60% state replay speedup (#4434)

* 60% state replay speedup

* don't use HashList for epoch participation - in addition to the code
currently clearing the caches several times redundantly, clearing has to
be done each block nullifying the benefit (35%)
* introduce active balance cache - computing it is slow due to cache
unfriendliness in the random access pattern and bounds checking and we
do it for every block - this cache follows the same update pattern as
the active validator index cache (20%)
* avoid recomputing base reward several times per attestation (5%)

Applying 1024 blocks goes from 20s to ~8s on my laptop - these kinds of
requests happen on historical REST queries but also whenever there's a
reorg.

* fix test and diffs

beacon_chain/spec/beaconstate.nim

@@ -637,8 +637,13 @@ func get_total_active_balance*(state: ForkyBeaconState, cache: var StateCache):
 
   let epoch = state.get_current_epoch()
 
-  get_total_balance(
-    state, cache.get_shuffled_active_validator_indices(state, epoch))
+  cache.total_active_balance.withValue(epoch, tab) do:
+    return tab[]
+  do:
+    let tab = get_total_balance(
+      state, cache.get_shuffled_active_validator_indices(state, epoch))
+    cache.total_active_balance[epoch] = tab
+    return tab
 
 # https://github.com/ethereum/consensus-specs/blob/v1.3.0-alpha.2/specs/altair/beacon-chain.md#get_base_reward_per_increment
 func get_base_reward_per_increment_sqrt*(
@@ -704,15 +709,15 @@ func get_proposer_reward*(state: ForkyBeaconState,
     state, attestation.data, state.slot - attestation.data.slot)
   for index in get_attesting_indices(
       state, attestation.data, attestation.aggregation_bits, cache):
+    let
+      base_reward = get_base_reward(state, index, base_reward_per_increment)
     for flag_index, weight in PARTICIPATION_FLAG_WEIGHTS:
       if flag_index in participation_flag_indices and
          not has_flag(epoch_participation.item(index), flag_index):
-        epoch_participation[index] =
+        asList(epoch_participation)[index] =
           add_flag(epoch_participation.item(index), flag_index)
         # these are all valid; TODO statically verify or do it type-safely
-        result += get_base_reward(
-          state, index, base_reward_per_increment) * weight.uint64
-  epoch_participation.asHashList.clearCache()
+        result += base_reward * weight.uint64
 
   let proposer_reward_denominator =
     (WEIGHT_DENOMINATOR.uint64 - PROPOSER_WEIGHT.uint64) *
@@ -860,8 +865,7 @@ func upgrade_to_altair*(cfg: RuntimeConfig, pre: phase0.BeaconState):
     empty_participation: EpochParticipationFlags
     inactivity_scores = HashList[uint64, Limit VALIDATOR_REGISTRY_LIMIT]()
 
-  doAssert empty_participation.data.setLen(pre.validators.len)
-  empty_participation.asHashList.resetCache()
+  doAssert empty_participation.asList.setLen(pre.validators.len)
 
   doAssert inactivity_scores.data.setLen(pre.validators.len)
   inactivity_scores.resetCache()

beacon_chain/spec/datatypes/altair.nim

@@ -78,7 +78,7 @@ type
   ParticipationFlags* = uint8
 
   EpochParticipationFlags* =
-    distinct HashList[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]
+    distinct List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]
 
   # https://github.com/ethereum/consensus-specs/blob/v1.3.0-alpha.2/specs/altair/beacon-chain.md#syncaggregate
   SyncAggregate* = object
@@ -558,10 +558,8 @@ type
     # Represent in full; for the next epoch, current_epoch_participation in
     # epoch n is previous_epoch_participation in epoch n+1 but this doesn't
     # generalize.
-    previous_epoch_participation*:
-      List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]
-    current_epoch_participation*:
-      List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]
+    previous_epoch_participation*: EpochParticipationFlags
+    current_epoch_participation*: EpochParticipationFlags
 
     justification_bits*: JustificationBits
     previous_justified_checkpoint*: Checkpoint
@@ -589,26 +587,44 @@ template `[]`*(arr: array[SYNC_COMMITTEE_SIZE, auto] | seq;
 makeLimitedU8(SyncSubcommitteeIndex, SYNC_COMMITTEE_SUBNET_COUNT)
 makeLimitedU16(IndexInSyncCommittee, SYNC_COMMITTEE_SIZE)
 
-template asHashList*(epochFlags: EpochParticipationFlags): untyped =
-  HashList[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT] epochFlags
+template asList*(epochFlags: EpochParticipationFlags): untyped =
+  List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT] epochFlags
+template asList*(epochFlags: var EpochParticipationFlags): untyped =
+  let tmp = cast[ptr List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]](addr epochFlags)
+  tmp[]
+
+template asSeq*(epochFlags: EpochParticipationFlags): untyped =
+  seq[ParticipationFlags] asList(epochFlags)
+
+template asSeq*(epochFlags: var EpochParticipationFlags): untyped =
+  let tmp = cast[ptr seq[ParticipationFlags]](addr epochFlags)
+  tmp[]
 
 template item*(epochFlags: EpochParticipationFlags, idx: ValidatorIndex): ParticipationFlags =
-  asHashList(epochFlags).item(idx)
+  asList(epochFlags)[idx]
 
-template `[]`*(epochFlags: EpochParticipationFlags, idx: ValidatorIndex|uint64): ParticipationFlags =
-  asHashList(epochFlags)[idx]
+template `[]`*(epochFlags: EpochParticipationFlags, idx: ValidatorIndex|uint64|int): ParticipationFlags =
+  asList(epochFlags)[idx]
 
 template `[]=`*(epochFlags: EpochParticipationFlags, idx: ValidatorIndex, flags: ParticipationFlags) =
-  asHashList(epochFlags)[idx] = flags
+  asList(epochFlags)[idx] = flags
 
 template add*(epochFlags: var EpochParticipationFlags, flags: ParticipationFlags): bool =
-  asHashList(epochFlags).add flags
+  asList(epochFlags).add flags
 
 template len*(epochFlags: EpochParticipationFlags): int =
-  asHashList(epochFlags).len
-
-template data*(epochFlags: EpochParticipationFlags): untyped =
-  asHashList(epochFlags).data
+  asList(epochFlags).len
+
+template low*(epochFlags: EpochParticipationFlags): int =
+  asSeq(epochFlags).low
+template high*(epochFlags: EpochParticipationFlags): int =
+  asSeq(epochFlags).high
+
+template assign*(v: var EpochParticipationFlags, src: EpochParticipationFlags) =
+  # TODO https://github.com/nim-lang/Nim/issues/21123
+  mixin assign
+  var tmp = cast[ptr seq[ParticipationFlags]](addr v)
+  assign(tmp[], distinctBase src)
 
 func shortLog*(v: SomeBeaconBlock): auto =
   (

beacon_chain/spec/datatypes/base.nim

@@ -406,6 +406,7 @@ type
   # This doesn't know about forks or branches in the DAG. It's for straight,
   # linear chunks of the chain.
   StateCache* = object
+    total_active_balance*: Table[Epoch, Gwei]
     shuffled_active_validator_indices*: Table[Epoch, seq[ValidatorIndex]]
     beacon_proposer_indices*: Table[Slot, Option[ValidatorIndex]]
     sync_committees*: Table[SyncCommitteePeriod, SyncCommitteeCache]
@@ -923,6 +924,14 @@ func prune*(cache: var StateCache, epoch: Epoch) =
     pruneEpoch = epoch - 2
 
   var drops: seq[Slot]
+  block:
+    for k in cache.total_active_balance.keys:
+      if k < pruneEpoch:
+        drops.add pruneEpoch.start_slot
+    for drop in drops:
+      cache.total_active_balance.del drop.epoch
+    drops.setLen(0)
+
   block:
     for k in cache.shuffled_active_validator_indices.keys:
       if k < pruneEpoch:
@@ -948,6 +957,7 @@ func prune*(cache: var StateCache, epoch: Epoch) =
     drops.setLen(0)
 
 func clear*(cache: var StateCache) =
+  cache.total_active_balance.clear
   cache.shuffled_active_validator_indices.clear
   cache.beacon_proposer_indices.clear
   cache.sync_committees.clear

beacon_chain/spec/eth2_apis/eth2_rest_serialization.nim

@@ -612,15 +612,12 @@ proc readValue*(reader: var JsonReader[RestJson], value: var Epoch) {.
 proc writeValue*(writer: var JsonWriter[RestJson],
                  epochFlags: EpochParticipationFlags)
                 {.raises: [IOError, Defect].} =
-  for e in writer.stepwiseArrayCreation(epochFlags.asHashList):
+  for e in writer.stepwiseArrayCreation(epochFlags.asList):
     writer.writeValue $e
 
 proc readValue*(reader: var JsonReader[RestJson],
                 epochFlags: var EpochParticipationFlags)
                {.raises: [SerializationError, IOError, Defect].} =
-  # Please note that this function won't compute the cached hash tree roots
-  # immediately. They will be computed on the first HTR attempt.
-
   for e in reader.readArray(string):
     let parsed = try:
       parseBiggestUInt(e)
@@ -632,7 +629,7 @@ proc readValue*(reader: var JsonReader[RestJson],
       reader.raiseUnexpectedValue(
         "The usigned integer value should fit in 8 bits")
 
-    if not epochFlags.data.add(uint8(parsed)):
+    if not epochFlags.asList.add(uint8(parsed)):
       reader.raiseUnexpectedValue(
         "The participation flags list size exceeds limit")
 

beacon_chain/spec/ssz_codec.nim

@@ -17,7 +17,7 @@ import
   ./datatypes/base
 
 from ./datatypes/altair import
-  ParticipationFlags, EpochParticipationFlags, asHashList
+  ParticipationFlags, EpochParticipationFlags
 
 export codec, base, typetraits, EpochParticipationFlags
 
@@ -28,7 +28,7 @@ template toSszType*(v: BlsCurveType): auto = toRaw(v)
 template toSszType*(v: ForkDigest|GraffitiBytes): auto = distinctBase(v)
 template toSszType*(v: Version): auto = distinctBase(v)
 template toSszType*(v: JustificationBits): auto = distinctBase(v)
-template toSszType*(epochFlags: EpochParticipationFlags): auto = asHashList epochFlags
+template toSszType*(v: EpochParticipationFlags): auto = asList v
 
 func fromSszBytes*(T: type GraffitiBytes, data: openArray[byte]): T {.raisesssz.} =
   if data.len != sizeof(result):
@@ -60,4 +60,6 @@ func fromSszBytes*(T: type JustificationBits, bytes: openArray[byte]): T {.raise
   copyMem(result.addr, unsafeAddr bytes[0], sizeof(result))
 
 func fromSszBytes*(T: type EpochParticipationFlags, bytes: openArray[byte]): T {.raisesssz.} =
-  readSszValue(bytes, HashList[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT] result)
+  # TODO https://github.com/nim-lang/Nim/issues/21123
+  let tmp = cast[ptr List[ParticipationFlags, Limit VALIDATOR_REGISTRY_LIMIT]](addr result)
+  readSszValue(bytes, tmp[])

beacon_chain/spec/state_transition.nim

@@ -120,6 +120,7 @@ func process_slot*(
     hash_tree_root(state.latest_block_header)
 
 func clear_epoch_from_cache(cache: var StateCache, epoch: Epoch) =
+  cache.total_active_balance.del epoch
   cache.shuffled_active_validator_indices.del epoch
 
   for slot in epoch.slots():

beacon_chain/spec/state_transition_epoch.nim

@@ -1005,13 +1005,11 @@ func process_participation_flag_updates*(
 
   const zero = 0.ParticipationFlags
   for i in 0 ..< state.current_epoch_participation.len:
-    state.current_epoch_participation.data[i] = zero
+    asList(state.current_epoch_participation)[i] = zero
 
   # Shouldn't be wasted zeroing, because state.current_epoch_participation only
   # grows. New elements are automatically initialized to 0, as required.
-  doAssert state.current_epoch_participation.data.setLen(state.validators.len)
-
-  state.current_epoch_participation.asHashList.resetCache()
+  doAssert state.current_epoch_participation.asList.setLen(state.validators.len)
 
 # https://github.com/ethereum/consensus-specs/blob/v1.3.0-alpha.2/specs/altair/beacon-chain.md#sync-committee-updates
 func process_sync_committee_updates*(

beacon_chain/statediff.nim

@@ -143,8 +143,8 @@ func diffStates*(state0, state1: bellatrix.BeaconState): BeaconStateDiff =
     slashing: state1.slashings[state0.slot.epoch.uint64 mod
       EPOCHS_PER_HISTORICAL_VECTOR.uint64],
 
-    previous_epoch_participation: state1.previous_epoch_participation.data,
-    current_epoch_participation: state1.current_epoch_participation.data,
+    previous_epoch_participation: state1.previous_epoch_participation,
+    current_epoch_participation: state1.current_epoch_participation,
 
     justification_bits: state1.justification_bits,
     previous_justified_checkpoint: state1.previous_justified_checkpoint,
@@ -192,9 +192,9 @@ func applyDiff*(
   assign(state.slashings.mitem(epochIndex), stateDiff.slashing)
 
   assign(
-    state.previous_epoch_participation.data, stateDiff.previous_epoch_participation)
+    state.previous_epoch_participation, stateDiff.previous_epoch_participation)
   assign(
-    state.current_epoch_participation.data, stateDiff.current_epoch_participation)
+    state.current_epoch_participation, stateDiff.current_epoch_participation)
 
   state.justification_bits = stateDiff.justification_bits
   assign(