fix(tsdb): survive low disk space

This commit is contained in:
naiba committed 2026-09-13 01:15:18 +00:00
1 parent 1155c17573
commit 06d387779d
6 files changed
+341 -14

No files matched your search

+123 -5
View File
@@ -1,15 +1,26 @@
package tsdb
import (
"errors"
"fmt"
"log"
"path/filepath"
"strings"
"sync"
"sync/atomic"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
)
// ErrDiskFull is returned when VictoriaMetrics cannot initialize because the
// filesystem containing the TSDB has run out of space.
var ErrDiskFull = errors.New("TSDB disk is full")
type readOnlyChecker interface {
IsReadOnly() bool
}
// TSDB 封装 VictoriaMetrics 存储
type TSDB struct {
storage *storage.Storage
@@ -18,6 +29,13 @@ type TSDB struct {
closed bool
writer *bufferedWriter
readOnly readOnlyChecker
addRowsFn func([]storage.MetricRow, uint8)
debugFlushFn func()
readOnlyObserved atomic.Bool
diskFullPaused atomic.Bool
}
// InitGlobalSettings 初始化 VictoriaMetrics 包级别的全局设置。
@@ -35,7 +53,18 @@ func InitGlobalSettings(config *Config) {
}
// Open 打开或创建 TSDB 存储
func Open(config *Config) (*TSDB, error) {
func Open(config *Config) (db *TSDB, err error) {
defer func() {
if recovered := recover(); recovered != nil {
if diskErr := diskFullPanicError(recovered); diskErr != nil {
db = nil
err = diskErr
return
}
panic(recovered)
}
}()
if config == nil {
config = DefaultConfig()
}
@@ -59,9 +88,12 @@ func Open(config *Config) (*TSDB, error) {
stor := storage.MustOpenStorage(dataPath, opts)
db := &TSDB{
storage: stor,
config: config,
db = &TSDB{
storage: stor,
config: config,
readOnly: stor,
addRowsFn: stor.AddRows,
debugFlushFn: stor.DebugFlush,
}
db.writer = newBufferedWriter(db, config.WriteBufferSize, config.WriteBufferFlushInterval)
@@ -72,6 +104,86 @@ func Open(config *Config) (*TSDB, error) {
return db, nil
}
func diskFullPanicError(recovered any) error {
message := strings.ToLower(fmt.Sprint(recovered))
for _, marker := range []string{
"no space left on device",
"disk quota exceeded",
"not enough space on the disk",
} {
if strings.Contains(message, marker) {
return fmt.Errorf("%w: %v", ErrDiskFull, recovered)
}
}
return nil
}
// WritesPaused reports whether new samples are currently being discarded.
// Queries remain available while VictoriaMetrics is in read-only mode.
func (db *TSDB) WritesPaused() bool {
if db.diskFullPaused.Load() {
return true
}
return db.readOnly != nil && db.readOnly.IsReadOnly()
}
func (db *TSDB) acceptsWrites() bool {
if db.diskFullPaused.Load() {
return false
}
readOnly := db.readOnly != nil && db.readOnly.IsReadOnly()
if readOnly {
if db.readOnlyObserved.CompareAndSwap(false, true) {
log.Println("NEZHA>> TSDB writes paused because free disk space is below the configured limit; dashboard, queries, and alerts remain available")
}
return false
}
if db.readOnlyObserved.CompareAndSwap(true, false) {
log.Println("NEZHA>> TSDB writes resumed after free disk space recovered")
}
return true
}
func (db *TSDB) pauseWritesAfterDiskFull(recovered any) {
if db.diskFullPaused.CompareAndSwap(false, true) {
log.Printf("NEZHA>> TSDB disk is full; disabling TSDB writes until dashboard restart while keeping dashboard, queries, and alerts running: %v", recovered)
}
}
func (db *TSDB) addRowsSafely(rows []storage.MetricRow) {
if len(rows) == 0 || !db.acceptsWrites() {
return
}
defer func() {
if recovered := recover(); recovered != nil {
if diskFullPanicError(recovered) == nil {
panic(recovered)
}
db.pauseWritesAfterDiskFull(recovered)
}
}()
db.addRowsFn(rows, 64)
}
func (db *TSDB) debugFlushSafely() {
if !db.acceptsWrites() {
return
}
defer func() {
if recovered := recover(); recovered != nil {
if diskFullPanicError(recovered) == nil {
panic(recovered)
}
db.pauseWritesAfterDiskFull(recovered)
}
}()
db.debugFlushFn()
}
// Close 关闭 TSDB 存储
func (db *TSDB) Close() error {
db.mu.Lock()
@@ -110,8 +222,14 @@ func (db *TSDB) IsClosed() bool {
// Flush 强制刷盘(主要用于测试)
func (db *TSDB) Flush() {
db.mu.RLock()
defer db.mu.RUnlock()
if db.closed {
return
}
if db.writer != nil {
db.writer.flush()
}
db.storage.DebugFlush()
db.debugFlushSafely()
}