Move processing cache out of DA (#5420)

* Move processing cache out of DA

* Merge branch 'sigp/unstable' into non-da-processing-cach

* Merge branch 'unstable' of https://github.com/sigp/lighthouse into non-da-processing-cache

* remove unused file, remove outdated TODO, add is_deneb check to missing blob id calculations

* Merge branch 'unstable' of https://github.com/sigp/lighthouse into non-da-processing-cache

* fix lints
This commit is contained in:
Lion - dapplion
2024-04-11 02:19:40 +09:00
committed by GitHub
parent b1f9751a69
commit 30dc260472
10 changed files with 82 additions and 209 deletions

View File

@@ -1,4 +1,4 @@
use crate::{BeaconChain, BeaconChainError, BeaconChainTypes};
use crate::{metrics, BeaconChain, BeaconChainError, BeaconChainTypes};
use execution_layer::{ExecutionLayer, ExecutionPayloadBodyV1};
use slog::{crit, debug, Logger};
use std::collections::HashMap;
@@ -412,8 +412,13 @@ impl<T: BeaconChainTypes> BeaconBlockStreamer<T> {
fn check_caches(&self, root: Hash256) -> Option<Arc<SignedBeaconBlock<T::EthSpec>>> {
if self.check_caches == CheckCaches::Yes {
self.beacon_chain
.data_availability_checker
.get_block(&root)
.reqresp_pre_import_cache
.read()
.get(&root)
.map(|block| {
metrics::inc_counter(&metrics::BEACON_REQRESP_PRE_IMPORT_CACHE_HITS);
block.clone()
})
.or(self.beacon_chain.early_attester_cache.get_block(root))
} else {
None

View File

@@ -359,6 +359,9 @@ pub type BeaconStore<T> = Arc<
>,
>;
/// Cache gossip verified blocks to serve over ReqResp before they are imported
type ReqRespPreImportCache<E> = HashMap<Hash256, Arc<SignedBeaconBlock<E>>>;
/// Represents the "Beacon Chain" component of Ethereum 2.0. Allows import of blocks and block
/// operations and chooses a canonical head.
pub struct BeaconChain<T: BeaconChainTypes> {
@@ -461,6 +464,8 @@ pub struct BeaconChain<T: BeaconChainTypes> {
pub(crate) attester_cache: Arc<AttesterCache>,
/// A cache used when producing attestations whilst the head block is still being imported.
pub early_attester_cache: EarlyAttesterCache<T::EthSpec>,
/// Cache gossip verified blocks to serve over ReqResp before they are imported
pub reqresp_pre_import_cache: Arc<RwLock<ReqRespPreImportCache<T::EthSpec>>>,
/// A cache used to keep track of various block timings.
pub block_times_cache: Arc<RwLock<BlockTimesCache>>,
/// A cache used to track pre-finalization block roots for quick rejection.
@@ -2898,8 +2903,6 @@ impl<T: BeaconChainTypes> BeaconChain<T> {
}
}
self.data_availability_checker
.notify_gossip_blob(block_root, &blob);
let r = self.check_gossip_blob_availability_and_import(blob).await;
self.remove_notified(&block_root, r)
}
@@ -2932,8 +2935,6 @@ impl<T: BeaconChainTypes> BeaconChain<T> {
}
}
self.data_availability_checker
.notify_rpc_blobs(block_root, &blobs);
let r = self
.check_rpc_blob_availability_and_import(slot, block_root, blobs)
.await;
@@ -2950,7 +2951,7 @@ impl<T: BeaconChainTypes> BeaconChain<T> {
let has_missing_components =
matches!(r, Ok(AvailabilityProcessingStatus::MissingComponents(_, _)));
if !has_missing_components {
self.data_availability_checker.remove_notified(block_root);
self.reqresp_pre_import_cache.write().remove(block_root);
}
r
}
@@ -2963,8 +2964,10 @@ impl<T: BeaconChainTypes> BeaconChain<T> {
unverified_block: B,
notify_execution_layer: NotifyExecutionLayer,
) -> Result<AvailabilityProcessingStatus, BlockError<T::EthSpec>> {
self.data_availability_checker
.notify_block(block_root, unverified_block.block_cloned());
self.reqresp_pre_import_cache
.write()
.insert(block_root, unverified_block.block_cloned());
let r = self
.process_block(block_root, unverified_block, notify_execution_layer, || {
Ok(())

View File

@@ -955,6 +955,7 @@ where
validator_pubkey_cache: TimeoutRwLock::new(validator_pubkey_cache),
attester_cache: <_>::default(),
early_attester_cache: <_>::default(),
reqresp_pre_import_cache: <_>::default(),
light_client_server_cache: LightClientServerCache::new(),
light_client_server_tx: self.light_client_server_tx,
shutdown_sender: self

View File

@@ -7,11 +7,8 @@ pub use crate::data_availability_checker::availability_view::{
};
pub use crate::data_availability_checker::child_components::ChildComponents;
use crate::data_availability_checker::overflow_lru_cache::OverflowLRUCache;
use crate::data_availability_checker::processing_cache::ProcessingCache;
use crate::{BeaconChain, BeaconChainTypes, BeaconStore};
use kzg::Kzg;
use parking_lot::RwLock;
pub use processing_cache::ProcessingComponents;
use slasher::test_utils::E;
use slog::{debug, error, Logger};
use slot_clock::SlotClock;
@@ -21,7 +18,6 @@ use std::fmt::Debug;
use std::num::NonZeroUsize;
use std::sync::Arc;
use task_executor::TaskExecutor;
use types::beacon_block_body::KzgCommitmentOpts;
use types::blob_sidecar::{BlobIdentifier, BlobSidecar, FixedBlobSidecarList};
use types::{BlobSidecarList, ChainSpec, Epoch, EthSpec, Hash256, SignedBeaconBlock};
@@ -29,7 +25,6 @@ mod availability_view;
mod child_components;
mod error;
mod overflow_lru_cache;
mod processing_cache;
mod state_lru_cache;
pub use error::{Error as AvailabilityCheckError, ErrorCategory as AvailabilityCheckErrorCategory};
@@ -50,7 +45,6 @@ pub const STATE_LRU_CAPACITY: usize = STATE_LRU_CAPACITY_NON_ZERO.get();
/// `DataAvailabilityChecker` is responsible for KZG verification of block components as well as
/// checking whether a "availability check" is required at all.
pub struct DataAvailabilityChecker<T: BeaconChainTypes> {
processing_cache: RwLock<ProcessingCache<T::EthSpec>>,
availability_cache: Arc<OverflowLRUCache<T>>,
slot_clock: T::SlotClock,
kzg: Option<Arc<Kzg>>,
@@ -89,7 +83,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
) -> Result<Self, AvailabilityCheckError> {
let overflow_cache = OverflowLRUCache::new(OVERFLOW_LRU_CAPACITY, store, spec.clone())?;
Ok(Self {
processing_cache: <_>::default(),
availability_cache: Arc::new(overflow_cache),
slot_clock,
log: log.clone(),
@@ -98,17 +91,25 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
})
}
/// Checks if the given block root is cached.
/// Checks if the block root is currenlty in the availability cache awaiting processing because
/// of missing components.
pub fn has_block(&self, block_root: &Hash256) -> bool {
self.processing_cache.read().has_block(block_root)
self.availability_cache.has_block(block_root)
}
/// Get the processing info for a block.
pub fn get_processing_components(
&self,
block_root: Hash256,
) -> Option<ProcessingComponents<T::EthSpec>> {
self.processing_cache.read().get(&block_root).cloned()
pub fn get_missing_blob_ids_with(&self, block_root: Hash256) -> MissingBlobs {
self.availability_cache
.with_pending_components(&block_root, |pending_components| match pending_components {
Some(pending_components) => self.get_missing_blob_ids(
block_root,
pending_components
.get_cached_block()
.as_ref()
.map(|b| b.as_block()),
&pending_components.verified_blobs,
),
None => MissingBlobs::new_without_block(block_root, self.is_deneb()),
})
}
/// If there's no block, all possible ids will be returned that don't exist in the given blobs.
@@ -116,7 +117,7 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
pub fn get_missing_blob_ids<V>(
&self,
block_root: Hash256,
block: &Option<Arc<SignedBeaconBlock<T::EthSpec>>>,
block: Option<&SignedBeaconBlock<T::EthSpec>>,
blobs: &FixedVector<Option<V>, <T::EthSpec as EthSpec>::MaxBlobsPerBlock>,
) -> MissingBlobs {
let Some(current_slot) = self.slot_clock.now_or_genesis() else {
@@ -132,10 +133,15 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
if self.da_check_required_for_epoch(current_epoch) {
match block {
Some(cached_block) => {
let block_commitments = cached_block.get_commitments();
let block_commitments_len = cached_block
.message()
.body()
.blob_kzg_commitments()
.map(|v| v.len())
.unwrap_or(0);
let blob_ids = blobs
.iter()
.take(block_commitments.len())
.take(block_commitments_len)
.enumerate()
.filter_map(|(index, blob_commitment_opt)| {
blob_commitment_opt.is_none().then_some(BlobIdentifier {
@@ -163,14 +169,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
self.availability_cache.peek_blob(blob_id)
}
/// Get a block from the availability cache. Includes any blocks we are currently processing.
pub fn get_block(&self, block_root: &Hash256) -> Option<Arc<SignedBeaconBlock<T::EthSpec>>> {
self.processing_cache
.read()
.get(block_root)
.and_then(|cached| cached.block.clone())
}
/// Put a list of blobs received via RPC into the availability cache. This performs KZG
/// verification on the blobs in the list.
pub fn put_rpc_blobs(
@@ -323,52 +321,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
block.num_expected_blobs() > 0 && self.da_check_required_for_epoch(block.epoch())
}
/// Adds a block to the processing cache. This block's commitments are unverified but caching
/// them here is useful to avoid duplicate downloads of blocks, as well as understanding
/// our blob download requirements. We will also serve this over RPC.
pub fn notify_block(&self, block_root: Hash256, block: Arc<SignedBeaconBlock<T::EthSpec>>) {
self.processing_cache
.write()
.entry(block_root)
.or_default()
.merge_block(block);
}
/// Add a single blob commitment to the processing cache. This commitment is unverified but caching
/// them here is useful to avoid duplicate downloads of blobs, as well as understanding
/// our block and blob download requirements.
pub fn notify_gossip_blob(&self, block_root: Hash256, blob: &GossipVerifiedBlob<T>) {
let index = blob.index();
let commitment = blob.kzg_commitment();
self.processing_cache
.write()
.entry(block_root)
.or_default()
.merge_single_blob(index as usize, commitment);
}
/// Adds blob commitments to the processing cache. These commitments are unverified but caching
/// them here is useful to avoid duplicate downloads of blobs, as well as understanding
/// our block and blob download requirements.
pub fn notify_rpc_blobs(&self, block_root: Hash256, blobs: &FixedBlobSidecarList<T::EthSpec>) {
let mut commitments = KzgCommitmentOpts::<T::EthSpec>::default();
for blob in blobs.iter().flatten() {
if let Some(commitment) = commitments.get_mut(blob.index as usize) {
*commitment = Some(blob.kzg_commitment);
}
}
self.processing_cache
.write()
.entry(block_root)
.or_default()
.merge_blobs(commitments);
}
/// Clears the block and all blobs from the processing cache for a give root if they exist.
pub fn remove_notified(&self, block_root: &Hash256) {
self.processing_cache.write().remove(block_root)
}
/// The epoch at which we require a data availability check in block processing.
/// `None` if the `Deneb` fork is disabled.
pub fn data_availability_boundary(&self) -> Option<Epoch> {
@@ -410,7 +362,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
/// Collects metrics from the data availability checker.
pub fn metrics(&self) -> DataAvailabilityCheckerMetrics {
DataAvailabilityCheckerMetrics {
processing_cache_size: self.processing_cache.read().len(),
num_store_entries: self.availability_cache.num_store_entries(),
state_cache_size: self.availability_cache.state_cache_size(),
block_cache_size: self.availability_cache.block_cache_size(),
@@ -420,7 +371,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
/// Helper struct to group data availability checker metrics.
pub struct DataAvailabilityCheckerMetrics {
pub processing_cache_size: usize,
pub num_store_entries: usize,
pub state_cache_size: usize,
pub block_cache_size: usize,

View File

@@ -2,7 +2,6 @@ use super::state_lru_cache::DietAvailabilityPendingExecutedBlock;
use crate::blob_verification::KzgVerifiedBlob;
use crate::block_verification_types::AsBlock;
use crate::data_availability_checker::overflow_lru_cache::PendingComponents;
use crate::data_availability_checker::ProcessingComponents;
use kzg::KzgCommitment;
use ssz_types::FixedVector;
use std::sync::Arc;
@@ -178,14 +177,6 @@ macro_rules! impl_availability_view {
};
}
impl_availability_view!(
ProcessingComponents,
Arc<SignedBeaconBlock<E>>,
KzgCommitment,
block,
blob_commitments
);
impl_availability_view!(
PendingComponents,
DietAvailabilityPendingExecutedBlock<E>,
@@ -293,32 +284,6 @@ pub mod tests {
(block, blobs, invalid_blobs)
}
type ProcessingViewSetup<E> = (
Arc<SignedBeaconBlock<E>>,
FixedVector<Option<KzgCommitment>, <E as EthSpec>::MaxBlobsPerBlock>,
FixedVector<Option<KzgCommitment>, <E as EthSpec>::MaxBlobsPerBlock>,
);
pub fn setup_processing_components(
block: SignedBeaconBlock<E>,
valid_blobs: FixedVector<Option<Arc<BlobSidecar<E>>>, <E as EthSpec>::MaxBlobsPerBlock>,
invalid_blobs: FixedVector<Option<Arc<BlobSidecar<E>>>, <E as EthSpec>::MaxBlobsPerBlock>,
) -> ProcessingViewSetup<E> {
let blobs = FixedVector::from(
valid_blobs
.iter()
.map(|blob_opt| blob_opt.as_ref().map(|blob| blob.kzg_commitment))
.collect::<Vec<_>>(),
);
let invalid_blobs = FixedVector::from(
invalid_blobs
.iter()
.map(|blob_opt| blob_opt.as_ref().map(|blob| blob.kzg_commitment))
.collect::<Vec<_>>(),
);
(Arc::new(block), blobs, invalid_blobs)
}
type PendingComponentsSetup<E> = (
DietAvailabilityPendingExecutedBlock<E>,
FixedVector<Option<KzgVerifiedBlob<E>>, <E as EthSpec>::MaxBlobsPerBlock>,
@@ -490,13 +455,6 @@ pub mod tests {
};
}
generate_tests!(
processing_components_tests,
ProcessingComponents::<E>,
kzg_commitments,
processing_blobs,
setup_processing_components
);
generate_tests!(
pending_components_tests,
PendingComponents<E>,

View File

@@ -305,6 +305,11 @@ impl<T: BeaconChainTypes> Critical<T> {
Ok(())
}
/// Returns true if the block root is known, without altering the LRU ordering
pub fn has_block(&self, block_root: &Hash256) -> bool {
self.in_memory.peek(block_root).is_some() || self.store_keys.contains(block_root)
}
/// This only checks for the blobs in memory
pub fn peek_blob(
&self,
@@ -322,6 +327,13 @@ impl<T: BeaconChainTypes> Critical<T> {
}
}
pub fn peek_pending_components(
&self,
block_root: &Hash256,
) -> Option<&PendingComponents<T::EthSpec>> {
self.in_memory.peek(block_root)
}
/// Puts the pending components in the LRU cache. If the cache
/// is at capacity, the LRU entry is written to the store first
pub fn put_pending_components(
@@ -409,6 +421,11 @@ impl<T: BeaconChainTypes> OverflowLRUCache<T> {
})
}
/// Returns true if the block root is known, without altering the LRU ordering
pub fn has_block(&self, block_root: &Hash256) -> bool {
self.critical.read().has_block(block_root)
}
/// Fetch a blob from the cache without affecting the LRU ordering
pub fn peek_blob(
&self,
@@ -425,6 +442,14 @@ impl<T: BeaconChainTypes> OverflowLRUCache<T> {
}
}
pub fn with_pending_components<R, F: FnOnce(Option<&PendingComponents<T::EthSpec>>) -> R>(
&self,
block_root: &Hash256,
f: F,
) -> R {
f(self.critical.read().peek_pending_components(block_root))
}
pub fn put_kzg_verified_blobs<I: IntoIterator<Item = KzgVerifiedBlob<T::EthSpec>>>(
&self,
block_root: Hash256,

View File

@@ -1,63 +0,0 @@
use crate::data_availability_checker::AvailabilityView;
use std::collections::hash_map::Entry;
use std::collections::HashMap;
use std::sync::Arc;
use types::beacon_block_body::KzgCommitmentOpts;
use types::{EthSpec, Hash256, SignedBeaconBlock};
/// This cache is used only for gossip blocks/blobs and single block/blob lookups, to give req/resp
/// a view of what we have and what we require. This cache serves a slightly different purpose than
/// gossip caches because it allows us to process duplicate blobs that are valid in gossip.
/// See `AvailabilityView`'s trait definition.
#[derive(Default)]
pub struct ProcessingCache<E: EthSpec> {
processing_cache: HashMap<Hash256, ProcessingComponents<E>>,
}
impl<E: EthSpec> ProcessingCache<E> {
pub fn get(&self, block_root: &Hash256) -> Option<&ProcessingComponents<E>> {
self.processing_cache.get(block_root)
}
pub fn entry(&mut self, block_root: Hash256) -> Entry<'_, Hash256, ProcessingComponents<E>> {
self.processing_cache.entry(block_root)
}
pub fn remove(&mut self, block_root: &Hash256) {
self.processing_cache.remove(block_root);
}
pub fn has_block(&self, block_root: &Hash256) -> bool {
self.processing_cache
.get(block_root)
.map_or(false, |b| b.block_exists())
}
pub fn len(&self) -> usize {
self.processing_cache.len()
}
}
#[derive(Default, Debug, Clone)]
pub struct ProcessingComponents<E: EthSpec> {
/// Blobs required for a block can only be known if we have seen the block. So `Some` here
/// means we've seen it, a `None` means we haven't. The `kzg_commitments` value helps us figure
/// out whether incoming blobs actually match the block.
pub block: Option<Arc<SignedBeaconBlock<E>>>,
/// `KzgCommitments` for blobs are always known, even if we haven't seen the block. See
/// `AvailabilityView`'s trait definition for more details.
pub blob_commitments: KzgCommitmentOpts<E>,
}
impl<E: EthSpec> ProcessingComponents<E> {
pub fn new() -> Self {
Self::default()
}
}
// Not safe for use outside of tests as this always required a slot.
#[cfg(test)]
impl<E: EthSpec> ProcessingComponents<E> {
pub fn empty(_block_root: Hash256) -> Self {
Self {
block: None,
blob_commitments: KzgCommitmentOpts::<E>::default(),
}
}
}

View File

@@ -304,6 +304,14 @@ lazy_static! {
"Count of times the early attester cache returns an attestation"
);
pub static ref BEACON_REQRESP_PRE_IMPORT_CACHE_SIZE: Result<IntGauge> = try_create_int_gauge(
"beacon_reqresp_pre_import_cache_size",
"Current count of items of the reqresp pre import cache"
);
pub static ref BEACON_REQRESP_PRE_IMPORT_CACHE_HITS: Result<IntCounter> = try_create_int_counter(
"beacon_reqresp_pre_import_cache_hits",
"Count of times the reqresp pre import cache returns an item"
);
}
// Second lazy-static block is used to account for macro recursion limit.
@@ -1129,15 +1137,9 @@ lazy_static! {
Ok(vec![0.1, 0.2, 0.3,0.4,0.5,0.75,1.0,1.25,1.5,1.75,2.0,2.5,3.0,3.5,4.0,5.0,6.0,7.0,8.0,9.0,10.0,15.0,20.0])
);
/*
* Data Availability cache metrics
*/
pub static ref DATA_AVAILABILITY_PROCESSING_CACHE_SIZE: Result<IntGauge> =
try_create_int_gauge(
"data_availability_processing_cache_size",
"Number of entries in the data availability processing cache."
);
pub static ref DATA_AVAILABILITY_OVERFLOW_MEMORY_BLOCK_CACHE_SIZE: Result<IntGauge> =
try_create_int_gauge(
"data_availability_overflow_memory_block_cache_size",
@@ -1196,11 +1198,12 @@ pub fn scrape_for_metrics<T: BeaconChainTypes>(beacon_chain: &BeaconChain<T>) {
)
}
let da_checker_metrics = beacon_chain.data_availability_checker.metrics();
set_gauge_by_usize(
&DATA_AVAILABILITY_PROCESSING_CACHE_SIZE,
da_checker_metrics.processing_cache_size,
&BEACON_REQRESP_PRE_IMPORT_CACHE_SIZE,
beacon_chain.reqresp_pre_import_cache.read().len(),
);
let da_checker_metrics = beacon_chain.data_availability_checker.metrics();
set_gauge_by_usize(
&DATA_AVAILABILITY_OVERFLOW_MEMORY_BLOCK_CACHE_SIZE,
da_checker_metrics.block_cache_size,

View File

@@ -278,19 +278,11 @@ impl<L: Lookup, T: BeaconChainTypes> SingleBlockLookup<L, T> {
if let Some(components) = self.child_components.as_ref() {
self.da_checker.get_missing_blob_ids(
block_root,
&components.downloaded_block,
components.downloaded_block.as_ref().map(|b| b.as_ref()),
&components.downloaded_blobs,
)
} else {
let Some(processing_components) = self.da_checker.get_processing_components(block_root)
else {
return MissingBlobs::new_without_block(block_root, self.da_checker.is_deneb());
};
self.da_checker.get_missing_blob_ids(
block_root,
&processing_components.block,
&processing_components.blob_commitments,
)
self.da_checker.get_missing_blob_ids_with(block_root)
}
}

View File

@@ -1077,7 +1077,6 @@ fn test_same_chain_race_condition() {
mod deneb_only {
use super::*;
use crate::sync::SyncMessage;
use beacon_chain::data_availability_checker::AvailabilityCheckError;
use ssz_types::VariableList;