Files
seaweedfs/weed/storage/needle_map_memory.go
T
b9ad62fc16 [Volume] Keep DAT and index state consistent after async batch Sync failure (#11425)
* fix 11400

* persist failed-recovery quarantine and harden rollback

- record the unavailable state in a .unavailable marker, fsync it, and
  re-arm it on load so a restart cannot serve an unverified pair
- quarantine the volume so heartbeats stop advertising it
- block MarkVolumeWritable while unavailable, rechecked under noWriteLock
- fail every request of a failed batch, not only the succeeded ones
- restore the needle map and truncate .dat on inline fsync rollback failure
- add truncateIndex for the sorted-file needle map
- mirror the fail-closed semantics in the Rust volume server

* volume: erase rolled-back mappings instead of leaving tombstones

A rolled-back batch or failed inline write used Delete() to undo a
needle that did not exist beforehand, leaving a tombstoned map entry
whose stale offset makes the next write to that needle fail reading a
header that no longer exists. Add removeMapping/restoreMapping to the
mappers so recovery erases entries that were absent before the batch
and reinstates the exact prior offset/size for ones that were,
including tombstones. The index row still goes through Delete so a
replay forgets the needle.

* volume: gate bulk readers on unavailable and fsync the marker's dir

- fsync_dir(&self.dir) synced the volume dir's parent, not the dir
  holding .unavailable; pass the marker path so the create survives
  a host crash
- export UnavailableError and check it in ReadAllNeedles,
  VolumeTailSender, VolumeIncrementalCopy, and IncrementalBackup so
  replica-sync paths cannot stream or append data from an unverified
  .dat/.idx pair; mirror on the Rust side via read_dat_slice,
  read_all_needles, dat_scan_plan, and the incremental-copy handler

* volume: drop issue references from comments near touched code

* volume: stop active scans when the volume becomes unavailable

The stream entry-point checks ran once per RPC, so a volume quarantined
by a failed recovery mid-scan kept serving data. Recheck availability
per needle/chunk on the detached read paths: tail scan and heartbeat,
read-all, incremental copy, incremental backup writes, and the Rust
StreamingBody chunk reads. Rust incremental copy also rejects a
quarantined volume before sync_to_disk touches the backend.

---------

Co-authored-by: Chris Lu <chris.lu@gmail.com>
2026-09-24 06:57:44 +08:00

147 lines
4.2 KiB
Go

package storage
import (
"os"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/storage/idx"
"github.com/seaweedfs/seaweedfs/weed/storage/needle"
"github.com/seaweedfs/seaweedfs/weed/storage/needle_map"
. "github.com/seaweedfs/seaweedfs/weed/storage/types"
"github.com/syndtr/goleveldb/leveldb/opt"
)
type NeedleMap struct {
baseNeedleMapper
m needle_map.NeedleValueMap
}
func NewCompactNeedleMap(file *os.File) *NeedleMap {
nm := &NeedleMap{
m: needle_map.NewCompactMap(),
}
nm.indexFile = file
stat, err := file.Stat()
if err != nil {
glog.Fatalf("stat file %s: %v", file.Name(), err)
}
nm.indexFileOffset = stat.Size()
return nm
}
func LoadCompactNeedleMap(file *os.File, version needle.Version) (*NeedleMap, error) {
nm := NewCompactNeedleMap(file)
return doLoading(file, nm, version)
}
func doLoading(file *os.File, nm *NeedleMap, version needle.Version) (*NeedleMap, error) {
e := idx.WalkIndexFile(file, 0, func(key NeedleId, offset Offset, size Size) error {
nm.MaybeSetMaxFileKey(key)
nm.MaybeSetMaxNeedleEnd(offset, size, version)
if !offset.IsZero() && !size.IsDeleted() {
nm.FileCounter++
nm.FileByteCounter = nm.FileByteCounter + uint64(size)
oldOffset, oldSize := nm.m.Set(NeedleId(key), offset, size)
if !oldOffset.IsZero() && !oldSize.IsDeleted() {
nm.DeletionCounter++
nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
}
} else {
oldSize := nm.m.Delete(NeedleId(key))
nm.DeletionCounter++
nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
}
return nil
})
glog.V(1).Infof("max file key: %v count: %d deleted: %d for file: %s", nm.MaxFileKey(), nm.FileCount(), nm.DeletedCount(), file.Name())
return nm, e
}
func (nm *NeedleMap) Put(key NeedleId, offset Offset, size Size) error {
_, oldSize := nm.m.Set(NeedleId(key), offset, size)
nm.logPut(key, oldSize, size)
return nm.appendToIndexFile(key, offset, size)
}
func (nm *NeedleMap) Get(key NeedleId) (element *needle_map.NeedleValue, ok bool) {
element, ok = nm.m.Get(NeedleId(key))
return
}
func (nm *NeedleMap) Delete(key NeedleId, offset Offset) error {
deletedBytes := nm.m.Delete(NeedleId(key))
nm.logDelete(deletedBytes)
return nm.appendToIndexFile(key, offset, TombstoneFileSize)
}
func (nm *NeedleMap) removeMapping(key NeedleId) error {
nm.m.Remove(NeedleId(key))
return nil
}
func (nm *NeedleMap) restoreMapping(key NeedleId, offset Offset, size Size) error {
nm.m.Set(NeedleId(key), offset, size)
return nil
}
func (nm *NeedleMap) Close() {
if nm.indexFile == nil {
return
}
indexFileName := nm.indexFile.Name()
if err := nm.indexFile.Sync(); err != nil {
glog.Warningf("sync file %s failed, %v", indexFileName, err)
}
_ = nm.indexFile.Close()
}
func (nm *NeedleMap) Destroy() error {
nm.Close()
return os.Remove(nm.indexFile.Name())
}
func (nm *NeedleMap) UpdateNeedleMap(v *Volume, indexFile *os.File, opts *opt.Options, ldbTimeout int64) error {
if v.nm != nil {
v.nm.Close()
v.nm = nil
}
defer func() {
if v.tmpNm != nil {
v.tmpNm.Close()
v.tmpNm = nil
}
}()
nm.indexFile = indexFile
stat, err := indexFile.Stat()
if err != nil {
glog.Fatalf("stat file %s: %v", indexFile.Name(), err)
return err
}
nm.indexFileOffset = stat.Size()
v.nm = nm
v.tmpNm = nil
return nil
}
func (nm *NeedleMap) DoOffsetLoading(v *Volume, indexFile *os.File, startFrom uint64) error {
glog.V(0).Infof("loading idx from offset %d for file: %s", startFrom, indexFile.Name())
version := needle.GetCurrentVersion()
if v != nil {
version = v.Version()
}
e := idx.WalkIndexFile(indexFile, startFrom, func(key NeedleId, offset Offset, size Size) error {
nm.MaybeSetMaxFileKey(key)
nm.MaybeSetMaxNeedleEnd(offset, size, version)
nm.FileCounter++
if !offset.IsZero() && !size.IsDeleted() {
nm.FileByteCounter = nm.FileByteCounter + uint64(size)
oldOffset, oldSize := nm.m.Set(NeedleId(key), offset, size)
if !oldOffset.IsZero() && !oldSize.IsDeleted() {
nm.DeletionCounter++
nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
}
} else {
oldSize := nm.m.Delete(NeedleId(key))
nm.DeletionCounter++
nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
}
return nil
})
return e
}