Merge remote-tracking branch 'origin' into command-processor

CHANGELOG.md

@@ -8,6 +8,7 @@ This release adds new control over Raft snapshotting, a key part of the Raft con
 ### Implementation changes and bug fixes
 - [PR #1528](https://github.com/rqlite/rqlite/pull/1528): Support setting trailing logs for user-requested snapshot.
 - [PR #1529](https://github.com/rqlite/rqlite/pull/1529): Remove obsolete code related to user-triggered snapshots.
+- [PR #1536](https://github.com/rqlite/rqlite/pull/1536): Store WAL path in store, to avoid races.
 
 ## 8.13.5 (December 26th 2023)
 ### Implementation changes and bug fixes

db/db.go

@@ -111,6 +111,11 @@ type PoolStats struct {
 	MaxLifetimeClosed  int64         `json:"max_lifetime_closed"`
 }
 
+// WALPath returns the path to the WAL file for the given database path.
+func WALPath(dbPath string) string {
+	return dbPath + "-wal"
+}
+
 // IsValidSQLiteFile checks that the supplied path looks like a SQLite file.
 // A non-existent file is considered invalid.
 func IsValidSQLiteFile(path string) bool {

db/db_test.go

@@ -86,6 +86,9 @@ func Test_DBPaths(t *testing.T) {
 	if exp, got := pathWAL+"-wal", dbWAL.WALPath(); exp != got {
 		t.Fatalf("expected WAL path %s, got %s", exp, got)
 	}
+	if p1, p2 := WALPath(pathWAL), dbWAL.WALPath(); p1 != p2 {
+		t.Fatalf("WAL paths are not equal (%s != %s)", p1, p2)
+	}
 
 	db, path := mustCreateOnDiskDatabase()
 	defer db.Close()

snapshot/store.go

@@ -176,7 +176,9 @@ func (s *Store) SetFullNeeded() error {
 	return f.Close()
 }
 
-// Stats returns stats about the Snapshot Store.
+// Stats returns stats about the Snapshot Store. This function may return
+// an error if the Store is in an inconsistent state. In that case the stats
+// returned may be incomplete or invalid.
 func (s *Store) Stats() (map[string]interface{}, error) {
 	snapshots, err := s.getSnapshots()
 	if err != nil {

store/store.go

@@ -215,14 +215,15 @@ type Store struct {
 	restorePath   string
 	restoreDoneCh chan struct{}
 
-	raft   *raft.Raft // The consensus mechanism.
-	ly     Layer
-	raftTn *NodeTransport
-	raftID string    // Node ID.
-	dbConf *DBConfig // SQLite database config.
-	dbPath string    // Path to underlying SQLite file.
-	dbDir  string    // Path to directory containing SQLite file.
-	db     *sql.DB   // The underlying SQLite store.
+	raft    *raft.Raft // The consensus mechanism.
+	ly      Layer
+	raftTn  *NodeTransport
+	raftID  string    // Node ID.
+	dbConf  *DBConfig // SQLite database config.
+	dbPath  string    // Path to underlying SQLite file.
+	walPath string    // Path to WAL file.
+	dbDir   string    // Path to directory containing SQLite file.
+	db      *sql.DB   // The underlying SQLite store.
 
 	queryTxMu sync.RWMutex
 
@@ -338,6 +339,7 @@ func New(ly Layer, c *Config) *Store {
 		raftID:          c.ID,
 		dbConf:          c.DBConf,
 		dbPath:          dbPath,
+		walPath:         sql.WALPath(dbPath),
 		dbDir:           filepath.Dir(dbPath),
 		leaderObservers: make([]chan<- struct{}, 0),
 		reqMarshaller:   command.NewRequestMarshaler(),
@@ -952,11 +954,6 @@ func (s *Store) Stats() (map[string]interface{}, error) {
 		return nil, err
 	}
 
-	snapsStats, err := s.snapshotStore.Stats()
-	if err != nil {
-		return nil, err
-	}
-
 	lAppliedIdx, err := s.boltStore.GetAppliedIndex()
 	if err != nil {
 		return nil, err
@@ -981,7 +978,6 @@ func (s *Store) Stats() (map[string]interface{}, error) {
 		"apply_timeout":          s.ApplyTimeout.String(),
 		"heartbeat_timeout":      s.HeartbeatTimeout.String(),
 		"election_timeout":       s.ElectionTimeout.String(),
-		"snapshot_store":         snapsStats,
 		"snapshot_threshold":     s.SnapshotThreshold,
 		"snapshot_interval":      s.SnapshotInterval.String(),
 		"reap_timeout":           s.ReapTimeout.String(),
@@ -995,6 +991,13 @@ func (s *Store) Stats() (map[string]interface{}, error) {
 		"sqlite3":                dbStatus,
 		"db_conf":                s.dbConf,
 	}
+
+	// Snapshot stats may be in flux if a snapshot is in progress. Only
+	// report them if they are available.
+	snapsStats, err := s.snapshotStore.Stats()
+	if err == nil {
+		status["snapshot_store"] = snapsStats
+	}
 	return status, nil
 }
 
@@ -1738,8 +1741,8 @@ func (s *Store) fsmSnapshot() (fSnap raft.FSMSnapshot, retErr error) {
 	} else {
 		compactedBuf := bytes.NewBuffer(nil)
 		var err error
-		if pathExistsWithData(s.db.WALPath()) {
-			walFD, err := os.Open(s.db.WALPath())
+		if pathExistsWithData(s.walPath) {
+			walFD, err := os.Open(s.walPath)
 			if err != nil {
 				return nil, err
 			}
@@ -1758,7 +1761,7 @@ func (s *Store) fsmSnapshot() (fSnap raft.FSMSnapshot, retErr error) {
 			}
 			walFD.Close() // We need it closed for the next step.
 
-			walSz, err := fileSize(s.db.WALPath())
+			walSz, err := fileSize(s.walPath)
 			if err != nil {
 				return nil, err
 			}
@@ -1975,7 +1978,7 @@ func (s *Store) runWALSnapshotting() (closeCh, doneCh chan struct{}) {
 		for {
 			select {
 			case <-ticker.C:
-				sz, err := s.db.WALSize()
+				sz, err := fileSize(s.walPath)
 				if err != nil {
 					s.logger.Printf("failed to get WAL size: %s", err.Error())
 					continue