neo-go/pkg/network/blockqueue.go
Roman Khimov 8bb1ecb45a network: remove priority queue from block queue
Use circular buffer which is a bit more appropriate. The problem is that
priority queue accepts and stores equal items which wastes memory even in
normal usage scenario, but it's especially dangerous if the node is stuck for
some reason. In this case it'll accept from peers and put into queue the same
blocks again and again leaking memory up to OOM condition.

Notice that queue length calculation might be wrong in case circular buffer
wraps, but it's not very likely to happen (usually blocks not coming from the
queue are added by consensus and it's not very fast in doing so).
2021-11-01 11:49:01 +03:00

134 lines
3.1 KiB
Go

package network
import (
"sync"
"github.com/nspcc-dev/neo-go/pkg/core/block"
"github.com/nspcc-dev/neo-go/pkg/core/blockchainer"
"go.uber.org/atomic"
"go.uber.org/zap"
)
type blockQueue struct {
log *zap.Logger
queueLock sync.Mutex
queue []*block.Block
checkBlocks chan struct{}
chain blockchainer.Blockqueuer
relayF func(*block.Block)
discarded *atomic.Bool
len int
}
const (
// blockCacheSize is the amount of blocks above current height
// which are stored in queue.
blockCacheSize = 2000
)
func newBlockQueue(capacity int, bc blockchainer.Blockqueuer, log *zap.Logger, relayer func(*block.Block)) *blockQueue {
if log == nil {
return nil
}
return &blockQueue{
log: log,
queue: make([]*block.Block, blockCacheSize),
checkBlocks: make(chan struct{}, 1),
chain: bc,
relayF: relayer,
discarded: atomic.NewBool(false),
}
}
func (bq *blockQueue) run() {
var lastHeight = bq.chain.BlockHeight()
for {
_, ok := <-bq.checkBlocks
if !ok {
break
}
for {
h := bq.chain.BlockHeight()
pos := int(h+1) % blockCacheSize
bq.queueLock.Lock()
b := bq.queue[pos]
// The chain moved forward using blocks from other sources (consensus).
for i := lastHeight; i < h; i++ {
old := int(i+1) % blockCacheSize
if bq.queue[old] != nil && bq.queue[old].Index == i {
bq.len--
bq.queue[old] = nil
}
}
bq.queueLock.Unlock()
lastHeight = h
if b == nil {
break
}
err := bq.chain.AddBlock(b)
if err != nil {
// The block might already be added by consensus.
if bq.chain.BlockHeight() < b.Index {
bq.log.Warn("blockQueue: failed adding block into the blockchain",
zap.String("error", err.Error()),
zap.Uint32("blockHeight", bq.chain.BlockHeight()),
zap.Uint32("nextIndex", b.Index))
}
} else if bq.relayF != nil {
bq.relayF(b)
}
bq.queueLock.Lock()
bq.len--
l := bq.len
if bq.queue[pos] == b {
bq.queue[pos] = nil
}
bq.queueLock.Unlock()
updateBlockQueueLenMetric(l)
}
}
}
func (bq *blockQueue) putBlock(block *block.Block) error {
h := bq.chain.BlockHeight()
bq.queueLock.Lock()
if block.Index <= h || h+blockCacheSize < block.Index {
// can easily happen when fetching the same blocks from
// different peers, thus not considered as error
bq.queueLock.Unlock()
return nil
}
pos := block.Index % blockCacheSize
// If we already have it, keep the old block, throw away new one.
if bq.queue[pos] == nil || bq.queue[pos].Index < block.Index {
bq.len++
bq.queue[pos] = block
}
l := bq.len
bq.queueLock.Unlock()
// update metrics
updateBlockQueueLenMetric(l)
select {
case bq.checkBlocks <- struct{}{}:
// ok, signalled to goroutine processing queue
default:
// it's already busy processing blocks
}
return nil
}
func (bq *blockQueue) discard() {
if bq.discarded.CAS(false, true) {
close(bq.checkBlocks)
bq.queueLock.Lock()
// Technically we could bq.queue = nil, but this would cost
// another if in run().
for i := 0; i < len(bq.queue); i++ {
bq.queue[i] = nil
}
bq.len = 0
bq.queueLock.Unlock()
}
}