gvsafronov/futriix
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

Update internal/storage/persistence.go

Browse Source
This commit is contained in:
Григорий Сафронов
2026-06-18 19:48:08 +00:00
parent 48fe3447ad
commit 2fd82f2939
1 changed files with 310 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

353
internal/storage/persistence.go
View File

@@ -16,6 +16,8 @@ package storage
import (
"bytes"
"compress/gzip"
"crypto/sha256"
"encoding/hex"
"encoding/json"
"fmt"
"os"
@@ -232,6 +234,7 @@ type PersistenceConfig struct {
SyncWrites bool `json:"sync_writes"`
WalPath string `json:"wal_path"`
UseWalForCheckpoint bool `json:"use_wal_for_checkpoint"`
AtomicWrites bool `json:"atomic_writes"`
}
// DefaultPersistenceConfig возвращает конфигурацию по умолчанию
@@ -244,6 +247,7 @@ func DefaultPersistenceConfig() *PersistenceConfig {
SyncWrites: true,
WalPath: "futriis.wal",
UseWalForCheckpoint: true,
AtomicWrites: true,
}
}
@@ -258,18 +262,33 @@ type DatabaseSnapshot struct {
WalSegment string `json:"wal_segment,omitempty"`
}
// CheckpointRecoveryInfo содержит информацию о восстановлении из чекпоинта
type CheckpointRecoveryInfo struct {
CheckpointTime int64 `json:"checkpoint_time"`
CheckpointVersion uint64 `json:"checkpoint_version"`
WalLSN uint64 `json:"wal_lsn"`
AppliedWalCount int `json:"applied_wal_count"`
RestoredDocs int64 `json:"restored_docs"`
RestoredColls int `json:"restored_collections"`
DurationMs int64 `json:"duration_ms"`
Success bool `json:"success"`
Error string `json:"error,omitempty"`
}
// PersistenceManager управляет персистентным хранением
type PersistenceManager struct {
config *PersistenceConfig
storage *Storage
logger LoggerInterface
mu sync.RWMutex
stopChan chan struct{}
wg sync.WaitGroup
lastCheckpoint int64
checkpointID atomic.Uint64
walManager interface{}
walReader WALReader
config *PersistenceConfig
storage *Storage
logger LoggerInterface
mu sync.RWMutex
stopChan chan struct{}
wg sync.WaitGroup
lastCheckpoint int64
checkpointID atomic.Uint64
walManager interface{}
walReader WALReader
recoveryInfo *CheckpointRecoveryInfo
isRestoring atomic.Bool
}
// NewPersistenceManager создаёт новый менеджер персистентности
@@ -292,7 +311,6 @@ func NewPersistenceManager(config *PersistenceConfig, storage *Storage, logger L
}
}
// Инициализируем WAL reader для восстановления
if config.UseWalForCheckpoint && config.WalPath != "" {
pm.walReader = NewWALReaderImpl(filepath.Dir(config.WalPath), logger)
}
@@ -345,6 +363,12 @@ func (pm *PersistenceManager) checkpointLoop() {
case <-pm.stopChan:
return
case <-ticker.C:
if pm.isRestoring.Load() {
if pm.logger != nil {
pm.logger.Info("Skipping checkpoint during restore")
}
continue
}
if err := pm.SaveAll(); err != nil {
if pm.logger != nil {
pm.logger.Error(fmt.Sprintf("Failed to save checkpoint: %v", err))
@@ -356,8 +380,12 @@ func (pm *PersistenceManager) checkpointLoop() {
}
}
// SaveDatabase сохраняет базу данных на диск с синхронизацией WAL
// SaveDatabase сохраняет базу данных на диск с атомарной записью
func (pm *PersistenceManager) SaveDatabase(dbName string) error {
if pm.isRestoring.Load() {
return fmt.Errorf("cannot save checkpoint during restore")
}
pm.mu.Lock()
defer pm.mu.Unlock()
@@ -403,6 +431,7 @@ func (pm *PersistenceManager) SaveDatabase(dbName string) error {
"fields": doc.GetFields(),
"created_at": doc.CreatedAt,
"updated_at": doc.UpdatedAt,
"deleted_at": doc.DeletedAt,
"version": doc.Version,
})
}
@@ -414,6 +443,16 @@ func (pm *PersistenceManager) SaveDatabase(dbName string) error {
return err
}
// Вычисляем контрольную сумму
hash := sha256.Sum256(data)
snapshot.Checksum = hex.EncodeToString(hash[:])
// Пере-сериализуем с контрольной суммой
data, err = json.Marshal(snapshot)
if err != nil {
return err
}
if pm.config.CompressEnabled {
data, err = pm.compress(data)
if err != nil {
@@ -421,29 +460,61 @@ func (pm *PersistenceManager) SaveDatabase(dbName string) error {
}
}
// Атомарная запись с использованием временного файла
filename := pm.getSnapshotFilename(dbName, snapshot.Version)
if err := os.WriteFile(filename, data, 0644); err != nil {
if err := pm.atomicWriteFile(filename, data); err != nil {
return err
}
if pm.config.SyncWrites {
if f, err := os.OpenFile(filename, os.O_RDWR, 0644); err == nil {
RealFsync(f)
f.Close()
}
FsyncDir(pm.config.DataDir)
}
pm.lastCheckpoint = snapshot.CreatedAt
if pm.logger != nil {
pm.logger.Info(fmt.Sprintf("Saved database %s checkpoint %d (%d bytes, WAL LSN: %d)",
dbName, snapshot.Version, len(data), walLSN))
pm.logger.Info(fmt.Sprintf("Saved database %s checkpoint %d (%d bytes, checksum: %s, WAL LSN: %d)",
dbName, snapshot.Version, len(data), snapshot.Checksum[:8], walLSN))
}
return nil
}
// atomicWriteFile выполняет атомарную запись файла
func (pm *PersistenceManager) atomicWriteFile(filename string, data []byte) error {
if pm.config.AtomicWrites {
// Создаём временный файл в той же директории
tempFile := filename + ".tmp"
if err := os.WriteFile(tempFile, data, 0644); err != nil {
return fmt.Errorf("failed to write temp file: %v", err)
}
// Синхронизируем временный файл
if pm.config.SyncWrites {
if f, err := os.OpenFile(tempFile, os.O_RDWR, 0644); err == nil {
RealFsync(f)
f.Close()
}
}
// Атомарное переименование
if err := os.Rename(tempFile, filename); err != nil {
os.Remove(tempFile)
return fmt.Errorf("failed to rename temp file: %v", err)
}
// Синхронизируем директорию
FsyncDir(pm.config.DataDir)
return nil
}
// Обычная запись (не атомарная)
return os.WriteFile(filename, data, 0644)
}
// SaveAll сохраняет все базы данных
func (pm *PersistenceManager) SaveAll() error {
if pm.isRestoring.Load() {
return fmt.Errorf("cannot save checkpoints during restore")
}
databases := pm.storage.ListDatabases()
var lastErr error
@@ -459,51 +530,80 @@ func (pm *PersistenceManager) SaveAll() error {
return lastErr
}
// LoadDatabase загружает базу данных с диска
// LoadDatabase загружает базу данных с диска с применением WAL
func (pm *PersistenceManager) LoadDatabase(dbName string) error {
pm.isRestoring.Store(true)
defer pm.isRestoring.Store(false)
pm.mu.Lock()
defer pm.mu.Unlock()
startTime := time.Now()
recoveryInfo := &CheckpointRecoveryInfo{
CheckpointTime: time.Now().UnixMilli(),
Success: false,
}
snapshot, err := pm.findLatestSnapshot(dbName)
if err != nil {
recoveryInfo.Error = err.Error()
pm.recoveryInfo = recoveryInfo
return err
}
if snapshot == nil {
recoveryInfo.Error = "no snapshot found"
pm.recoveryInfo = recoveryInfo
return fmt.Errorf("no snapshot found for database: %s", dbName)
}
if pm.config.UseWalForCheckpoint && snapshot.WalLSN > 0 && pm.walReader != nil {
currentLSN, err := pm.walReader.GetCurrentIndex()
if err == nil && currentLSN > snapshot.WalLSN {
if pm.logger != nil {
pm.logger.Info(fmt.Sprintf("Database %s: WAL has %d entries after checkpoint, applying...",
dbName, currentLSN-snapshot.WalLSN))
}
}
recoveryInfo.CheckpointVersion = snapshot.Version
recoveryInfo.WalLSN = snapshot.WalLSN
// Проверяем целостность чекпоинта
if err := pm.verifySnapshot(snapshot); err != nil {
recoveryInfo.Error = err.Error()
pm.recoveryInfo = recoveryInfo
return fmt.Errorf("snapshot integrity check failed: %v", err)
}
// Создаём базу данных если не существует
if !pm.storage.ExistsDatabase(dbName) {
if err := pm.storage.CreateDatabase(dbName); err != nil {
recoveryInfo.Error = err.Error()
pm.recoveryInfo = recoveryInfo
return err
}
}
db, err := pm.storage.GetDatabase(dbName)
if err != nil {
recoveryInfo.Error = err.Error()
pm.recoveryInfo = recoveryInfo
return err
}
// Восстанавливаем коллекции
restoredColls := 0
restoredDocs := int64(0)
for collName, collDataRaw := range snapshot.Collections {
collData, ok := collDataRaw.([]interface{})
if !ok {
continue
}
if _, err := db.GetCollection(collName); err != nil {
if err := db.CreateCollection(collName); err != nil {
continue
// Удаляем существующую коллекцию если есть
if _, err := db.GetCollection(collName); err == nil {
db.DropCollection(collName)
}
if err := db.CreateCollection(collName); err != nil {
if pm.logger != nil {
pm.logger.Error(fmt.Sprintf("Failed to create collection %s: %v", collName, err))
}
continue
}
coll, err := db.GetCollection(collName)
@@ -511,6 +611,8 @@ func (pm *PersistenceManager) LoadDatabase(dbName string) error {
continue
}
restoredColls++
for _, docRaw := range collData {
docMap, ok := docRaw.(map[string]interface{})
if !ok {
@@ -532,27 +634,172 @@ func (pm *PersistenceManager) LoadDatabase(dbName string) error {
if createdAt, ok := docMap["created_at"].(float64); ok {
doc.CreatedAt = int64(createdAt)
} else if createdAt, ok := docMap["created_at"].(int64); ok {
doc.CreatedAt = createdAt
}
if updatedAt, ok := docMap["updated_at"].(float64); ok {
doc.UpdatedAt = int64(updatedAt)
} else if updatedAt, ok := docMap["updated_at"].(int64); ok {
doc.UpdatedAt = updatedAt
}
if deletedAt, ok := docMap["deleted_at"].(float64); ok {
doc.DeletedAt = int64(deletedAt)
} else if deletedAt, ok := docMap["deleted_at"].(int64); ok {
doc.DeletedAt = deletedAt
}
if version, ok := docMap["version"].(float64); ok {
doc.Version = uint64(version)
} else if version, ok := docMap["version"].(uint64); ok {
doc.Version = version
}
if err := coll.Insert(doc); err != nil {
if pm.logger != nil {
pm.logger.Error(fmt.Sprintf("Failed to restore document %s: %v", docID, err))
pm.logger.Warn(fmt.Sprintf("Failed to restore document %s: %v", docID, err))
}
continue
}
restoredDocs++
}
}
recoveryInfo.RestoredDocs = restoredDocs
recoveryInfo.RestoredColls = restoredColls
// Применяем WAL записи после чекпоинта
appliedWalCount := 0
if pm.config.UseWalForCheckpoint && snapshot.WalLSN > 0 && pm.walReader != nil {
currentLSN, err := pm.walReader.GetCurrentIndex()
if err == nil && currentLSN > snapshot.WalLSN {
if pm.logger != nil {
pm.logger.Info(fmt.Sprintf("Database %s: Applying %d WAL entries after checkpoint (LSN %d -> %d)",
dbName, currentLSN-snapshot.WalLSN, snapshot.WalLSN, currentLSN))
}
entries, err := pm.walReader.ReadSince(snapshot.WalLSN)
if err != nil {
if pm.logger != nil {
pm.logger.Error(fmt.Sprintf("Failed to read WAL entries: %v", err))
}
} else {
for _, entry := range entries {
if err := pm.applyWALEntry(db, entry); err != nil {
if pm.logger != nil {
pm.logger.Error(fmt.Sprintf("Failed to apply WAL entry %d: %v", entry.Index, err))
}
continue
}
appliedWalCount++
}
}
}
}
recoveryInfo.AppliedWalCount = appliedWalCount
recoveryInfo.DurationMs = time.Since(startTime).Milliseconds()
recoveryInfo.Success = true
pm.recoveryInfo = recoveryInfo
if pm.logger != nil {
pm.logger.Info(fmt.Sprintf("Loaded database %s from snapshot (version %d, WAL LSN: %d)",
dbName, snapshot.Version, snapshot.WalLSN))
pm.logger.Info(fmt.Sprintf("Loaded database %s from snapshot (version %d, WAL LSN: %d, applied %d WAL entries, %d docs, %d colls, duration: %dms)",
dbName, snapshot.Version, snapshot.WalLSN, appliedWalCount, restoredDocs, restoredColls, recoveryInfo.DurationMs))
}
return nil
}
// applyWALEntry применяет одну запись WAL к базе данных
func (pm *PersistenceManager) applyWALEntry(db *Database, entry WALEntry) error {
var txRecord struct {
ID uint64 `json:"id"`
State int32 `json:"state"`
Timestamp int64 `json:"timestamp"`
Operations []struct {
Type string `json:"type"`
Database string `json:"database"`
Collection string `json:"collection"`
DocumentID string `json:"document_id"`
Data map[string]interface{} `json:"data"`
Version uint64 `json:"version"`
} `json:"operations"`
}
if err := json.Unmarshal(entry.Data, &txRecord); err != nil {
return err
}
// Применяем только закоммиченные транзакции
if txRecord.State != 1 { // TransactionCommitted
return nil
}
for _, op := range txRecord.Operations {
// Пропускаем операции для других баз данных
if op.Database != db.name {
continue
}
coll, err := db.GetCollection(op.Collection)
if err != nil {
continue
}
switch op.Type {
case "insert":
doc := NewDocumentWithID(op.DocumentID)
for k, v := range op.Data {
doc.SetField(k, v)
}
doc.Version = op.Version
if err := coll.Insert(doc); err != nil {
return err
}
case "update":
if err := coll.Update(op.DocumentID, op.Data); err != nil {
return err
}
case "delete":
if err := coll.Delete(op.DocumentID); err != nil {
return err
}
}
}
return nil
}
// verifySnapshot проверяет целостность снапшота
func (pm *PersistenceManager) verifySnapshot(snapshot *DatabaseSnapshot) error {
if snapshot.Checksum == "" {
return fmt.Errorf("snapshot has no checksum")
}
// Пересоздаём данные без контрольной суммы для проверки
tempSnapshot := &DatabaseSnapshot{
Name: snapshot.Name,
Collections: snapshot.Collections,
CreatedAt: snapshot.CreatedAt,
Version: snapshot.Version,
WalLSN: snapshot.WalLSN,
WalSegment: snapshot.WalSegment,
}
data, err := json.Marshal(tempSnapshot)
if err != nil {
return fmt.Errorf("failed to marshal for checksum: %v", err)
}
hash := sha256.Sum256(data)
checksum := hex.EncodeToString(hash[:])
if checksum != snapshot.Checksum {
return fmt.Errorf("checksum mismatch: expected %s, got %s", snapshot.Checksum[:8], checksum[:8])
}
return nil
@@ -703,11 +950,31 @@ func (pm *PersistenceManager) GetLastCheckpointInfo() map[string]interface{} {
pm.mu.RLock()
defer pm.mu.RUnlock()
return map[string]interface{}{
"last_checkpoint_time": pm.lastCheckpoint,
"last_checkpoint_time_str": time.UnixMilli(pm.lastCheckpoint).Format("2006-01-02 15:04:05.000"),
"checkpoint_id": pm.checkpointID.Load(),
"wal_path": pm.config.WalPath,
"use_wal": pm.config.UseWalForCheckpoint,
info := map[string]interface{}{
"last_checkpoint_time": pm.lastCheckpoint,
"last_checkpoint_time_str": time.UnixMilli(pm.lastCheckpoint).Format("2006-01-02 15:04:05.000"),
"checkpoint_id": pm.checkpointID.Load(),
"wal_path": pm.config.WalPath,
"use_wal": pm.config.UseWalForCheckpoint,
"atomic_writes": pm.config.AtomicWrites,
"is_restoring": pm.isRestoring.Load(),
}
if pm.recoveryInfo != nil {
info["last_recovery"] = pm.recoveryInfo
}
return info
}
// GetRecoveryInfo возвращает информацию о последнем восстановлении
func (pm *PersistenceManager) GetRecoveryInfo() *CheckpointRecoveryInfo {
pm.mu.RLock()
defer pm.mu.RUnlock()
return pm.recoveryInfo
}
// IsRestoring возвращает статус восстановления
func (pm *PersistenceManager) IsRestoring() bool {
return pm.isRestoring.Load()
}