neo-go/pkg/core/headerhashes.go
Roman Khimov 1c38b45074 core: don't always store all hashes in memory
We're paging these hashes, so we need a previous full page and a current one
plus some cache for various requests. Storing 1M of hashes is 32M of memory
and it grows quickly. It also seriously affects node startup time, most of
what it's doing is reading these hashes, the longer the chain the more time it
needs to do that.

Notice that this doesn't change the underlying DB scheme in any way.
2022-11-25 14:30:51 +03:00

211 lines
5.5 KiB
Go

package core
import (
"fmt"
"sync"
lru "github.com/hashicorp/golang-lru"
"github.com/nspcc-dev/neo-go/pkg/core/block"
"github.com/nspcc-dev/neo-go/pkg/core/dao"
"github.com/nspcc-dev/neo-go/pkg/util"
)
const (
headerBatchCount = 2000
pagesCache = 8
)
// HeaderHashes is a header hash manager part of the Blockchain. It can't be used
// without Blockchain.
type HeaderHashes struct {
// Backing storage.
dao *dao.Simple
// Lock for all internal state fields.
lock sync.RWMutex
// The latest header hashes (storedHeaderCount+).
latest []util.Uint256
// Previously completed page of header hashes (pre-storedHeaderCount).
previous []util.Uint256
// Number of headers stored in the chain file.
storedHeaderCount uint32
// Cache for accessed pages of header hashes.
cache *lru.Cache
}
func (h *HeaderHashes) initGenesis(dao *dao.Simple, hash util.Uint256) {
h.dao = dao
h.cache, _ = lru.New(pagesCache) // Never errors for positive size.
h.previous = make([]util.Uint256, headerBatchCount)
h.latest = make([]util.Uint256, 0, headerBatchCount)
h.latest = append(h.latest, hash)
dao.PutCurrentHeader(hash, 0)
}
func (h *HeaderHashes) init(dao *dao.Simple) error {
h.dao = dao
h.cache, _ = lru.New(pagesCache) // Never errors for positive size.
currHeaderHeight, currHeaderHash, err := h.dao.GetCurrentHeaderHeight()
if err != nil {
return fmt.Errorf("failed to retrieve current header info: %w", err)
}
h.storedHeaderCount = ((currHeaderHeight + 1) / headerBatchCount) * headerBatchCount
if h.storedHeaderCount >= headerBatchCount {
h.previous, err = h.dao.GetHeaderHashes(h.storedHeaderCount - headerBatchCount)
if err != nil {
return fmt.Errorf("failed to retrieve header hash page %d: %w", h.storedHeaderCount-headerBatchCount, err)
}
} else {
h.previous = make([]util.Uint256, headerBatchCount)
}
h.latest = make([]util.Uint256, 0, headerBatchCount)
// There is a high chance that the Node is stopped before the next
// batch of 2000 headers was stored. Via the currentHeaders stored we can sync
// that with stored blocks.
if currHeaderHeight >= h.storedHeaderCount {
hash := currHeaderHash
var targetHash util.Uint256
if h.storedHeaderCount >= headerBatchCount {
targetHash = h.previous[len(h.previous)-1]
}
headers := make([]util.Uint256, 0, headerBatchCount)
for hash != targetHash {
blk, err := h.dao.GetBlock(hash)
if err != nil {
return fmt.Errorf("could not get header %s: %w", hash, err)
}
headers = append(headers, blk.Hash())
hash = blk.PrevHash
}
hashSliceReverse(headers)
h.latest = append(h.latest, headers...)
}
return nil
}
func (h *HeaderHashes) lastHeaderIndex() uint32 {
return h.storedHeaderCount + uint32(len(h.latest)) - 1
}
// HeaderHeight returns the index/height of the highest header.
func (h *HeaderHashes) HeaderHeight() uint32 {
h.lock.RLock()
n := h.lastHeaderIndex()
h.lock.RUnlock()
return n
}
func (h *HeaderHashes) addHeaders(headers ...*block.Header) error {
var (
batch = h.dao.GetPrivate()
lastHeader *block.Header
err error
)
h.lock.Lock()
defer h.lock.Unlock()
for _, head := range headers {
if head.Index != h.lastHeaderIndex()+1 {
continue
}
err = batch.StoreHeader(head)
if err != nil {
return err
}
lastHeader = head
h.latest = append(h.latest, head.Hash())
if len(h.latest) == headerBatchCount {
err = batch.StoreHeaderHashes(h.latest, h.storedHeaderCount)
if err != nil {
return err
}
copy(h.previous, h.latest)
h.latest = h.latest[:0]
h.storedHeaderCount += headerBatchCount
}
}
if lastHeader != nil {
batch.PutCurrentHeader(lastHeader.Hash(), lastHeader.Index)
updateHeaderHeightMetric(lastHeader.Index)
if _, err = batch.Persist(); err != nil {
return err
}
}
return nil
}
// CurrentHeaderHash returns the hash of the latest known header.
func (h *HeaderHashes) CurrentHeaderHash() util.Uint256 {
var hash util.Uint256
h.lock.RLock()
if len(h.latest) > 0 {
hash = h.latest[len(h.latest)-1]
} else {
hash = h.previous[len(h.previous)-1]
}
h.lock.RUnlock()
return hash
}
// GetHeaderHash returns hash of the header/block with specified index, if
// HeaderHashes doesn't have a hash for this height, zero Uint256 value is returned.
func (h *HeaderHashes) GetHeaderHash(i uint32) util.Uint256 {
h.lock.RLock()
res, ok := h.getLocalHeaderHash(i)
h.lock.RUnlock()
if ok {
return res
}
// If it's not in the latest/previous, then it's in the cache or DB, those
// need no additional locks.
page := (i / headerBatchCount) * headerBatchCount
cache, ok := h.cache.Get(page)
if ok {
hashes := cache.([]util.Uint256)
return hashes[i-page]
}
hashes, err := h.dao.GetHeaderHashes(page)
if err != nil {
return util.Uint256{}
}
_ = h.cache.Add(page, hashes)
return hashes[i-page]
}
// getLocalHeaderHash looks for the index in the latest and previous caches.
// Locking is left to the user.
func (h *HeaderHashes) getLocalHeaderHash(i uint32) (util.Uint256, bool) {
if i > h.lastHeaderIndex() {
return util.Uint256{}, false
}
if i >= h.storedHeaderCount {
return h.latest[i-h.storedHeaderCount], true
}
previousStored := h.storedHeaderCount - headerBatchCount
if i >= previousStored {
return h.previous[i-previousStored], true
}
return util.Uint256{}, false
}
func (h *HeaderHashes) haveRecentHash(hash util.Uint256, i uint32) bool {
h.lock.RLock()
defer h.lock.RUnlock()
for ; i > 0; i-- {
lh, ok := h.getLocalHeaderHash(i)
if ok && hash.Equals(lh) {
return true
}
}
return false
}