Files
seaweedfs/weed/s3api/filer_util_delete_test.go
T
Chris LuandGitHub 7bb0a1c127 s3: replay a delete whose reply the transport dropped (#11022)
* s3: stop retrying a delete the filer refused for a non-empty folder

The filer looked and the children are there, so the answer will not change.
retryFilerOp spent six attempts and up to 3.1s of backoff on it before the
caller could act on the condition it was already holding.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: thread the request context through the unversioned delete path

doDeleteEntry issued every DeleteEntry on context.Background(), so an S3
client that hung up left the gateway working on its behalf, out of reach of
both cancellation and the per-request retry allowance that
DeleteMultipleObjectsHandler installs.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: treat a cancelled filer RPC as terminal, not transient

isRetryableFilerErr matched context.Canceled and DeadlineExceeded by
sentinel, which only holds while the error is still local. Once it has
crossed gRPC it is a status, so an abandoned request was retried six times
on behalf of a caller that had already gone.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: replay a delete whose reply the transport dropped

A delete is idempotent at the filer, which answers an entry that is already
gone with an empty resp.Error, so a reply lost in transit can be reissued
rather than surfaced. Surfaced, it becomes a 500 on the bucket delete, which
boto3 resends and is then answered NoSuchBucket, or a per-key InternalError
inside the 200 of a multi-object delete, which no SDK retries at all.

The replay runs through retryFilerOp, so it draws on the allowance the
request already installs rather than paying a backoff per key, and stops for
a caller that has gone. rm and rmObject re-enter WithFilerClient per attempt,
so each one walks the failover list again on a connection the failed attempt
had invalidated; the multi-object loop holds one client for the batch, so
there the replay reuses it.

Classification stays structural. The filer reports its own refusals in
resp.Error, which carries no status and has the deleted path - and, for a
recursive delete, the children it stopped on - formatted into it, so no key
name can steer the decision either way.

rm and rmObject now take the caller's context. Cleanup and rollback paths
pass context.Background() deliberately: they have to run whether or not the
caller is still waiting.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD

* s3: share one retry allowance across multipart completion cleanup

The unused-entry loop deletes once per entry, and each delete now retries,
so a filer that stays unavailable held the response for 3.1s per entry after
the object was already committed.

Claude-Session: https://claude.ai/code/session_01XqaJrwgXQ5GSUpyzRbe5nD
2026-08-28 14:30:21 -07:00

339 lines
13 KiB
Go

package s3api
import (
"context"
"errors"
"fmt"
"testing"
"time"
"github.com/seaweedfs/seaweedfs/weed/filer"
"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
"github.com/seaweedfs/seaweedfs/weed/s3api/s3_constants"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
grpc "google.golang.org/grpc"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
type deleteObjectEntryTestClient struct {
filer_pb.SeaweedFilerClient
deleteResp *filer_pb.DeleteEntryResponse
deleteErr error
lookupResp *filer_pb.LookupDirectoryEntryResponse
lookupErr error
updateErr error
// deleteCallErrs scripts the RPC itself per attempt, deleteRespErrs the
// filer's own message: the two ways a delete fails, told apart because
// only the first is a transport that dropped a reply.
deleteCallErrs []error
deleteRespErrs []string
deleteAttempts int
deleteReqs []*filer_pb.DeleteEntryRequest
deleteReq *filer_pb.DeleteEntryRequest
lookupReq *filer_pb.LookupDirectoryEntryRequest
updateReq *filer_pb.UpdateEntryRequest
}
func (c *deleteObjectEntryTestClient) DeleteEntry(ctx context.Context, req *filer_pb.DeleteEntryRequest, _ ...grpc.CallOption) (*filer_pb.DeleteEntryResponse, error) {
attempt := c.deleteAttempts
c.deleteAttempts++
c.deleteReq = req
c.deleteReqs = append(c.deleteReqs, req)
if err := ctx.Err(); err != nil {
return nil, status.FromContextError(err).Err()
}
if attempt < len(c.deleteCallErrs) && c.deleteCallErrs[attempt] != nil {
return nil, c.deleteCallErrs[attempt]
}
if attempt < len(c.deleteRespErrs) && c.deleteRespErrs[attempt] != "" {
return &filer_pb.DeleteEntryResponse{Error: c.deleteRespErrs[attempt]}, nil
}
if c.deleteResp == nil {
return &filer_pb.DeleteEntryResponse{}, c.deleteErr
}
return c.deleteResp, c.deleteErr
}
func (c *deleteObjectEntryTestClient) LookupDirectoryEntry(_ context.Context, req *filer_pb.LookupDirectoryEntryRequest, _ ...grpc.CallOption) (*filer_pb.LookupDirectoryEntryResponse, error) {
c.lookupReq = req
if c.lookupResp == nil {
return &filer_pb.LookupDirectoryEntryResponse{}, c.lookupErr
}
return c.lookupResp, c.lookupErr
}
func (c *deleteObjectEntryTestClient) UpdateEntry(_ context.Context, req *filer_pb.UpdateEntryRequest, _ ...grpc.CallOption) (*filer_pb.UpdateEntryResponse, error) {
c.updateReq = req
return &filer_pb.UpdateEntryResponse{}, c.updateErr
}
func TestDeleteObjectEntryDemotesNonEmptyDirectoryMarker(t *testing.T) {
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: filer.MsgFailDelNonEmptyFolder + ": /buckets/test/photos",
},
lookupResp: &filer_pb.LookupDirectoryEntryResponse{
Entry: &filer_pb.Entry{
Name: "photos",
IsDirectory: true,
Attributes: &filer_pb.FuseAttributes{
Mime: "application/octet-stream",
Md5: []byte{1, 2, 3, 4},
FileSize: 4,
},
Content: []byte("test"),
Extended: map[string][]byte{
s3_constants.ExtETagKey: []byte("etag"),
s3_constants.ExtAmzOwnerKey: []byte("owner"),
s3_constants.AmzUserMetaPrefix + "Color": []byte("blue"),
s3_constants.AmzObjectTaggingPrefix + "k": []byte("v"),
"xattr-keep": []byte("keep-me"),
"x-seaweedfs-internal": []byte("keep-me-too"),
},
},
},
}
err := deleteObjectEntry(context.Background(), client, "/buckets/test", "photos", true, false)
require.NoError(t, err)
require.NotNil(t, client.lookupReq)
require.NotNil(t, client.updateReq)
updated := client.updateReq.Entry
require.NotNil(t, updated)
assert.False(t, updated.IsDirectoryKeyObject())
assert.Equal(t, "", updated.Attributes.Mime)
assert.Empty(t, updated.Attributes.Md5)
assert.Zero(t, updated.Attributes.FileSize)
assert.Nil(t, updated.Content)
assert.Nil(t, updated.Chunks)
assert.Equal(t, map[string][]byte{
"xattr-keep": []byte("keep-me"),
"x-seaweedfs-internal": []byte("keep-me-too"),
}, updated.Extended)
}
// A prefix object is demoted the same way, and the mark has to go with the data:
// the path is a plain directory again, and no longer a key of its own.
func TestDeleteObjectEntryDemotesPrefixObject(t *testing.T) {
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: filer.MsgFailDelNonEmptyFolder + ": /buckets/test/photos",
},
lookupResp: &filer_pb.LookupDirectoryEntryResponse{
Entry: &filer_pb.Entry{
Name: "photos",
IsDirectory: true,
Attributes: &filer_pb.FuseAttributes{},
Extended: map[string][]byte{
s3_constants.SeaweedFSPrefixObject: []byte("true"),
s3_constants.ExtETagKey: []byte("etag"),
},
},
},
}
require.NoError(t, deleteObjectEntry(context.Background(), client, "/buckets/test", "photos", true, false))
require.NotNil(t, client.updateReq)
updated := client.updateReq.Entry
require.NotNil(t, updated)
assert.False(t, updated.IsPrefixObject())
assert.False(t, updated.IsDirectoryKeyObject())
assert.Empty(t, updated.Extended)
}
func TestDeleteObjectEntryTreatsImplicitDirectoryAsSuccessfulNoop(t *testing.T) {
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: filer.MsgFailDelNonEmptyFolder + ": /buckets/test/photos",
},
lookupResp: &filer_pb.LookupDirectoryEntryResponse{
Entry: &filer_pb.Entry{
Name: "photos",
IsDirectory: true,
Attributes: &filer_pb.FuseAttributes{},
},
},
}
err := deleteObjectEntry(context.Background(), client, "/buckets/test", "photos", true, false)
require.NoError(t, err)
require.NotNil(t, client.lookupReq)
assert.Nil(t, client.updateReq)
}
func TestDeleteObjectEntryIgnoresConcurrentUpdateNotFound(t *testing.T) {
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: filer.MsgFailDelNonEmptyFolder + ": /buckets/test/photos",
},
lookupResp: &filer_pb.LookupDirectoryEntryResponse{
Entry: &filer_pb.Entry{
Name: "photos",
IsDirectory: true,
Attributes: &filer_pb.FuseAttributes{
Mime: "application/octet-stream",
},
},
},
updateErr: status.Error(codes.NotFound, "already removed"),
}
err := deleteObjectEntry(context.Background(), client, "/buckets/test", "photos", true, false)
require.NoError(t, err)
require.NotNil(t, client.lookupReq)
require.NotNil(t, client.updateReq)
}
// The key is the client's and the filer echoes it in the message it sends back,
// so a key named after the marker must not turn a real failure into the demote
// no-op, which would answer a failed delete with a 204.
func TestDeleteObjectEntryIgnoresMarkerSpoofedByKey(t *testing.T) {
name := filer.MsgFailDelNonEmptyFolder
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: "delete file /buckets/test/" + name + ": filer store delete: disk full",
},
}
err := deleteObjectEntry(context.Background(), client, "/buckets/test", name, true, false)
require.Error(t, err)
assert.Contains(t, err.Error(), "disk full")
assert.Nil(t, client.lookupReq)
assert.Nil(t, client.updateReq)
}
func TestDeleteObjectEntryPropagatesNonDirectoryDeleteErrors(t *testing.T) {
client := &deleteObjectEntryTestClient{
deleteErr: errors.New("boom"),
}
err := deleteObjectEntry(context.Background(), client, "/buckets/test", "photos", true, false)
require.Error(t, err)
assert.Contains(t, err.Error(), "boom")
assert.Nil(t, client.lookupReq)
assert.Nil(t, client.updateReq)
}
// A delete is idempotent at the filer, so a reply the transport dropped is
// reissued rather than surfaced. Issue #7204 saw the surfaced version as a 500
// on the bucket delete, which boto3 resent and then answered NoSuchBucket;
// issue #7224 saw it as a per-key InternalError inside a 200, which no SDK
// retries, so the object silently stayed.
func TestDeleteUnversionedObjectReplaysADroppedReply(t *testing.T) {
s3a := &S3ApiServer{option: &S3ApiServerOption{BucketsPath: "/buckets"}}
client := &deleteObjectEntryTestClient{
deleteCallErrs: []error{status.Error(codes.Unavailable, "transport is closing"), nil},
}
err := s3a.deleteUnversionedObjectWithClient(testRetryCtx(10*time.Millisecond), client, "mybucket", "a/b/c.txt", false)
require.NoError(t, err)
assert.Equal(t, 2, client.deleteAttempts)
// the replay must ask for the same delete, not a differently shaped one
require.Len(t, client.deleteReqs, 2)
assert.Equal(t, client.deleteReqs[0].String(), client.deleteReqs[1].String())
}
// The filer formats the deleted path, and for a recursive delete the child it
// stopped on, into the message it sends back, so both are the client's text.
// Nothing here reads that text, and this pins it: the same failure under a key
// named after a transport condition is replayed exactly as often as under a
// plain one.
func TestDeleteUnversionedObjectReplayIgnoresTheObjectName(t *testing.T) {
attemptsFor := func(key string) int {
s3a := &S3ApiServer{option: &S3ApiServerOption{BucketsPath: "/buckets"}}
client := &deleteObjectEntryTestClient{
deleteRespErrs: []string{"delete file /buckets/b/" + key + ": permission denied", ""},
}
require.NoError(t, s3a.deleteUnversionedObjectWithClient(testRetryCtx(10*time.Millisecond), client, "b", key, false))
return client.deleteAttempts
}
plain := attemptsFor("plain.log")
for _, key := range []string{"transport.log", "unavailable.txt", "slowdown.csv", filer_pb.ErrNotFound.Error()} {
assert.Equal(t, plain, attemptsFor(key), "key %q must not steer the replay", key)
}
}
// A folder the filer refused because it still has children is an answer about
// the tree, not a hiccup: it is handed straight to the demote path instead of
// being replayed until the allowance runs out.
func TestDeleteUnversionedObjectDoesNotReplayANonEmptyFolder(t *testing.T) {
s3a := &S3ApiServer{option: &S3ApiServerOption{BucketsPath: "/buckets"}}
client := &deleteObjectEntryTestClient{
deleteResp: &filer_pb.DeleteEntryResponse{
Error: filer.MsgFailDelNonEmptyFolder + ": /buckets/b/photos",
},
}
require.NoError(t, s3a.deleteUnversionedObjectWithClient(context.Background(), client, "b", "photos", false))
assert.Equal(t, 1, client.deleteAttempts)
assert.NotNil(t, client.lookupReq, "the demote must still run")
}
// A caller that has gone away is not worth replaying for.
func TestDeleteUnversionedObjectDoesNotReplayForACancelledCaller(t *testing.T) {
s3a := &S3ApiServer{option: &S3ApiServerOption{BucketsPath: "/buckets"}}
client := &deleteObjectEntryTestClient{}
ctx, cancel := context.WithCancel(context.Background())
cancel()
err := s3a.deleteUnversionedObjectWithClient(ctx, client, "b", "k", false)
require.Error(t, err)
assert.Equal(t, 1, client.deleteAttempts)
}
// The client picks how many keys a multi-object delete carries, so the backoff
// is drawn from one allowance held by the request rather than one per key.
func TestDeleteUnversionedObjectDrawsFromTheRequestAllowance(t *testing.T) {
s3a := &S3ApiServer{option: &S3ApiServerOption{BucketsPath: "/buckets"}}
ctx := testRetryCtx(20 * time.Millisecond)
start := time.Now()
var attempts int
for i := 0; i < 50; i++ {
client := &deleteObjectEntryTestClient{deleteErr: status.Error(codes.Unavailable, "transport is closing")}
require.Error(t, s3a.deleteUnversionedObjectWithClient(ctx, client, "b", fmt.Sprintf("k%d", i), false))
attempts += client.deleteAttempts
}
assert.Less(t, time.Since(start), time.Second, "one allowance, not one per key")
assert.Greater(t, attempts, 50, "the allowance must still buy a replay")
}
// testRetryCtx caps the backoff so a test exercises the replay without waiting
// out the real per-request allowance.
func testRetryCtx(allowance time.Duration) context.Context {
return withFilerRetryBudget(context.Background(), allowance)
}
func TestIsRetryableFilerErr(t *testing.T) {
assert.False(t, isRetryableFilerErr(nil))
assert.True(t, isRetryableFilerErr(status.Error(codes.Unavailable, "transport is closing")))
assert.True(t, isRetryableFilerErr(fmt.Errorf("delete entry /b/k: %w", status.Error(codes.Unavailable, "x"))))
// the filer's own message carries no status, and a store error may well clear
assert.True(t, isRetryableFilerErr(filer.DeleteEntryError("delete file /b/k: filer store delete: disk full")))
assert.False(t, isRetryableFilerErr(filer.DeleteEntryError(filer.MsgFailDelNonEmptyFolder+": /b/photos")))
assert.False(t, isRetryableFilerErr(filer_pb.ErrNotFound))
assert.False(t, isRetryableFilerErr(status.Error(codes.NotFound, "gone")))
assert.False(t, isRetryableFilerErr(context.Canceled))
assert.False(t, isRetryableFilerErr(status.Error(codes.Canceled, "context canceled")))
assert.False(t, isRetryableFilerErr(status.Error(codes.DeadlineExceeded, "context deadline exceeded")))
// the message is never read, so the same code classifies the same way
// whatever the client named the key
assert.Equal(t,
isRetryableFilerErr(status.Error(codes.PermissionDenied, "denied for /b/plain.log")),
isRetryableFilerErr(status.Error(codes.PermissionDenied, "denied for /b/transport.log")))
}