Files
seaweedfs/weed/worker/tasks/s3_lifecycle/handler.go
T
Chris LuandClaude Opus 5.5 799fd10342 s3 lifecycle: fail over between S3 fronts during a pass (#12647)
* s3 lifecycle: add a LifecycleDelete client that fails over between S3 fronts

When the current front returns Unavailable, the client closes it, marks it
failed for the rest of the pass, picks the next front that passes
checkLifecycleReady and resends the same request there. Callers cut off by
that switch closing their connection resend on the new front. When no front
is left, every later call returns one error naming the fronts tried.

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

* s3 lifecycle: serve the pass through the failover client

Execute hands the replay drain and the walker one failover client built on
the selected front, reports each switch as job progress, and closes the
current connection when the pass ends.

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

* s3 lifecycle: keep failing over until a front serves the resent delete

The resend after a switch went straight to the new front, so a second
lost front ended the call even with a healthy third one, and the walker
has no retry of its own. Resend through the failover loop instead.

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

* s3 lifecycle: do not recheck fronts skipped when the pass started

The failover client started with no failed fronts, so losing the selected
front checked the fronts selectS3Front had already skipped again, holding
every delete for another readiness timeout per front.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-15 12:37:07 -08:00

516 lines
21 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package s3_lifecycle
import (
"context"
"errors"
"fmt"
"math"
"strconv"
"time"
"github.com/seaweedfs/seaweedfs/weed/credential"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
"github.com/seaweedfs/seaweedfs/weed/pb/plugin_pb"
"github.com/seaweedfs/seaweedfs/weed/pb/s3_lifecycle_pb"
pluginworker "github.com/seaweedfs/seaweedfs/weed/plugin/worker"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3lifecycle"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3lifecycle/dailyrun"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3lifecycle/dispatcher"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3lifecycle/engine"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3lifecycle/scheduler"
"github.com/seaweedfs/seaweedfs/weed/stats"
"golang.org/x/time/rate"
"google.golang.org/grpc"
)
func init() {
pluginworker.RegisterHandler(pluginworker.HandlerFactory{
JobType: jobType,
Category: pluginworker.CategoryDefault,
Aliases: []string{"s3-lifecycle", "s3.lifecycle", "lifecycle"},
Build: func(opts pluginworker.HandlerBuildOptions) (pluginworker.JobHandler, error) {
return NewHandler(opts.GrpcDialOption), nil
},
})
}
// Handler is the worker-side runner for S3 object lifecycle expiration.
// One Execute call drives one bounded dailyrun.Run pass against the S3
// endpoints discovered from the master; admin caps concurrency at one
// job per worker so a fresh proposal only spawns a new run after the
// prior one exits.
type Handler struct {
grpcDialOption grpc.DialOption
}
func NewHandler(grpcDialOption grpc.DialOption) *Handler {
return &Handler{grpcDialOption: grpcDialOption}
}
func (h *Handler) Capability() *plugin_pb.JobTypeCapability {
return &plugin_pb.JobTypeCapability{
JobType: jobType,
CanDetect: true,
CanExecute: true,
MaxDetectionConcurrency: 1,
MaxExecutionConcurrency: 1,
DisplayName: "S3 Lifecycle",
Description: "Daily batch: scan the filer meta-log and delete objects whose lifecycle rule has fired.",
Weight: 20,
}
}
func (h *Handler) Descriptor() *plugin_pb.JobTypeDescriptor {
return &plugin_pb.JobTypeDescriptor{
JobType: jobType,
DisplayName: "S3 Lifecycle",
Description: "Daily S3 object expiration scan with per-shard cursors and bounded retry.",
Icon: "fas fa-recycle",
DescriptorVersion: 1,
AdminConfigForm: &plugin_pb.ConfigForm{
FormId: "s3-lifecycle-admin",
Title: "S3 Lifecycle",
Description: "Cluster-wide delete-throughput cap.",
Sections: []*plugin_pb.ConfigSection{
{
SectionId: "scope",
Title: "Scope",
Description: "Cluster-wide delete-throughput cap.",
Fields: []*plugin_pb.ConfigField{
{
Name: ClusterDeletesPerSecondAdminKey,
Label: "Cluster Delete Rate (per second)",
Description: "Cluster-wide ceiling on lifecycle delete RPCs per second, divided evenly across active s3_lifecycle workers at job-dispatch time. 0 = unlimited.",
FieldType: plugin_pb.ConfigFieldType_CONFIG_FIELD_TYPE_INT64,
Widget: plugin_pb.ConfigWidget_CONFIG_WIDGET_NUMBER,
MinValue: &plugin_pb.ConfigValue{Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
},
{
Name: ClusterDeletesBurstAdminKey,
Label: "Cluster Delete Burst",
Description: "Token-bucket burst capacity across the cluster (max simultaneous deletes). 0 = 2 × rate. Same allocation rule as the rate.",
FieldType: plugin_pb.ConfigFieldType_CONFIG_FIELD_TYPE_INT64,
Widget: plugin_pb.ConfigWidget_CONFIG_WIDGET_NUMBER,
MinValue: &plugin_pb.ConfigValue{Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
},
{
Name: MetaLogRetentionDaysAdminKey,
Label: "Meta-Log Retention (days)",
Description: "How far back the filer's meta-log subscription can reach. Rules whose TTL exceeds this run via the walker; shrinking this value will trigger a one-time recovery walk on the next run for any rule that's now too old to replay. 0 = unbounded (no partition; every rule serviced by replay).",
FieldType: plugin_pb.ConfigFieldType_CONFIG_FIELD_TYPE_INT64,
Widget: plugin_pb.ConfigWidget_CONFIG_WIDGET_NUMBER,
MinValue: &plugin_pb.ConfigValue{Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
},
{
Name: WalkerIntervalMinutesAdminKey,
Label: "Walker Interval (minutes)",
Description: "Minimum time between steady-state walker fires per shard. Cold-start and rule-change recovery walks ignore this — they run unconditionally. 0 = fire on every run (use when the worker is scheduled at the desired walk cadence, e.g. hourly). Set to a positive value when the worker runs at a tighter cadence than the desired walk frequency, to avoid hammering filer with a full subtree scan per run.",
FieldType: plugin_pb.ConfigFieldType_CONFIG_FIELD_TYPE_INT64,
Widget: plugin_pb.ConfigWidget_CONFIG_WIDGET_NUMBER,
MinValue: &plugin_pb.ConfigValue{Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
},
},
},
},
DefaultValues: map[string]*plugin_pb.ConfigValue{
ClusterDeletesPerSecondAdminKey: {Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
ClusterDeletesBurstAdminKey: {Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
MetaLogRetentionDaysAdminKey: {Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
WalkerIntervalMinutesAdminKey: {Kind: &plugin_pb.ConfigValue_Int64Value{Int64Value: 0}},
},
},
// WorkerConfigForm intentionally absent — the prior
// "Per-Run Time Limit (minutes)" knob duplicated the admin
// scheduler's Execution Timeout (both wall-clock caps on the
// same Execute call), and operators had to keep the two values
// in agreement. Removed in favor of a single source of truth:
// AdminRuntimeDefaults.ExecutionTimeoutSeconds below.
AdminRuntimeDefaults: &plugin_pb.AdminRuntimeDefaults{
// On by default: S3 lifecycle is a standard bucket feature
// (PutBucketLifecycleConfiguration is part of the S3 API),
// and a bucket with rules set but no worker running silently
// retains data past its declared expiration. Operators who
// want the worker off can still disable it in the admin UI;
// the default error is "data lingers" not "worker burns CPU
// on empty rule sets" (the worker fast-exits with no
// configured rules).
Enabled: true,
DetectionIntervalMinutes: 24 * 60, // daily
DetectionTimeoutSeconds: 60,
MaxJobsPerDetection: 1,
// Effectively no per-pass wall-clock cap. Lifecycle is a
// scheduled batch — its natural duration is "as long as it
// takes to process today's events." The scheduler's global
// 90s default would kill every real run, and a numeric
// cap operators have to estimate (1h? 8h?) is a recurring
// footgun: too low truncates a legitimate large-bucket
// pass; too high makes the value meaningless.
//
// Use math.MaxInt32 seconds (~68 years) for both knobs to
// say "no timeout in practice" in code-review-readable form.
// Operators who genuinely want a cap can set one in the
// admin UI; the underlying context.WithTimeout machinery
// is unchanged.
ExecutionTimeoutSeconds: math.MaxInt32,
JobTypeMaxRuntimeSeconds: math.MaxInt32,
},
}
}
func (h *Handler) Detect(ctx context.Context, request *plugin_pb.RunDetectionRequest, sender pluginworker.DetectionSender) error {
if request == nil || sender == nil {
return fmt.Errorf("detect: nil request or sender")
}
if request.JobType != "" && request.JobType != jobType {
return fmt.Errorf("job type %q is not handled by s3 lifecycle handler", request.JobType)
}
s3Endpoints := clusterS3Endpoints(request.ClusterContext)
if len(s3Endpoints) == 0 {
_ = sender.SendActivity(pluginworker.BuildDetectorActivity("skipped", "no s3 servers registered with master", nil))
return sender.SendComplete(&plugin_pb.DetectionComplete{JobType: jobType, Success: true})
}
filerAddresses := []string{}
if request.ClusterContext != nil {
filerAddresses = append(filerAddresses, request.ClusterContext.FilerAddresses...)
}
if len(filerAddresses) == 0 {
_ = sender.SendActivity(pluginworker.BuildDetectorActivity("skipped", "no filer addresses in cluster context", nil))
return sender.SendComplete(&plugin_pb.DetectionComplete{JobType: jobType, Success: true})
}
// FilerAddresses are pb.ServerAddress strings (host:httpPort.grpcPort);
// Execute dials filer_grpc_address verbatim, so resolve it to a gRPC address.
filerGrpcAddress := pb.ServerAddress(filerAddresses[0]).ToGrpcAddress()
proposal := &plugin_pb.JobProposal{
JobType: jobType,
ProposalId: fmt.Sprintf("s3-lifecycle-%d", time.Now().UnixNano()),
Priority: plugin_pb.JobPriority_JOB_PRIORITY_NORMAL,
Parameters: map[string]*plugin_pb.ConfigValue{
"filer_grpc_address": {Kind: &plugin_pb.ConfigValue_StringValue{StringValue: filerGrpcAddress}},
},
}
if err := sender.SendProposals(&plugin_pb.DetectionProposals{
JobType: jobType,
Proposals: []*plugin_pb.JobProposal{proposal},
}); err != nil {
return err
}
return sender.SendComplete(&plugin_pb.DetectionComplete{
JobType: jobType,
Success: true,
TotalProposals: 1,
})
}
func (h *Handler) Execute(ctx context.Context, request *plugin_pb.ExecuteJobRequest, sender pluginworker.ExecutionSender) error {
if request == nil || request.Job == nil || sender == nil {
return fmt.Errorf("execute: nil request/job/sender")
}
if request.Job.JobType != "" && request.Job.JobType != jobType {
return fmt.Errorf("job type %q is not handled by s3 lifecycle handler", request.Job.JobType)
}
cfg := ParseConfig(request.GetAdminConfigValues(), request.GetWorkerConfigValues())
s3Endpoints := clusterS3Endpoints(request.ClusterContext)
if len(s3Endpoints) == 0 {
return fmt.Errorf("execute: no s3 servers registered with master")
}
filerAddress := readString(request.Job.Parameters, "filer_grpc_address", "")
if filerAddress == "" {
return fmt.Errorf("execute: missing filer_grpc_address in job parameters")
}
// Run lifetime is bounded by the scheduler's Execution Timeout
// (admin UI). The handler used to wrap ctx in another
// context.WithTimeout(cfg.MaxRuntime), but that doubled the
// concept — both knobs were wall-clock caps on the same Execute
// call, and the smaller one always won (typically the 90s
// scheduler default would clobber a 60-min worker setting).
// Single source of truth is now AdminRuntimeDefaults.ExecutionTimeoutSeconds.
runCtx := ctx
_ = sender.SendProgress(&plugin_pb.JobProgressUpdate{
JobId: request.Job.JobId, JobType: jobType,
State: plugin_pb.JobState_JOB_STATE_RUNNING, Stage: "starting",
Message: fmt.Sprintf("scheduler workers=%d s3=%v", cfg.Workers, s3Endpoints),
})
dialCtx, dialCancel := context.WithTimeout(runCtx, 30*time.Second)
filerConn, err := pb.GrpcDial(dialCtx, filerAddress, false, h.grpcDialOption)
dialCancel()
if err != nil {
return fmt.Errorf("dial filer %s: %w", filerAddress, err)
}
defer filerConn.Close()
filerClient := filer_pb.NewSeaweedFilerClient(filerConn)
bucketsPath, err := lookupBucketsPath(runCtx, filerClient)
if err != nil {
return fmt.Errorf("buckets path: %w", err)
}
s3Conn, s3Endpoint, skipped := h.selectS3Front(runCtx, s3Endpoints)
if s3Conn == nil {
return noReadyS3Error(skipped)
}
client := h.newFailoverClient(runCtx, s3Conn, s3Endpoint, s3Endpoints, func(msg string) {
_ = sender.SendProgress(&plugin_pb.JobProgressUpdate{
JobId: request.Job.JobId, JobType: jobType,
State: plugin_pb.JobState_JOB_STATE_RUNNING, Stage: "running",
Message: msg,
})
})
defer client.Close()
if len(skipped) > 0 {
skippedErr := errors.Join(skipped...)
glog.Warningf("s3 lifecycle: using s3 %s, skipped: %v", s3Endpoint, skippedErr)
_ = sender.SendProgress(&plugin_pb.JobProgressUpdate{
JobId: request.Job.JobId, JobType: jobType,
State: plugin_pb.JobState_JOB_STATE_RUNNING, Stage: "starting",
Message: fmt.Sprintf("using s3 %s, skipped: %v", s3Endpoint, skippedErr),
})
}
report, err := h.executeDailyReplay(runCtx, request, bucketsPath, filerClient, client, cfg, sender)
if err != nil {
return err
}
return sendCompletion(request, sender, report)
}
// JobCompleted must set JobType to the handler's jobType constant;
// routed requests may leave request.Job.JobType empty, and the admin
// ignores completions with empty JobType. Failed passes carry their
// result too, so every bucket failure stays in the job record.
func sendCompletion(request *plugin_pb.ExecuteJobRequest, sender pluginworker.ExecutionSender, report *runReport) error {
return sender.SendCompleted(&plugin_pb.JobCompleted{
JobId: request.Job.JobId,
JobType: jobType,
Success: report.err == nil,
ErrorMessage: report.errorMessage(),
Result: &plugin_pb.JobResult{
Summary: report.summaryLine(),
OutputValues: report.outputValues(),
},
})
}
// executeDailyReplay runs one bounded daily-replay pass via
// dailyrun.Run. The walker fires inside runShard on rule-content edits
// and against the steady-state walk view; all rule kinds are serviced.
// The returned error covers setup only; the pass's own error is in the
// report.
func (h *Handler) executeDailyReplay(ctx context.Context, request *plugin_pb.ExecuteJobRequest, bucketsPath string, filerClient filer_pb.SeaweedFilerClient, client dailyrun.LifecycleClient, cfg Config, sender pluginworker.ExecutionSender) (*runReport, error) {
startedAt := time.Now()
eng := engine.New()
inputs, parseErrors, err := scheduler.LoadCompileInputs(ctx, filerClient, bucketsPath)
if err != nil {
return nil, fmt.Errorf("daily_replay: load lifecycle inputs: %w", err)
}
for _, pe := range parseErrors {
glog.Warningf("daily_replay: invalid lifecycle policy on bucket %s: %v", pe.Bucket, pe.Err)
}
eng.Compile(inputs, engine.CompileOptions{PriorStates: scheduler.AllActivePriorStates(inputs)})
shards := make([]int, 0, s3lifecycle.ShardCount)
for i := 0; i < s3lifecycle.ShardCount; i++ {
shards = append(shards, i)
}
limiter, limiterDesc := buildLimiterFromClusterContext(request.GetClusterContext())
// Bucket list for the walker — derived from inputs so it matches
// the snapshot the engine compiled.
buckets := make([]string, 0, len(inputs))
for _, in := range inputs {
if in.Bucket != "" {
buckets = append(buckets, in.Bucket)
}
}
persister := &dailyrun.FilerCursorPersister{Store: dispatcher.NewFilerStoreClient(filerClient)}
report := &runReport{bucketsTotal: len(buckets) + len(parseErrors), invalidPolicies: parseErrors}
loadLastFullPass(ctx, persister, report)
runStats := &dailyrun.RunStats{}
// Share the limiter with processMatches so walker + replay can't
// combine to burst past the cluster cap.
walkerDispatch := &dailyrun.WalkerDispatcher{Client: client, Limiter: limiter, Stats: runStats}
walker := dailyrun.WalkerFunc(func(walkCtx context.Context, view *engine.Snapshot, shards []int) (time.Time, error) {
return dailyrun.WalkBuckets(walkCtx, view, shards, buckets,
dailyrun.FilerListFunc(runStats.CountListed(filerClient), bucketsPath, view), walkerDispatch, runStats, persister)
})
_ = sender.SendProgress(&plugin_pb.JobProgressUpdate{
JobId: request.Job.JobId, JobType: jobType,
State: plugin_pb.JobState_JOB_STATE_RUNNING, Stage: "starting",
Message: fmt.Sprintf("daily_replay shards=%d workers=%d rate=%s buckets=%d invalid_policies=%d walker=on",
len(shards), cfg.Workers, limiterDesc, len(buckets), len(parseErrors)),
})
stopProgress := startProgressReporter(request, sender, runStats.Summary, progressInterval)
report.err = dailyrun.Run(ctx, dailyrun.Config{
Shards: shards,
BucketsPath: bucketsPath,
Engine: eng,
FilerClient: filerClient,
Client: client,
Persister: persister,
Lister: dispatcher.NewFilerSiblingLister(filerClient, bucketsPath),
Workers: cfg.Workers,
Limiter: limiter,
RetentionWindow: cfg.MetaLogRetention,
Walker: walker,
WalkerInterval: cfg.WalkerInterval,
ClientName: "worker-s3-lifecycle-daily",
Stats: runStats,
})
stopProgress()
report.canceled = ctx.Err() != nil
report.elapsed = time.Since(startedAt)
report.summary = runStats.Summary()
if report.err != nil {
glog.Warningf("daily_replay: %v", report.err)
}
recordFullPass(ctx, persister, report, startedAt)
return report, nil
}
func loadLastFullPass(ctx context.Context, persister *dailyrun.FilerCursorPersister, report *runReport) {
fp, found, err := persister.LoadFullPass(ctx)
if err != nil {
glog.Warningf("daily_replay: %v", err)
return
}
if found {
report.lastFullPass = fp.CompletedAt
publishFullPass(fp)
}
}
// recordFullPass advances the full-pass record only when the pass
// covered every shard and every policy.
func recordFullPass(ctx context.Context, persister *dailyrun.FilerCursorPersister, report *runReport, startedAt time.Time) {
if !report.fullPass() {
return
}
fp := dailyrun.FullPass{StartedAt: startedAt.UTC(), CompletedAt: time.Now().UTC()}
saveCtx, saveCancel := context.WithTimeout(ctx, 10*time.Second)
defer saveCancel()
if err := persister.SaveFullPass(saveCtx, fp); err != nil {
glog.Warningf("daily_replay: %v", err)
report.fullPassSaveErr = err
return
}
report.lastFullPass = fp.CompletedAt
publishFullPass(fp)
}
func publishFullPass(fp dailyrun.FullPass) {
stats.S3LifecycleDailyRunLastFullPassTimestampSeconds.Set(float64(fp.CompletedAt.UnixNano()) / 1e9)
}
// buildLimiterFromClusterContext parses the per-worker share the admin
// wrote into ClusterContext.Metadata (see weed/admin/plugin/plugin.go's
// s3_lifecycle injection) and returns a rate.Limiter, or nil when no
// rate cap applies. The description string is for the JobProgressUpdate
// so operators can see "rate=unlimited" / "rate=12.5/s burst=25" in
// the activity log.
//
// Tolerant of missing keys, empty strings, malformed numbers, and
// non-positive values — all treated as "no limit" rather than failing
// the run. The admin allocator is the single point that decides whether
// to populate these keys; the worker doesn't second-guess.
func buildLimiterFromClusterContext(cc *plugin_pb.ClusterContext) (*rate.Limiter, string) {
if cc == nil || cc.Metadata == nil {
return nil, "unlimited"
}
rps, ok := parsePositiveFloat(cc.Metadata[MetadataKeyDeletesPerSecond])
if !ok {
return nil, "unlimited"
}
burst, _ := parsePositiveInt(cc.Metadata[MetadataKeyDeletesBurst])
if burst <= 0 {
// Sensible default: enough headroom for one tick's worth of
// throughput. Caller may also supply 0 to opt into this default.
burst = int(rps * 2)
if burst < 1 {
burst = 1
}
}
return rate.NewLimiter(rate.Limit(rps), burst), fmt.Sprintf("%.3g/s burst=%d", rps, burst)
}
func parsePositiveFloat(s string) (float64, bool) {
if s == "" {
return 0, false
}
v, err := strconv.ParseFloat(s, 64)
if err != nil || v <= 0 {
return 0, false
}
return v, true
}
func parsePositiveInt(s string) (int, bool) {
if s == "" {
return 0, false
}
v, err := strconv.Atoi(s)
if err != nil || v <= 0 {
return 0, false
}
return v, true
}
// clusterS3Endpoints returns the master-discovered S3 gRPC addresses for the
// cluster. The handler uses the first one that passes checkLifecycleReady and
// moves to the next ready one if it becomes unavailable during the pass; the
// master refreshes the list on KeepConnected.
func clusterS3Endpoints(cc *plugin_pb.ClusterContext) []string {
if cc == nil {
return nil
}
out := make([]string, 0, len(cc.S3GrpcAddresses))
for _, addr := range cc.S3GrpcAddresses {
if addr != "" {
out = append(out, addr)
}
}
return out
}
type lifecycleRPCAdapter struct {
c s3_lifecycle_pb.SeaweedS3LifecycleInternalClient
}
func (a lifecycleRPCAdapter) LifecycleDelete(ctx context.Context, req *s3_lifecycle_pb.LifecycleDeleteRequest) (*s3_lifecycle_pb.LifecycleDeleteResponse, error) {
ctx, _ = credential.WithS3InternalAdminAuth(ctx)
return a.c.LifecycleDelete(ctx, req)
}
func lookupBucketsPath(ctx context.Context, client filer_pb.SeaweedFilerClient) (string, error) {
resp, err := client.GetFilerConfiguration(ctx, &filer_pb.GetFilerConfigurationRequest{})
if err != nil {
return "", err
}
if path := resp.GetDirBuckets(); path != "" {
return path, nil
}
return "/buckets", nil
}
func readString(values map[string]*plugin_pb.ConfigValue, field, fallback string) string {
v, ok := values[field]
if !ok || v == nil {
return fallback
}
if k, ok := v.Kind.(*plugin_pb.ConfigValue_StringValue); ok {
return k.StringValue
}
return fallback
}
var _ pluginworker.JobHandler = (*Handler)(nil)