PR may implement `minimal` as well.
We currently use `BlockError` for both beacon blocks and LC objects.
In light of EIP4844, we will likely also use it for blob sidecars.
To avoid confusion, renaming it to a more generic `VerifierError`,
and update its documentation to be more generic.
To avoid long lines as a followup, also renaming the `block_processor`'s
`BlockProcessingCompleted.completed`->`ProcessingStatus.completed` and
`BlockProcessingCompleted.notCompleted`->`ProcessingStatus.notCompleted`
Currently, we require genesis and a checkpoint block and state to start
from an arbitrary slot - this PR relaxes this requirement so that we can
start with a state alone.
The current trusted-node-sync algorithm works by first downloading
blocks until we find an epoch aligned non-empty slot, then downloads the
state via slot.
However, current
[proposals](https://github.com/ethereum/beacon-APIs/pull/226) for
checkpointing prefer finalized state as
the main reference - this allows more simple access control and caching
on the server side - in particular, this should help checkpoint-syncing
from sources that have a fast `finalized` state download (like infura
and teku) but are slow when accessing state via slot.
Earlier versions of Nimbus will not be able to read databases created
without a checkpoint block and genesis. In most cases, backfilling makes
the database compatible except where genesis is also missing (custom
networks).
* backfill checkpoint block from libp2p instead of checkpoint source,
when doing trusted node sync
* allow starting the client without genesis / checkpoint block
* perform epoch start slot lookahead when loading tail state, so as to
deal with the case where the epoch start slot does not have a block
* replace `--blockId` with `--state-id` in TNS command line
* when replaying, also look at the parent of the last-known-block (even
if we don't have the parent block data, we can still replay from a
"parent" state) - in particular, this clears the way for implementing
state pruning
* deprecate `--finalized-checkpoint-block` option (no longer needed)
* Allow chain dag without genesis / block
This PR enables the initialization of the dag without access to blocks
or genesis state - it is a prerequisite for implementing a number of
interesting features:
* checkpoint sync without any block download
* pruning of blocks and states
* backfill checkpoint block
When EL `newPayload` is slow (e.g., Raspberry Pi with Besu), the epoch
and shuffling caches tend to fill up with multiple copies per epoch when
processing gossip and performing validator duties close to wall slot.
The old strategy of evicting oldest epoch led to the same item being
evicted over and over, leading to blocking of over 5 minutes in extreme
cases where alternate epochs/shuffling got loaded repeatedly.
Changing the cache eviction strategy to least-recently-used seems to
improve the situation drastically. A simple implementation was selected
based on single linked-list without a hashtable.
* avoid database race-condition inconsistency after fcU `INVALID` then crash
* ensure head doesn't fall behind finalized; add more tests for head movement/reloading DAG
In order to avoid full replays when validating attestations hailing from
untaken forks, it's better to keep shufflings separate from `EpochRef`
and perform a lookahead on the shuffling when processing the block that
determines them.
This also helps performance in the case where REST clients are trying to
perform lookahead on attestation duties and decreases memory usage by
sharing shufflings between EpochRef instances of the same dependent
root.
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 :)
One more step on the journey to reduce `BlockRef` usage across the
codebase - this one gets rid of `StateData` whose job was to keep track
of which block was last assigned to a state - these duties have now been
taken over by `latest_block_root`, a fairly recent addition that
computes this block root from state data (at a small cost that should be
insignificant)
99% mechanical change.
* fewer deps on `BlockRef` traversal in anticipation of pruning
* allows identifying EpochRef:s by their shuffling as a first step of
* tighten error handling around missing blocks
using the zero hash for signalling "missing block" is fragile and easy
to miss - with checkpoint sync now, and pruning in the future, missing
blocks become "normal".
When performing trusted node sync, historical access is limited to
states after the checkpoint.
Reindexing restores full historical access by replaying historical
blocks against the state and storing snapshots in the database.
The process can be initiated or resumed at any point in time.
This PR names and documents the concept of the archive: a range of slots
for which we have degraded functionality in terms of historical access -
in particular:
* we don't support rewinding to states in this range
* we don't keep an in-memory representation of the block dag
The archive de-facto exists in a trusted-node-synced node, but this PR
gives it a name and drops the in-memory digest index.
In order to satisfy `GetBlocksByRange` requests, we ensure that we have
blocks for the entire archive period via backfill. Future versions may
relax this further, adding a "pre-archive" period that is fully pruned.
During by-slot searches in the archive (both for libp2p and rest
requests), an extra database lookup is used to covert the given `slot`
to a `root` - future versions will avoid this using era files which
natively are indexed by `slot`. That said, the lookup is quite
fast compared to the actual block loading given how trivial the table
is - it's hard to measure, even.
A collateral benefit of this PR is that checkpoint-synced nodes will see
100-200MB memory usage savings, thanks to the dropped in-memory cache -
future pruning work will bring this benefit to full nodes as well.
* document chaindag storage architecture and assumptions
* look up parent using block id instead of full block in clearance
(future-proofing the code against a future in which blocks come from era
files)
* simplify finalized block init, always writing the backfill portion to
db at startup (to ensure lookups work as expected)
* preallocate some extra memory for finalized blocks, to avoid immediate
realloc
* 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
Etan Kissling
tersec
Jacek Sieka
henridf