mirror of
https://github.com/seaweedfs/seaweedfs.git
synced 2026-09-27 18:34:19 +00:00
* 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>
343 lines
10 KiB
Go
343 lines
10 KiB
Go
package storage
|
|
|
|
import (
|
|
"fmt"
|
|
"io"
|
|
"time"
|
|
|
|
"github.com/seaweedfs/seaweedfs/weed/util/mem"
|
|
|
|
"github.com/seaweedfs/seaweedfs/weed/glog"
|
|
"github.com/seaweedfs/seaweedfs/weed/stats"
|
|
"github.com/seaweedfs/seaweedfs/weed/storage/backend"
|
|
"github.com/seaweedfs/seaweedfs/weed/storage/needle"
|
|
"github.com/seaweedfs/seaweedfs/weed/storage/super_block"
|
|
. "github.com/seaweedfs/seaweedfs/weed/storage/types"
|
|
)
|
|
|
|
const PagedReadLimit = 1024 * 1024
|
|
|
|
// read fills in Needle content by looking up n.Id from NeedleMapper
|
|
func (v *Volume) readNeedle(n *needle.Needle, readOption *ReadOption, onReadSizeFn func(size Size)) (count int, err error) {
|
|
v.dataFileAccessLock.RLock()
|
|
defer v.dataFileAccessLock.RUnlock()
|
|
|
|
if err := v.UnavailableError(); err != nil {
|
|
return 0, err
|
|
}
|
|
|
|
if v.nm == nil {
|
|
glog.V(0).Infof("volume %d: needle map not loaded; read returns not-found", v.Id)
|
|
return -1, ErrorNotFound
|
|
}
|
|
|
|
nv, ok := v.nm.Get(n.Id)
|
|
if !ok || nv.Offset.IsZero() {
|
|
return -1, ErrorNotFound
|
|
}
|
|
readSize := nv.Size
|
|
if readSize.IsDeleted() {
|
|
if readOption != nil && readOption.ReadDeleted && readSize != TombstoneFileSize {
|
|
glog.V(3).Infof("reading deleted %s", n.String())
|
|
stats.VolumeServerHandlerCounter.WithLabelValues(stats.ReadDeletedNeedle).Inc()
|
|
readSize = -readSize
|
|
} else {
|
|
return -1, ErrorDeleted
|
|
}
|
|
}
|
|
if readSize == 0 {
|
|
return 0, nil
|
|
}
|
|
if onReadSizeFn != nil {
|
|
onReadSizeFn(readSize)
|
|
}
|
|
if readOption != nil && readOption.AttemptMetaOnly && readSize > PagedReadLimit {
|
|
readOption.VolumeRevision = v.SuperBlock.CompactionRevision
|
|
err = n.ReadNeedleMeta(v.DataBackend, nv.Offset.ToActualOffset(), readSize, v.Version())
|
|
if err == needle.ErrorSizeMismatch && OffsetSize == 4 {
|
|
readOption.IsOutOfRange = true
|
|
err = n.ReadNeedleMeta(v.DataBackend, nv.Offset.ToActualOffset()+int64(MaxPossibleVolumeSize), readSize, v.Version())
|
|
}
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
if !n.IsCompressed() && !n.IsChunkedManifest() {
|
|
readOption.IsMetaOnly = true
|
|
}
|
|
}
|
|
if readOption == nil || !readOption.IsMetaOnly {
|
|
err = n.ReadData(v.DataBackend, nv.Offset.ToActualOffset(), readSize, v.Version())
|
|
v.checkReadWriteError(err)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
}
|
|
count = int(n.DataSize)
|
|
if !n.HasTtl() {
|
|
return
|
|
}
|
|
ttlMinutes := n.Ttl.Minutes()
|
|
if ttlMinutes == 0 {
|
|
return
|
|
}
|
|
if !n.HasLastModifiedDate() {
|
|
return
|
|
}
|
|
if time.Now().Before(time.Unix(0, int64(n.AppendAtNs)).Add(time.Duration(ttlMinutes) * time.Minute)) {
|
|
return
|
|
}
|
|
return -1, ErrorNotFound
|
|
}
|
|
|
|
// read needle at a specific offset
|
|
func (v *Volume) readNeedleMetaAt(n *needle.Needle, offset int64, size int32) (err error) {
|
|
v.dataFileAccessLock.RLock()
|
|
defer v.dataFileAccessLock.RUnlock()
|
|
|
|
if err := v.UnavailableError(); err != nil {
|
|
return err
|
|
}
|
|
|
|
// read deleted needle meta data
|
|
if size < 0 {
|
|
size = 0
|
|
}
|
|
err = n.ReadNeedleMeta(v.DataBackend, offset, Size(size), v.Version())
|
|
if err == needle.ErrorSizeMismatch && OffsetSize == 4 {
|
|
err = n.ReadNeedleMeta(v.DataBackend, offset+int64(MaxPossibleVolumeSize), Size(size), v.Version())
|
|
}
|
|
if err != nil {
|
|
return err
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// read fills in Needle content by looking up n.Id from NeedleMapper
|
|
func (v *Volume) readNeedleDataInto(n *needle.Needle, readOption *ReadOption, writer io.Writer, offset int64, size int64) (err error) {
|
|
|
|
if !readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RLock()
|
|
defer v.dataFileAccessLock.RUnlock()
|
|
}
|
|
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RLock()
|
|
}
|
|
if err := v.UnavailableError(); err != nil {
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
return err
|
|
}
|
|
if v.nm == nil {
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
glog.V(0).Infof("volume %d: needle map not loaded; read returns not-found", v.Id)
|
|
return ErrorNotFound
|
|
}
|
|
nv, ok := v.nm.Get(n.Id)
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
|
|
if !ok || nv.Offset.IsZero() {
|
|
return ErrorNotFound
|
|
}
|
|
readSize := nv.Size
|
|
if readSize.IsDeleted() {
|
|
if readOption != nil && readOption.ReadDeleted && readSize != TombstoneFileSize {
|
|
glog.V(3).Infof("reading deleted %s", n.String())
|
|
readSize = -readSize
|
|
} else {
|
|
return ErrorDeleted
|
|
}
|
|
}
|
|
if readSize == 0 {
|
|
return nil
|
|
}
|
|
|
|
actualOffset := nv.Offset.ToActualOffset()
|
|
if readOption.IsOutOfRange {
|
|
actualOffset += int64(MaxPossibleVolumeSize)
|
|
}
|
|
|
|
buf := mem.Allocate(min(readOption.ReadBufferSize, int(size)))
|
|
defer mem.Free(buf)
|
|
|
|
// read needle data
|
|
crc := needle.CRC(0)
|
|
for x := offset; x < offset+size; x += int64(len(buf)) {
|
|
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RLock()
|
|
if err := v.UnavailableError(); err != nil {
|
|
v.dataFileAccessLock.RUnlock()
|
|
return err
|
|
}
|
|
}
|
|
// possibly re-read needle offset if volume is compacted
|
|
if readOption.VolumeRevision != v.SuperBlock.CompactionRevision {
|
|
if v.nm == nil {
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
glog.V(0).Infof("volume %d: needle map not loaded mid-read", v.Id)
|
|
return ErrorNotFound
|
|
}
|
|
// the volume is compacted
|
|
nv, ok = v.nm.Get(n.Id)
|
|
if !ok || nv.Offset.IsZero() {
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
return ErrorNotFound
|
|
}
|
|
actualOffset = nv.Offset.ToActualOffset()
|
|
readOption.VolumeRevision = v.SuperBlock.CompactionRevision
|
|
}
|
|
count, err := n.ReadNeedleData(v.DataBackend, actualOffset, buf, x)
|
|
if readOption.HasSlowRead {
|
|
v.dataFileAccessLock.RUnlock()
|
|
}
|
|
// Thread the underlying read error through the EIO tracker.
|
|
// Without this, large/range GETs through readNeedleDataInto
|
|
// would never trip IoErrorTolerance even on a failing disk.
|
|
// io.EOF is treated as a clean end-of-stream below, not an
|
|
// error.
|
|
if err != nil && err != io.EOF {
|
|
v.checkReadWriteError(err)
|
|
}
|
|
|
|
toWrite := min(count, int(offset+size-x))
|
|
if toWrite > 0 {
|
|
crc = crc.Update(buf[0:toWrite])
|
|
// Note: CRC validation happens after the loop completes (see below)
|
|
// to avoid performance overhead in the hot read path
|
|
if _, err = writer.Write(buf[0:toWrite]); err != nil {
|
|
return fmt.Errorf("ReadNeedleData write: %w", err)
|
|
}
|
|
}
|
|
if err != nil {
|
|
if err == io.EOF {
|
|
err = nil
|
|
break
|
|
}
|
|
return fmt.Errorf("ReadNeedleData: %w", err)
|
|
}
|
|
if count <= 0 {
|
|
break
|
|
}
|
|
}
|
|
// Whole-needle read completed without a backend error — clear any
|
|
// pending EIO streak. If a non-EIO failure happens later (CRC etc.)
|
|
// we still return that error to the caller, but the disk itself
|
|
// produced clean bytes.
|
|
v.checkReadWriteError(nil)
|
|
if offset == 0 && size == int64(n.DataSize) && (n.Checksum != crc && uint32(n.Checksum) != crc.Value()) {
|
|
// the crc.Value() function is to be deprecated. this double checking is for backward compatibility
|
|
// with seaweed version using crc.Value() instead of uint32(crc), which appears in commit 056c480eb
|
|
// and switch appeared in version 3.09.
|
|
stats.VolumeServerHandlerCounter.WithLabelValues(stats.ErrorCRC).Inc()
|
|
return fmt.Errorf("ReadNeedleData checksum %v expected %v for Needle: %v,%v", crc, n.Checksum, v.Id, n)
|
|
}
|
|
return nil
|
|
|
|
}
|
|
|
|
func min(x, y int) int {
|
|
if x < y {
|
|
return x
|
|
}
|
|
return y
|
|
}
|
|
|
|
// read fills in Needle content by looking up n.Id from NeedleMapper
|
|
func (v *Volume) ReadNeedleBlob(offset int64, size Size) ([]byte, error) {
|
|
// A deletion marker is not a record length; reject it before taking the lock.
|
|
if size.IsDeleted() {
|
|
return nil, fmt.Errorf("invalid needle size %d", size)
|
|
}
|
|
|
|
v.dataFileAccessLock.RLock()
|
|
defer v.dataFileAccessLock.RUnlock()
|
|
|
|
if err := v.UnavailableError(); err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
blob, err := needle.ReadNeedleBlob(v.DataBackend, offset, size, v.Version())
|
|
v.checkReadWriteError(err)
|
|
return blob, err
|
|
}
|
|
|
|
type VolumeFileScanner interface {
|
|
VisitSuperBlock(super_block.SuperBlock) error
|
|
ReadNeedleBody() bool
|
|
VisitNeedle(n *needle.Needle, offset int64, needleHeader, needleBody []byte) error
|
|
}
|
|
|
|
func ScanVolumeFile(dirname string, collection string, id needle.VolumeId,
|
|
needleMapKind NeedleMapKind,
|
|
volumeFileScanner VolumeFileScanner) (err error) {
|
|
var v *Volume
|
|
if v, err = loadVolumeWithoutIndex(dirname, collection, id, needleMapKind, needle.GetCurrentVersion()); err != nil {
|
|
return fmt.Errorf("failed to load volume %d: %w", id, err)
|
|
}
|
|
if err = volumeFileScanner.VisitSuperBlock(v.SuperBlock); err != nil {
|
|
return fmt.Errorf("failed to process volume %d super block: %w", id, err)
|
|
}
|
|
defer v.Close()
|
|
|
|
version := v.Version()
|
|
|
|
offset := int64(v.SuperBlock.BlockSize())
|
|
|
|
return ScanVolumeFileFrom(version, v.DataBackend, offset, volumeFileScanner)
|
|
}
|
|
|
|
func ScanVolumeFileFrom(version needle.Version, datBackend backend.BackendStorageFile, offset int64, volumeFileScanner VolumeFileScanner) (err error) {
|
|
n, nh, rest, e := needle.ReadNeedleHeader(datBackend, version, offset)
|
|
if e != nil {
|
|
if e == io.EOF {
|
|
return nil
|
|
}
|
|
return fmt.Errorf("cannot read %s at offset %d: %w", datBackend.Name(), offset, e)
|
|
}
|
|
for n != nil {
|
|
var needleBody []byte
|
|
if volumeFileScanner.ReadNeedleBody() {
|
|
// println("needle", n.Id.String(), "offset", offset, "size", n.Size, "rest", rest)
|
|
if needleBody, err = n.ReadNeedleBody(datBackend, version, offset+NeedleHeaderSize, rest); err != nil {
|
|
glog.V(0).Infof("cannot read needle head [%d, %d) body [%d, %d) body length %d: %v", offset, offset+NeedleHeaderSize, offset+NeedleHeaderSize, offset+NeedleHeaderSize+rest, rest, err)
|
|
// err = fmt.Errorf("cannot read needle body: %v", err)
|
|
// return
|
|
}
|
|
}
|
|
err := volumeFileScanner.VisitNeedle(n, offset, nh, needleBody)
|
|
if err == io.EOF {
|
|
return nil
|
|
}
|
|
if err != nil {
|
|
glog.V(0).Infof("visit needle error: %v", err)
|
|
return fmt.Errorf("visit needle error: %w", err)
|
|
}
|
|
// A corrupt header can carry a size so negative that the record length
|
|
// is zero or less; the scan cannot advance past it.
|
|
recordSize := NeedleHeaderSize + rest
|
|
if recordSize <= 0 {
|
|
return fmt.Errorf("%s: needle header at offset %d has size %d, record length %d: %w", datBackend.Name(), offset, n.Size, recordSize, needle.ErrorCorrupted)
|
|
}
|
|
offset += recordSize
|
|
glog.V(4).Infof("==> new entry offset %d", offset)
|
|
if n, nh, rest, err = needle.ReadNeedleHeader(datBackend, version, offset); err != nil {
|
|
if err == io.EOF {
|
|
return nil
|
|
}
|
|
return fmt.Errorf("cannot read needle header at offset %d: %w", offset, err)
|
|
}
|
|
glog.V(4).Infof("new entry needle size:%d rest:%d", n.Size, rest)
|
|
}
|
|
return nil
|
|
}
|