feat: chainstore: sharded mutex for filling chain height index

This PR introduces as sharded mutex within the ChainIndex#GetTipsetByHeight.
It also replaces a go map with xsync.Map which doesn't require locking.

The lock is taken when it appears that ChainIndex filling work should be
started. After claiming the lock, the status of the cache is rechecked,
if the entry is still missing, the fillCache is started.

Thanks to @snissn and @arajasek for debugging and taking initial stabs at this.

Supersedes #10866 and 10885

Signed-off-by: Jakub Sztandera <kubuxu@protocol.ai>

chain/store/index.go

@@ -2,18 +2,21 @@ package store
 
 import (
 	"context"
+	"hash/maphash"
 	"os"
 	"strconv"
-	"sync"
 
+	"github.com/puzpuzpuz/xsync/v2"
 	"golang.org/x/xerrors"
 
 	"github.com/filecoin-project/go-state-types/abi"
 
 	"github.com/filecoin-project/lotus/chain/types"
+	"github.com/filecoin-project/lotus/lib/shardedmutex"
 )
 
-var DefaultChainIndexCacheSize = 32 << 15
+// DefaultChainIndexCacheSize no longer sets the maximum size, just the inital size of the map.
+var DefaultChainIndexCacheSize = 1 << 15
 
 func init() {
 	if s := os.Getenv("LOTUS_CHAIN_INDEX_CACHE"); s != "" {
@@ -27,8 +30,9 @@ func init() {
 }
 
 type ChainIndex struct {
-	indexCacheLk sync.Mutex
-	indexCache   map[types.TipSetKey]*lbEntry
+	indexCache *xsync.MapOf[types.TipSetKey, *lbEntry]
+
+	fillCacheLock shardedmutex.ShardedMutexFor[types.TipSetKey]
 
 	loadTipSet loadTipSetFunc
 
@@ -36,11 +40,16 @@ type ChainIndex struct {
 }
 type loadTipSetFunc func(context.Context, types.TipSetKey) (*types.TipSet, error)
 
+func maphashTSK(s maphash.Seed, tsk types.TipSetKey) uint64 {
+	return maphash.Bytes(s, tsk.Bytes())
+}
+
 func NewChainIndex(lts loadTipSetFunc) *ChainIndex {
 	return &ChainIndex{
-		indexCache: make(map[types.TipSetKey]*lbEntry, DefaultChainIndexCacheSize),
-		loadTipSet: lts,
-		skipLength: 20,
+		indexCache:    xsync.NewTypedMapOfPresized[types.TipSetKey, *lbEntry](maphashTSK, DefaultChainIndexCacheSize),
+		fillCacheLock: shardedmutex.NewFor(maphashTSK, 32),
+		loadTipSet:    lts,
+		skipLength:    20,
 	}
 }
 
@@ -59,17 +68,23 @@ func (ci *ChainIndex) GetTipsetByHeight(ctx context.Context, from *types.TipSet,
 		return nil, xerrors.Errorf("failed to round down: %w", err)
 	}
 
-	ci.indexCacheLk.Lock()
-	defer ci.indexCacheLk.Unlock()
 	cur := rounded.Key()
 	for {
-		lbe, ok := ci.indexCache[cur]
+		lbe, ok := ci.indexCache.Load(cur) // check the cache
 		if !ok {
-			fc, err := ci.fillCache(ctx, cur)
-			if err != nil {
-				return nil, xerrors.Errorf("failed to fill cache: %w", err)
+			lk := ci.fillCacheLock.GetLock(cur)
+			lk.Lock()                         // if entry is missing, take the lock
+			lbe, ok = ci.indexCache.Load(cur) // check if someone else added it while we waited for lock
+			if !ok {
+				fc, err := ci.fillCache(ctx, cur)
+				if err != nil {
+					lk.Unlock()
+					return nil, xerrors.Errorf("failed to fill cache: %w", err)
+				}
+				lbe = fc
+				ci.indexCache.Store(cur, lbe)
 			}
-			lbe = fc
+			lk.Unlock()
 		}
 
 		if to == lbe.targetHeight {
@@ -137,7 +152,6 @@ func (ci *ChainIndex) fillCache(ctx context.Context, tsk types.TipSetKey) (*lbEn
 		targetHeight: skipTarget.Height(),
 		target:       skipTarget.Key(),
 	}
-	ci.indexCache[tsk] = lbe
 
 	return lbe, nil
 }

go.mod

@@ -139,6 +139,7 @@ require (
 	github.com/open-rpc/meta-schema v0.0.0-20201029221707-1b72ef2ea333
 	github.com/polydawn/refmt v0.89.0
 	github.com/prometheus/client_golang v1.14.0
+	github.com/puzpuzpuz/xsync/v2 v2.4.0
 	github.com/raulk/clock v1.1.0
 	github.com/raulk/go-watchdog v1.3.0
 	github.com/stretchr/testify v1.8.2

go.sum

@@ -1484,6 +1484,8 @@ github.com/prometheus/procfs v0.8.0 h1:ODq8ZFEaYeCaZOJlZZdJA2AbQR98dSHSM1KW/You5
 github.com/prometheus/procfs v0.8.0/go.mod h1:z7EfXMXOkbkqb9IINtpCn86r/to3BnA0uaxHdg830/4=
 github.com/prometheus/statsd_exporter v0.21.0 h1:hA05Q5RFeIjgwKIYEdFd59xu5Wwaznf33yKI+pyX6T8=
 github.com/prometheus/statsd_exporter v0.21.0/go.mod h1:rbT83sZq2V+p73lHhPZfMc3MLCHmSHelCh9hSGYNLTQ=
+github.com/puzpuzpuz/xsync/v2 v2.4.0 h1:5sXAMHrtx1bg9nbRZTOn8T4MkWe5V+o8yKRH02Eznag=
+github.com/puzpuzpuz/xsync/v2 v2.4.0/go.mod h1:gD2H2krq/w52MfPLE+Uy64TzJDVY7lP2znR9qmR35kU=
 github.com/quic-go/qpack v0.4.0 h1:Cr9BXA1sQS2SmDUWjSofMPNKmvF6IiIfDRmgU0w1ZCo=
 github.com/quic-go/qpack v0.4.0/go.mod h1:UZVnYIfi5GRk+zI9UMaCPsmZ2xKJP7XBUvVyT1Knj9A=
 github.com/quic-go/qtls-go1-19 v0.2.1 h1:aJcKNMkH5ASEJB9FXNeZCyTEIHU1J7MmHyz1Q1TSG1A=

lib/shardedmutex/shardedmutex.go

@@ -0,0 +1,75 @@
+package shardedmutex
+
+import (
+	"hash/maphash"
+	"sync"
+)
+
+const cacheline = 64
+
+// padding a mutex to a cacheline improves performance as the cachelines are not contested
+// name     old time/op  new time/op  delta
+// Locks-8  74.6ns ± 7%  12.3ns ± 2%  -83.54%  (p=0.000 n=20+18)
+type paddedMutex struct {
+	mt  sync.Mutex
+	pad [cacheline - 8]uint8
+}
+
+type ShardedMutex struct {
+	shards []paddedMutex
+}
+
+// New creates a new ShardedMutex with N shards
+func New(n_shards int) ShardedMutex {
+	if n_shards < 1 {
+		panic("n_shards cannot be less than 1")
+	}
+	return ShardedMutex{
+		shards: make([]paddedMutex, n_shards),
+	}
+}
+
+func (sm ShardedMutex) Shards() int {
+	return len(sm.shards)
+}
+
+func (sm ShardedMutex) Lock(shard int) {
+	sm.shards[shard].mt.Lock()
+}
+
+func (sm ShardedMutex) Unlock(shard int) {
+	sm.shards[shard].mt.Unlock()
+}
+
+func (sm ShardedMutex) GetLock(shard int) sync.Locker {
+	return &sm.shards[shard].mt
+}
+
+type ShardedMutexFor[K any] struct {
+	inner ShardedMutex
+
+	hasher func(maphash.Seed, K) uint64
+	seed   maphash.Seed
+}
+
+func NewFor[K any](hasher func(maphash.Seed, K) uint64, n_shards int) ShardedMutexFor[K] {
+	return ShardedMutexFor[K]{
+		inner:  New(n_shards),
+		hasher: hasher,
+		seed:   maphash.MakeSeed(),
+	}
+}
+
+func (sm ShardedMutexFor[K]) shardFor(key K) int {
+	return int(sm.hasher(sm.seed, key) % uint64(len(sm.inner.shards)))
+}
+
+func (sm ShardedMutexFor[K]) Lock(key K) {
+	sm.inner.Lock(sm.shardFor(key))
+}
+func (sm ShardedMutexFor[K]) Unlock(key K) {
+	sm.inner.Unlock(sm.shardFor(key))
+}
+func (sm ShardedMutexFor[K]) GetLock(key K) sync.Locker {
+	return sm.inner.GetLock(sm.shardFor(key))
+}

lib/shardedmutex/shardedmutex_test.go

@@ -0,0 +1,159 @@
+package shardedmutex
+
+import (
+	"fmt"
+	"hash/maphash"
+	"runtime"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestLockingDifferentShardsDoesNotBlock(t *testing.T) {
+	shards := 16
+	sm := New(shards)
+	done := make(chan struct{})
+	go func() {
+		select {
+		case <-done:
+			return
+		case <-time.After(5 * time.Second):
+			panic("test locked up")
+		}
+	}()
+	for i := 0; i < shards; i++ {
+		sm.Lock(i)
+	}
+
+	close(done)
+}
+func TestLockingSameShardsBlocks(t *testing.T) {
+	shards := 16
+	sm := New(shards)
+	wg := sync.WaitGroup{}
+	wg.Add(shards)
+	ch := make(chan int, shards)
+
+	for i := 0; i < shards; i++ {
+		go func(i int) {
+			if i != 15 {
+				sm.Lock(i)
+			}
+			wg.Done()
+			wg.Wait()
+			sm.Lock((15 + i) % shards)
+			ch <- i
+			sm.Unlock(i)
+		}(i)
+	}
+
+	wg.Wait()
+	for i := 0; i < 2*shards; i++ {
+		runtime.Gosched()
+	}
+	for i := 0; i < shards; i++ {
+		if a := <-ch; a != i {
+			t.Errorf("got %d instead of %d", a, i)
+		}
+	}
+}
+
+func TestShardedByString(t *testing.T) {
+	shards := 16
+	sm := NewFor(maphash.String, shards)
+
+	wg1 := sync.WaitGroup{}
+	wg1.Add(shards * 20)
+	wg2 := sync.WaitGroup{}
+	wg2.Add(shards * 20)
+
+	active := atomic.Int32{}
+	max := atomic.Int32{}
+
+	for i := 0; i < shards*20; i++ {
+		go func(i int) {
+			wg1.Done()
+			wg1.Wait()
+			sm.Lock(fmt.Sprintf("goroutine %d", i))
+			activeNew := active.Add(1)
+			for {
+				curMax := max.Load()
+				if curMax >= activeNew {
+					break
+				}
+				if max.CompareAndSwap(curMax, activeNew) {
+					break
+				}
+			}
+			for j := 0; j < 100; j++ {
+				runtime.Gosched()
+			}
+			active.Add(-1)
+			sm.Unlock(fmt.Sprintf("goroutine %d", i))
+			wg2.Done()
+		}(i)
+	}
+
+	wg2.Wait()
+
+	if max.Load() != 16 {
+		t.Fatal("max load not achieved", max.Load())
+	}
+
+}
+
+func BenchmarkShardedMutex(b *testing.B) {
+	shards := 16
+	sm := New(shards)
+
+	done := atomic.Int32{}
+	go func() {
+		for {
+			sm.Lock(0)
+			sm.Unlock(0)
+			if done.Load() != 0 {
+				return
+			}
+		}
+	}()
+	for i := 0; i < 100; i++ {
+		runtime.Gosched()
+	}
+
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		sm.Lock(1)
+		sm.Unlock(1)
+	}
+	done.Add(1)
+}
+
+func BenchmarkShardedMutexOf(b *testing.B) {
+	shards := 16
+	sm := NewFor(maphash.String, shards)
+
+	str1 := "string1"
+	str2 := "string2"
+
+	done := atomic.Int32{}
+	go func() {
+		for {
+			sm.Lock(str1)
+			sm.Unlock(str1)
+			if done.Load() != 0 {
+				return
+			}
+		}
+	}()
+	for i := 0; i < 100; i++ {
+		runtime.Gosched()
+	}
+
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		sm.Lock(str2)
+		sm.Unlock(str2)
+	}
+	done.Add(1)
+}