status-im
/
nimbus-eth2
mirror of https://github.com/status-im/nimbus-eth2.git
2
0
Fork 0
Code Issues Projects Releases Wiki Activity

fix action tracker validator decay (#3792) 

Validators were staying around 32x as long as they should - the effect
is minimal because it's rare validators disappear, but still..
This commit is contained in:
Jacek Sieka 2022-06-23 14:08:13 +02:00 committed by GitHub
parent beaf05b8d1
commit 369d696ee7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 29 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
beacon_chain/validators/action_tracker.nim
View File

@ -9,26 +9,24 @@ import
std/[sequtils, tables], std/[sequtils, tables],
stew/shims/[sets, hashes], chronicles, stew/shims/[sets, hashes], chronicles,
eth/p2p/discoveryv5/random2, eth/p2p/discoveryv5/random2,
../spec/datatypes/base, ../spec/forks,
../spec/[helpers, network], ../consensus_object_pools/spec_cache
../consensus_object_pools/[blockchain_dag, spec_cache]
export base, helpers, network, sets, tables export forks, tables, sets
{.push raises: [Defect].} {.push raises: [Defect].}
const const
SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS* = 4 ##\ SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS* = 4
## The number of slots before we're up for aggregation duty that we'll ## The number of slots before we're up for aggregation duty that we'll
## actually subscribe to the subnet we're aggregating for - this gives ## actually subscribe to the subnet we're aggregating for - this gives
## the node time to find a mesh etc - can likely be further trimmed ## the node time to find a mesh etc - can likely be further trimmed
KNOWN_VALIDATOR_DECAY = 3 * 32 * SLOTS_PER_EPOCH ##\ KNOWN_VALIDATOR_DECAY* = 3 * SLOTS_PER_EPOCH
## The number of slots before we "forget" about validators that have ## The number of slots before we "forget" about validators that have
## registered for duties - once we've forgotten about a validator, we'll ## registered for duties - once we've forgotten about a validator, we'll
## eventually decrease the number of stability subnets we're subscribed to - ## eventually decrease the number of stability subnets we're subscribed to.
## 3 epochs because we perform attestations once every epoch, +1 to deal ## Active validators are expected to register for duty every epoch - we use
## with rounding + 1 to deal with the network growing beyond 260k validators ## 3 epochs here to counter rounding errors and communication delays.
## and us not validating every epoch any more.
## When known validators decrease, we will keep the stability subnet around ## When known validators decrease, we will keep the stability subnet around
## until it "naturally" expires. ## until it "naturally" expires.
@ -42,13 +40,13 @@ type
subscribeAllAttnets: bool subscribeAllAttnets: bool
currentSlot: Slot ##\ currentSlot: Slot
## Duties that we accept are limited to a range around the current slot ## Duties that we accept are limited to a range around the current slot
subscribedSubnets*: AttnetBits ##\ subscribedSubnets*: AttnetBits
## All subnets we're currently subscribed to ## All subnets we're currently subscribed to
stabilitySubnets: seq[tuple[subnet_id: SubnetId, expiration: Epoch]] ##\ stabilitySubnets: seq[tuple[subnet_id: SubnetId, expiration: Epoch]]
## The subnets on which we listen and broadcast gossip traffic to maintain ## The subnets on which we listen and broadcast gossip traffic to maintain
## the health of the network - these are advertised in the ENR ## the health of the network - these are advertised in the ENR
nextCycleEpoch: Epoch nextCycleEpoch: Epoch
@ -63,12 +61,12 @@ type
## The latest dependent root we used to compute attestation duties ## The latest dependent root we used to compute attestation duties
## for internal validators ## for internal validators
knownValidators*: Table[ValidatorIndex, Slot] ##\ knownValidators*: Table[ValidatorIndex, Slot]
## Validators that we've recently seen - we'll subscribe to one stability ## Validators that we've recently seen - we'll subscribe to one stability
## subnet for each such validator - the slot is used to expire validators ## subnet for each such validator - the slot is used to expire validators
## that no longer are posting duties ## that no longer are posting duties
duties: HashSet[AggregatorDuty] ##\ duties: HashSet[AggregatorDuty]
## Known aggregation duties in the near future - before each such ## Known aggregation duties in the near future - before each such
## duty, we'll subscribe to the corresponding subnet to collect ## duty, we'll subscribe to the corresponding subnet to collect
## attestations for the aggregate ## attestations for the aggregate
@ -129,7 +127,7 @@ func stabilitySubnets*(tracker: ActionTracker, slot: Slot): AttnetBits =
res[v.subnet_id.int] = true res[v.subnet_id.int] = true
res res
func updateSlot*(tracker: var ActionTracker, wallSlot: Slot) = proc updateSlot*(tracker: var ActionTracker, wallSlot: Slot) =
# Prune duties from the past - this collection is kept small because there # Prune duties from the past - this collection is kept small because there
# are only so many slot/subnet combos - prune both internal and API-supplied # are only so many slot/subnet combos - prune both internal and API-supplied
# duties at the same time # duties at the same time
@ -139,7 +137,9 @@ func updateSlot*(tracker: var ActionTracker, wallSlot: Slot) =
var toPrune: seq[ValidatorIndex] var toPrune: seq[ValidatorIndex]
for k, v in tracker.knownValidators: for k, v in tracker.knownValidators:
if v + KNOWN_VALIDATOR_DECAY < wallSlot: toPrune.add k if v + KNOWN_VALIDATOR_DECAY < wallSlot: toPrune.add k
for k in toPrune: tracker.knownValidators.del k for k in toPrune:
debug "Validator no longer active", index = k
tracker.knownValidators.del k
# One stability subnet per known validator # One stability subnet per known validator
static: doAssert RANDOM_SUBNETS_PER_VALIDATOR == 1 static: doAssert RANDOM_SUBNETS_PER_VALIDATOR == 1

13
tests/test_action_tracker.nim
View File

@ -29,15 +29,26 @@ suite "subnet tracker":
check: check:
tracker.aggregateSubnets(Slot(0)).countOnes() == 2 tracker.aggregateSubnets(Slot(0)).countOnes() == 2
tracker.aggregateSubnets(Slot(1)).countOnes() == 1 tracker.aggregateSubnets(Slot(1)).countOnes() == 1
tracker.knownValidators.len() == 1
tracker.registerDuty(Slot(SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS), SubnetId(2), ValidatorIndex(0), true) tracker.registerDuty(Slot(SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS), SubnetId(2), ValidatorIndex(0), true)
check: check:
tracker.aggregateSubnets(Slot(0)).countOnes() == 2 tracker.aggregateSubnets(Slot(0)).countOnes() == 2
tracker.aggregateSubnets(Slot(1)).countOnes() == 2 tracker.aggregateSubnets(Slot(1)).countOnes() == 2
tracker.knownValidators.len() == 1
tracker.updateSlot(
Slot(SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS) + KNOWN_VALIDATOR_DECAY + 1)
check:
# Validator should be "forgotten" if they don't register for duty
tracker.knownValidators.len() == 0
# Guaranteed to expire # Guaranteed to expire
tracker.updateSlot( tracker.updateSlot(
Slot(EPOCHS_PER_RANDOM_SUBNET_SUBSCRIPTION * 2 * SLOTS_PER_EPOCH)) (Epoch(EPOCHS_PER_RANDOM_SUBNET_SUBSCRIPTION * 2) + 1).start_slot() +
SUBNET_SUBSCRIPTION_LEAD_TIME_SLOTS + KNOWN_VALIDATOR_DECAY + 1)
check: check:
tracker.stabilitySubnets(Slot(0)).countOnes() == 0 tracker.stabilitySubnets(Slot(0)).countOnes() == 0