prysm-pulse/beacon-chain/cache/committee.go
kasey f962d13407
Fix committee test race (#12338)
* fix races in committee cache tests

* lint

* gratuitous defer ClearCache if ClearCache

* log warning to avoid failed block processing

* gaz

* add Clear to cache stubs

* fix Clear mistakes

* last fake cache fix

---------

Co-authored-by: Kasey Kirkham <kasey@users.noreply.github.com>
2023-04-27 00:50:04 +00:00

238 lines
6.9 KiB
Go

//go:build !fuzz
package cache
import (
"context"
"errors"
"math"
"sync"
"time"
lru "github.com/hashicorp/golang-lru"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
lruwrpr "github.com/prysmaticlabs/prysm/v4/cache/lru"
"github.com/prysmaticlabs/prysm/v4/config/params"
"github.com/prysmaticlabs/prysm/v4/consensus-types/primitives"
"github.com/prysmaticlabs/prysm/v4/container/slice"
mathutil "github.com/prysmaticlabs/prysm/v4/math"
)
const (
// maxCommitteesCacheSize defines the max number of shuffled committees on per randao basis can cache.
// Due to reorgs and long finality, it's good to keep the old cache around for quickly switch over.
maxCommitteesCacheSize = int(32)
)
var (
// CommitteeCacheMiss tracks the number of committee requests that aren't present in the cache.
CommitteeCacheMiss = promauto.NewCounter(prometheus.CounterOpts{
Name: "committee_cache_miss",
Help: "The number of committee requests that aren't present in the cache.",
})
// CommitteeCacheHit tracks the number of committee requests that are in the cache.
CommitteeCacheHit = promauto.NewCounter(prometheus.CounterOpts{
Name: "committee_cache_hit",
Help: "The number of committee requests that are present in the cache.",
})
)
// CommitteeCache is a struct with 1 queue for looking up shuffled indices list by seed.
type CommitteeCache struct {
CommitteeCache *lru.Cache
lock sync.RWMutex
inProgress map[string]bool
}
// committeeKeyFn takes the seed as the key to retrieve shuffled indices of a committee in a given epoch.
func committeeKeyFn(obj interface{}) (string, error) {
info, ok := obj.(*Committees)
if !ok {
return "", ErrNotCommittee
}
return key(info.Seed), nil
}
// NewCommitteesCache creates a new committee cache for storing/accessing shuffled indices of a committee.
func NewCommitteesCache() *CommitteeCache {
cc := &CommitteeCache{}
cc.Clear()
return cc
}
// Clear resets the CommitteeCache to its initial state
func (c *CommitteeCache) Clear() {
c.lock.Lock()
defer c.lock.Unlock()
c.CommitteeCache = lruwrpr.New(maxCommitteesCacheSize)
c.inProgress = make(map[string]bool)
}
// Committee fetches the shuffled indices by slot and committee index. Every list of indices
// represent one committee. Returns true if the list exists with slot and committee index. Otherwise returns false, nil.
func (c *CommitteeCache) Committee(ctx context.Context, slot primitives.Slot, seed [32]byte, index primitives.CommitteeIndex) ([]primitives.ValidatorIndex, error) {
if err := c.checkInProgress(ctx, seed); err != nil {
return nil, err
}
obj, exists := c.CommitteeCache.Get(key(seed))
if exists {
CommitteeCacheHit.Inc()
} else {
CommitteeCacheMiss.Inc()
return nil, nil
}
item, ok := obj.(*Committees)
if !ok {
return nil, ErrNotCommittee
}
committeeCountPerSlot := uint64(1)
if item.CommitteeCount/uint64(params.BeaconConfig().SlotsPerEpoch) > 1 {
committeeCountPerSlot = item.CommitteeCount / uint64(params.BeaconConfig().SlotsPerEpoch)
}
indexOffSet, err := mathutil.Add64(uint64(index), uint64(slot.ModSlot(params.BeaconConfig().SlotsPerEpoch).Mul(committeeCountPerSlot)))
if err != nil {
return nil, err
}
start, end := startEndIndices(item, indexOffSet)
if end > uint64(len(item.ShuffledIndices)) || end < start {
return nil, errors.New("requested index out of bound")
}
return item.ShuffledIndices[start:end], nil
}
// AddCommitteeShuffledList adds Committee shuffled list object to the cache. T
// his method also trims the least recently list if the cache size has ready the max cache size limit.
func (c *CommitteeCache) AddCommitteeShuffledList(ctx context.Context, committees *Committees) error {
c.lock.Lock()
defer c.lock.Unlock()
if err := ctx.Err(); err != nil {
return err
}
key, err := committeeKeyFn(committees)
if err != nil {
return err
}
_ = c.CommitteeCache.Add(key, committees)
return nil
}
// ActiveIndices returns the active indices of a given seed stored in cache.
func (c *CommitteeCache) ActiveIndices(ctx context.Context, seed [32]byte) ([]primitives.ValidatorIndex, error) {
if err := c.checkInProgress(ctx, seed); err != nil {
return nil, err
}
obj, exists := c.CommitteeCache.Get(key(seed))
if exists {
CommitteeCacheHit.Inc()
} else {
CommitteeCacheMiss.Inc()
return nil, nil
}
item, ok := obj.(*Committees)
if !ok {
return nil, ErrNotCommittee
}
return item.SortedIndices, nil
}
// ActiveIndicesCount returns the active indices count of a given seed stored in cache.
func (c *CommitteeCache) ActiveIndicesCount(ctx context.Context, seed [32]byte) (int, error) {
if err := c.checkInProgress(ctx, seed); err != nil {
return 0, err
}
obj, exists := c.CommitteeCache.Get(key(seed))
if exists {
CommitteeCacheHit.Inc()
} else {
CommitteeCacheMiss.Inc()
return 0, nil
}
item, ok := obj.(*Committees)
if !ok {
return 0, ErrNotCommittee
}
return len(item.SortedIndices), nil
}
// HasEntry returns true if the committee cache has a value.
func (c *CommitteeCache) HasEntry(seed string) bool {
_, ok := c.CommitteeCache.Get(seed)
return ok
}
// MarkInProgress a request so that any other similar requests will block on
// Get until MarkNotInProgress is called.
func (c *CommitteeCache) MarkInProgress(seed [32]byte) error {
c.lock.Lock()
defer c.lock.Unlock()
s := key(seed)
if c.inProgress[s] {
return ErrAlreadyInProgress
}
c.inProgress[s] = true
return nil
}
// MarkNotInProgress will release the lock on a given request. This should be
// called after put.
func (c *CommitteeCache) MarkNotInProgress(seed [32]byte) error {
c.lock.Lock()
defer c.lock.Unlock()
s := key(seed)
delete(c.inProgress, s)
return nil
}
func startEndIndices(c *Committees, index uint64) (uint64, uint64) {
validatorCount := uint64(len(c.ShuffledIndices))
start := slice.SplitOffset(validatorCount, c.CommitteeCount, index)
end := slice.SplitOffset(validatorCount, c.CommitteeCount, index+1)
return start, end
}
// Using seed as source for key to handle reorgs in the same epoch.
// The seed is derived from state's array of randao mixes and epoch value
// hashed together. This avoids collisions on different validator set. Spec definition:
// https://github.com/ethereum/consensus-specs/blob/v0.9.3/specs/core/0_beacon-chain.md#get_seed
func key(seed [32]byte) string {
return string(seed[:])
}
func (c *CommitteeCache) checkInProgress(ctx context.Context, seed [32]byte) error {
delay := minDelay
// Another identical request may be in progress already. Let's wait until
// any in progress request resolves or our timeout is exceeded.
for {
if ctx.Err() != nil {
return ctx.Err()
}
c.lock.RLock()
if !c.inProgress[key(seed)] {
c.lock.RUnlock()
break
}
c.lock.RUnlock()
// This increasing backoff is to decrease the CPU cycles while waiting
// for the in progress boolean to flip to false.
time.Sleep(time.Duration(delay) * time.Nanosecond)
delay *= delayFactor
delay = math.Min(delay, maxDelay)
}
return nil
}