Files
seaweedfs/weed/topology/store_replicate.go
T
Chris LuandClaude Opus 5.5 9cd183b04f volume: fsync a needle delete that asks for it (#12736)
* volume: fsync a needle delete that asks for it

A delete now takes fsync like a write: the tombstone is synced before the
needle map drops the needle, and the .idx row before the delete is
acknowledged. A failed .dat sync takes the tombstone back off the .dat.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: honor fsync=true on DELETE and forward it to replicas

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): fsync a needle delete that asks for it

Mirrors the Go volume server: DELETE with fsync=true flushes the tombstone
before the index drops the needle and the .idx before the ack, and a failed
.dat flush takes the tombstone back off the .dat.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: let a durable delete through on a full volume

A queued delete hit the batch worker's volume size check, which a
tombstone append is exempt from inline, so fsync=true deletes failed once
the volume reached its hard limit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: sync a durable delete of an already-deleted needle

An earlier delete without fsync may have dropped the needle with its
tombstone still unsynced, so the inline path syncs before acking anyway,
as the batch worker already does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: refuse a batch request that races the worker stopping

stopWorker can close the batch channel between startWorker handing it out
and the send, which panicked. The request now falls back to the inline path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): sync a durable delete of an already-deleted needle

Mirrors the Go volume server: an earlier delete without fsync may have
dropped the needle with its tombstone still unsynced.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): run a durable delete on a blocking thread

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: keep syncDelete's signature for existing callers

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: refuse a write a delete cancelled before it appends

A write cancelled while still queued behind the volume locks used to
append, publish and then tombstone itself without a sync, after its
delete may already have been acknowledged as durable. A crash could
keep the row and lose the tombstone. Refuse it before the append.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): refuse a write a delete cancelled before it appends

Mirrors the Go change: a write cancelled while still queued behind the
volume locks no longer appends, publishes and tombstones itself without
a sync after its delete may already be durable.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): drop the write guards for a durable delete's flushes

The durable delete held the write lease and the io write guard across
both fsyncs, so reads of the volume waited on them. Append and publish
the tombstone under the guards and flush with only the io read side, as
write_synced_run and the Go path already do. do_delete_request keeps its
signature.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: stop taking writes when a durable delete fails to index

The needle map drops the needle before it appends the .idx row, so after
a failed append a durable retry found nothing to delete and acked, and a
later write could bury the unindexed tombstone so a reload brought the
needle back. Fence the volume the way a failed durable write does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): stop taking writes when a durable delete fails to index

Rust counterpart of the Go change: the needle map drops the needle
before it appends the .idx row, so a failed append left a durable retry
nothing to delete, and a later write could bury the unindexed tombstone.
publish_delete_request now fences the volume for a durable delete, as
publish_write does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: stop taking writes when any delete fails to index

A non-durable delete whose .idx append failed left the volume writable with
the needle already dropped from the map, so a later durable retry found
nothing to delete and acknowledged without the missing row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): stop taking writes when any delete fails to index

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume: take the fsync flag on the existing delete functions

DeleteVolumeNeedle, deleteNeedle2 and syncDelete take fsync instead of
gaining *WithFsync variants.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* volume(rs): take the fsync flag on the existing delete functions

delete_volume_needle and delete_needle take fsync instead of gaining
*_with_fsync variants; the durable path and the deletable checks fold
back into delete_needle, and the cancelled-write check mirrors Go's
isCancelledWrite.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-18 15:18:29 -08:00

445 lines
15 KiB
Go

package topology
import (
"context"
"encoding/json"
"errors"
"fmt"
"net/http"
"net/url"
"strconv"
"strings"
"sync"
"time"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/operation"
"github.com/seaweedfs/seaweedfs/weed/security"
"github.com/seaweedfs/seaweedfs/weed/stats"
"github.com/seaweedfs/seaweedfs/weed/storage"
"github.com/seaweedfs/seaweedfs/weed/storage/needle"
"github.com/seaweedfs/seaweedfs/weed/storage/types"
"github.com/seaweedfs/seaweedfs/weed/util"
"github.com/seaweedfs/seaweedfs/weed/util/buffer_pool"
util_http "github.com/seaweedfs/seaweedfs/weed/util/http"
"google.golang.org/grpc"
)
var ErrorWriteRefused = errors.New("write refused before append")
// ReplicatedWrite writes a needle to the local volume and fans it out to all
// remote replica locations. When type=replicate is set, the request is itself
// a forwarded replication and no further remote lookups are performed.
// Returns isUnchanged=true when the local write determined the needle content
// was already present.
func ReplicatedWrite(ctx context.Context, masterFn operation.GetMasterFn, grpcDialOption grpc.DialOption, s *storage.Store, volumeId needle.VolumeId, n *needle.Needle, r *http.Request, contentMd5 string) (isUnchanged bool, err error) {
//check JWT
jwt := security.GetJwt(r)
// check whether this is a replicated write request
var remoteLocations []operation.Location
if r.FormValue("type") != "replicate" {
// this is the initial request
remoteLocations, err = GetWritableRemoteReplications(s, grpcDialOption, volumeId, masterFn)
if err != nil {
glog.V(0).Infoln(err)
return
}
}
// with no local volume and no replica to forward to, the write would be acknowledged without being stored
vol := s.GetVolume(volumeId)
if vol == nil && len(remoteLocations) == 0 {
err = fmt.Errorf("volume %d not found on %s:%d", volumeId, s.Ip, s.Port)
glog.V(0).Infoln(err)
return
}
// read fsync value
fsync := false
if r.FormValue("fsync") == "true" {
fsync = true
}
replicaCount := len(remoteLocations)
if replicaCount > 0 {
// Record replication duration histogram for the overall write operation
defer func(t time.Time) {
stats.VolumeServerReplicationHistogram.WithLabelValues(stats.ReplicationOpWrite).Observe(time.Since(t).Seconds())
}(time.Now())
}
// Snapshot the fields the fan-out reads before the concurrent local write
// mutates the needle (it stamps the volume TTL and flags), so the two cannot
// race; the fan-out then touches only these locals plus read-only n.Data.
replicaTtl := n.Ttl
if vol != nil && replicaTtl == needle.EMPTY_TTL && vol.Ttl != needle.EMPTY_TTL {
replicaTtl = vol.Ttl
}
replicaLastModified := n.LastModified
replicaIsChunked := n.IsChunkedManifest()
replicaIsCompressed := n.IsCompressed()
replicaName := string(n.Name)
replicaMime := string(n.Mime)
replicaData := n.Data
replicaPairMap := make(map[string]string)
if replicaCount > 0 && n.HasPairs() {
tmpMap := make(map[string]string)
if pe := json.Unmarshal(n.Pairs, &tmpMap); pe != nil {
stats.VolumeServerHandlerCounter.WithLabelValues(stats.ErrorUnmarshalPairs).Inc()
glog.V(0).Infoln("Unmarshal pairs error:", pe)
}
for k, v := range tmpMap {
replicaPairMap[needle.PairNamePrefix+k] = v
}
}
// Observe replication targets histogram for all operations (including zero)
stats.VolumeServerReplicationTargets.Observe(float64(replicaCount))
// Run the local disk write and the replica fan-out concurrently against the
// buffered needle (latency max(local, replicas), not the sum); ack after both.
var wg sync.WaitGroup
var localErr, remoteErr error
var localIsUnchanged bool
if vol != nil {
wg.Add(1)
go func() {
defer wg.Done()
start := time.Now()
inFlightGauge := stats.VolumeServerInFlightRequestsGauge.WithLabelValues(stats.WriteToLocalDisk)
inFlightGauge.Inc()
defer inFlightGauge.Dec()
localIsUnchanged, localErr = s.WriteVolumeNeedle(volumeId, n, true, fsync)
stats.VolumeServerRequestHistogram.WithLabelValues(stats.WriteToLocalDisk).Observe(time.Since(start).Seconds())
if localErr != nil {
stats.VolumeServerHandlerCounter.WithLabelValues(stats.ErrorWriteToLocalDisk).Inc()
localErr = fmt.Errorf("failed to write to local disk: %w", localErr)
glog.V(0).Infoln(localErr)
}
}()
}
if replicaCount > 0 {
wg.Add(1)
go func() {
defer wg.Done()
start := time.Now()
inFlightGauge := stats.VolumeServerInFlightRequestsGauge.WithLabelValues(stats.WriteToReplicas)
inFlightGauge.Inc()
defer inFlightGauge.Dec()
remoteErr = DistributedOperation(ctx, remoteLocations, func(ctx context.Context, location operation.Location) error {
u := url.URL{
Scheme: "http",
Host: location.Url,
Path: r.URL.Path,
}
q := url.Values{
"type": {"replicate"},
"ttl": {replicaTtl.String()},
}
if fsync {
q.Set("fsync", "true")
}
if replicaLastModified > 0 {
q.Set("ts", strconv.FormatUint(replicaLastModified, 10))
}
if replicaIsChunked {
q.Set("cm", "true")
}
u.RawQuery = q.Encode()
bytesBuffer := buffer_pool.SyncPoolGetBuffer()
defer buffer_pool.SyncPoolPutBuffer(bytesBuffer)
// volume server do not know about encryption
uploadOption := &operation.UploadOption{
UploadUrl: u.String(),
Filename: replicaName,
Cipher: false,
IsInputCompressed: replicaIsCompressed,
IsReplication: true,
MimeType: replicaMime,
PairMap: replicaPairMap,
Jwt: jwt,
Md5: contentMd5,
BytesBuffer: bytesBuffer,
MaxAttempts: 1, // fail fast on a dead replica; the client write retries
}
uploader, uploaderErr := operation.NewUploader()
if uploaderErr != nil {
glog.Errorf("replication-UploadData, err:%v, url:%s", uploaderErr, u.String())
return uploaderErr
}
uploadErr := replicateTo(ctx, uploader, replicaData, uploadOption)
if uploadErr != nil {
glog.Errorf("replication-UploadData, err:%v, url:%s", uploadErr, u.String())
}
return uploadErr
})
stats.VolumeServerRequestHistogram.WithLabelValues(stats.WriteToReplicas).Observe(time.Since(start).Seconds())
if remoteErr != nil {
stats.VolumeServerHandlerCounter.WithLabelValues(stats.ErrorWriteToReplicas).Inc()
stats.VolumeServerReplicationCounter.WithLabelValues(stats.ReplicationOpWrite, stats.ReplicationFailure).Inc()
reason := classifyReplicationError(remoteErr)
stats.VolumeServerReplicationFailures.WithLabelValues(stats.ReplicationOpWrite, reason).Inc()
remoteErr = fmt.Errorf("failed to write to replicas for volume %d: %v", volumeId, remoteErr)
glog.V(0).Infoln(remoteErr)
} else {
stats.VolumeServerReplicationCounter.WithLabelValues(stats.ReplicationOpWrite, stats.ReplicationSuccess).Inc()
}
}()
}
wg.Wait()
if localErr != nil {
if errors.Is(localErr, storage.ErrorReadOnly) && replicaCount == 0 && !fsync && r.FormValue("type") != "replicate" {
localErr = fmt.Errorf("%w: %w", ErrorWriteRefused, localErr)
}
return false, localErr
}
if remoteErr != nil {
return false, remoteErr
}
return localIsUnchanged, nil
}
// replicaRequestTimeout bounds one replica forward with no body, from connect
// to the end of the response, so a silently partitioned peer cannot hold the
// write. Mirrors the Rust volume server's REPLICA_REQUEST_TIMEOUT.
var replicaRequestTimeout = 10 * time.Second
// replicaMinBytesPerSec is the slowest link a forward's body may cross within
// its bound, so a large write is not cut off on a link slower than a LAN.
const replicaMinBytesPerSec = 1 << 20
func replicaRequestBound(size int) time.Duration {
return replicaRequestTimeout + time.Duration(int64(size)*1000/replicaMinBytesPerSec)*time.Millisecond
}
func replicateTo(ctx context.Context, uploader *operation.Uploader, data []byte, option *operation.UploadOption) error {
bound := replicaRequestBound(len(data))
ctx, cancel := context.WithTimeout(ctx, bound)
defer cancel()
_, err := uploader.UploadData(ctx, data, option)
if err != nil && errors.Is(ctx.Err(), context.DeadlineExceeded) && !errors.Is(err, context.DeadlineExceeded) {
return fmt.Errorf("%w after %v: %v", context.DeadlineExceeded, bound, err)
}
return err
}
// ReplicatedDelete deletes a needle from the local volume and sends delete
// requests to all remote replica locations. Replica deletes use
// context.Background() so that a client disconnect does not orphan replica
// deletes.
func ReplicatedDelete(masterFn operation.GetMasterFn, grpcDialOption grpc.DialOption, store *storage.Store, volumeId needle.VolumeId, n *needle.Needle, r *http.Request) (size types.Size, err error) {
//check JWT
jwt := security.GetJwt(r)
var remoteLocations []operation.Location
if r.FormValue("type") != "replicate" {
remoteLocations, err = GetRemoteReplications(store, grpcDialOption, volumeId, masterFn)
if err != nil {
glog.V(0).Infoln(err)
return
}
}
replicaCount := len(remoteLocations)
if replicaCount > 0 {
// Record replication duration and operation counter for delete
defer func(t time.Time) {
stats.VolumeServerReplicationHistogram.WithLabelValues(stats.ReplicationOpDelete).Observe(time.Since(t).Seconds())
if err != nil {
stats.VolumeServerReplicationCounter.WithLabelValues(stats.ReplicationOpDelete, stats.ReplicationFailure).Inc()
} else {
stats.VolumeServerReplicationCounter.WithLabelValues(stats.ReplicationOpDelete, stats.ReplicationSuccess).Inc()
}
}(time.Now())
}
fsync := r.FormValue("fsync") == "true"
size, err = store.DeleteVolumeNeedle(volumeId, n, fsync)
if err != nil {
glog.V(0).Infoln("delete error:", err)
return
}
// Observe replication targets histogram for all operations (including zero)
stats.VolumeServerReplicationTargets.Observe(float64(replicaCount))
if replicaCount > 0 { //send to other replica locations
// background, not r.Context(): a client disconnect must not orphan replica deletes
if err = DistributedOperation(context.Background(), remoteLocations, func(ctx context.Context, location operation.Location) error {
query := "?type=replicate"
if fsync {
query += "&fsync=true"
}
url, normalizeErr := util_http.GetGlobalHttpClient().NormalizeHttpScheme(location.Url + r.URL.Path + query)
if normalizeErr != nil {
return normalizeErr
}
if jwt != "" && (!strings.HasPrefix(url, "https://") || !util_http.GetGlobalHttpClient().IsTLSVerified()) {
return fmt.Errorf("refusing to forward delete authorization to %s without HTTPS", location.Url)
}
return util_http.Delete(url, string(jwt))
}); err != nil {
reason := classifyReplicationError(err)
stats.VolumeServerReplicationFailures.WithLabelValues(stats.ReplicationOpDelete, reason).Inc()
size = 0
}
}
return
}
type DistributedOperationResult map[string]error
func (dr DistributedOperationResult) Error() error {
var msgs []string
var errs []error
for k, v := range dr {
if v != nil {
msgs = append(msgs, fmt.Sprintf("[%s]: %v", k, v))
errs = append(errs, v)
}
}
if len(errs) == 0 {
return nil
}
return &distributedOperationError{msg: strings.Join(msgs, "\n"), errs: errs}
}
// distributedOperationError keeps each replica's error reachable through
// errors.Is, so a forward that hit its bound is classified as a timeout.
type distributedOperationError struct {
msg string
errs []error
}
func (e *distributedOperationError) Error() string { return e.msg }
func (e *distributedOperationError) Unwrap() []error { return e.errs }
type RemoteResult struct {
Host string
Error error
}
func DistributedOperation(ctx context.Context, locations []operation.Location, op func(ctx context.Context, location operation.Location) error) error {
length := len(locations)
if length == 0 {
return nil
}
// cancel outstanding replica ops once the outcome is decided
ctx, cancel := context.WithCancel(ctx)
defer cancel()
// buffered so a straggler (e.g. a replica stalled on a TCP dial timeout) can
// still deliver its result and exit after we have already returned.
resultCh := make(chan RemoteResult, length)
for _, location := range locations {
go func(location operation.Location) {
resultCh <- RemoteResult{location.Url, op(ctx, location)}
}(location)
}
ret := DistributedOperationResult(make(map[string]error))
for i := 0; i < length; i++ {
result := <-resultCh
ret[result.Host] = result.Error
if result.Error != nil {
// fail fast on the first error instead of waiting for slow replicas
return ret.Error()
}
}
return ret.Error()
}
func GetRemoteReplications(s *storage.Store, grpcDialOption grpc.DialOption, volumeId needle.VolumeId, masterFn operation.GetMasterFn) ([]operation.Location, error) {
return getRemoteReplications(s, grpcDialOption, volumeId, masterFn, false)
}
func GetWritableRemoteReplications(s *storage.Store, grpcDialOption grpc.DialOption, volumeId needle.VolumeId, masterFn operation.GetMasterFn) ([]operation.Location, error) {
return getRemoteReplications(s, grpcDialOption, volumeId, masterFn, true)
}
func getRemoteReplications(s *storage.Store, grpcDialOption grpc.DialOption, volumeId needle.VolumeId, masterFn operation.GetMasterFn, writableOnly bool) (remoteLocations []operation.Location, err error) {
v := s.GetVolume(volumeId)
if v != nil && v.ReplicaPlacement.GetCopyCount() == 1 {
return
}
// not on local store, or has replications
lookupResults, lookupErr := operation.LookupVolumeIds(masterFn, grpcDialOption, []string{volumeId.String()}, false)
lookupResult := lookupResults[volumeId.String()]
writableLocations := 0
if lookupErr == nil {
if lookupResult == nil {
err = fmt.Errorf("replicating lookup returned no result for %d", volumeId)
return
}
selfUrl := util.JoinHostPort(s.Ip, s.Port)
for _, location := range lookupResult.Locations {
if writableOnly && location.ReadOnly {
continue
}
if !writableOnly && location.ReadOnly && !location.ReadOnlyCanDelete {
continue
}
writableLocations++
if location.Url != selfUrl {
remoteLocations = append(remoteLocations, location)
}
}
} else {
err = fmt.Errorf("replicating lookup failed for %d: %v", volumeId, lookupErr)
return
}
if v != nil {
// has one local and has remote replications
copyCount := v.ReplicaPlacement.GetCopyCount()
if writableLocations < copyCount {
operation.InvalidateVolumeIdLocationCache(volumeId.String())
label := "replication"
if writableOnly {
label = "writable replication"
}
err = fmt.Errorf("%s locations [%d] is less than volume %d replication copy count [%d]",
label, writableLocations, volumeId, copyCount)
}
}
return
}
// classifyReplicationError maps a Go error to a bounded-cardinality failure
// reason label for the replication_failures_total metric. Returns an empty
// string for nil errors.
func classifyReplicationError(err error) string {
if err == nil {
return ""
}
if errors.Is(err, context.DeadlineExceeded) {
return stats.FailureTimeout
}
if errors.Is(err, context.Canceled) {
return stats.FailureContextCancelled
}
errStr := err.Error()
if strings.Contains(errStr, "connection refused") {
return stats.FailureConnectionRefused
}
return stats.FailureServerError
}