Repository navigation
Fix lost Redis appends during Raft leader handoff #1286
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -1514,7 +1514,7 @@ func (f *kvFSM) handleOnePhaseTxnRequest(ctx context.Context, r *pb.Request, com | |
| // applying this log entry. The retention-window > max-retry-latency | ||
| // invariant prevents the rare case where a real never-landed retry | ||
| // arrives with PrevCommitTS below pebble's compacted floor. | ||
| dedup, err := f.dedupProbeOnePhase(ctx, meta, r.ReadKeys, muts) | ||
| if err != nil { | ||
| return err | ||
| } | ||
|
|
@@ -1531,7 +1531,7 @@ func (f *kvFSM) handleOnePhaseTxnRequest(ctx context.Context, r *pb.Request, com | |
| if err != nil { | ||
| return err | ||
| } | ||
| if err := f.store.ApplyMutationsRaftAtFenced(ctx, storeMuts, r.ReadKeys, startTS, commitTS, f.pendingApplyIdx); err != nil { | ||
| return errors.WithStack(err) | ||
| } | ||
| f.notifyApplyObservers(commitTS, uniq) | ||
|
|
@@ -1630,7 +1630,7 @@ func (f *kvFSM) uniqueTxnMutationsAboveFloorWithSnapshot(muts []*pb.Mutation, co | |
| // Returns (true, nil) → the entry must no-op (prior attempt landed). | ||
| // Returns (false, nil) → fall through to normal apply. | ||
| // Returns (false, err) → propagate err; apply must not proceed. | ||
| func (f *kvFSM) dedupProbeOnePhase(ctx context.Context, meta TxnMeta, readKeys [][]byte, muts []*pb.Mutation) (bool, error) { | ||
| if meta.PrevCommitTS == 0 { | ||
| return false, nil | ||
| } | ||
|
|
@@ -1639,20 +1639,56 @@ func (f *kvFSM) dedupProbeOnePhase(ctx context.Context, meta TxnMeta, readKeys [ | |
| return false, errors.WithStack(err) | ||
| } | ||
| if landed { | ||
| return f.priorMutationValuesMatch(ctx, muts, readKeys, meta.PrevCommitTS) | ||
| } | ||
| for _, stagedKey := range stagedVisibilityKeysForReadAlias(readKeys, meta.PrimaryKey, meta.PrimaryKey) { | ||
| landed, err = f.store.CommittedVersionAt(ctx, stagedKey, meta.PrevCommitTS) | ||
| if err != nil { | ||
| return false, errors.WithStack(err) | ||
| } | ||
| if landed { | ||
| return f.priorMutationValuesMatch(ctx, muts, readKeys, meta.PrevCommitTS) | ||
| } | ||
| } | ||
| return false, nil | ||
| } | ||
|
|
||
| // A timestamp may be issued by two leaders during a handoff. An exact | ||
| // version at PrevCommitTS alone cannot identify the retry, even for scalar | ||
| // writes. Compare every final PUT value in the retained write set. | ||
| func (f *kvFSM) priorMutationValuesMatch(ctx context.Context, muts []*pb.Mutation, readKeys [][]byte, prevTS uint64) (bool, error) { | ||
| uniq, err := uniqueMutations(muts) | ||
| if err != nil { | ||
| return false, err | ||
| } | ||
| for _, mut := range uniq { | ||
| if mut.Op != pb.Op_PUT { | ||
| continue | ||
| } | ||
| keys := append([][]byte{mut.Key}, stagedVisibilityKeysForReadAlias(readKeys, mut.Key, mut.Key)...) | ||
| matched := false | ||
| for _, key := range keys { | ||
| landed, err := f.store.CommittedVersionAt(ctx, key, prevTS) | ||
| if err != nil { | ||
| return false, errors.WithStack(err) | ||
| } | ||
| if !landed { | ||
| continue | ||
| } | ||
| value, err := f.store.GetAt(ctx, key, prevTS) | ||
| if err != nil { | ||
| return false, errors.WithStack(err) | ||
|
Comment on lines
+1678
to
+1680
There was a problem hiding this comment.
When Useful? React with 👍 / 👎.
Comment on lines
+1678
to
+1680
There was a problem hiding this comment.
When a PUT retry's AGENTS.md reference: AGENTS.md:L33-L35 Useful? React with 👍 / 👎. |
||
| } | ||
| matched = bytes.Equal(value, mut.Value) | ||
| break | ||
|
Comment on lines
+1682
to
+1683
There was a problem hiding this comment.
When staged visibility is active and both the live key and a staged alias have versions at Useful? React with 👍 / 👎. |
||
| } | ||
| if !matched { | ||
| return false, nil | ||
| } | ||
| } | ||
| return true, nil | ||
| } | ||
|
|
||
| func stagedVisibilityJobIDsForReadAlias(readKeys [][]byte, aliasRawKey []byte) []uint64 { | ||
| if len(readKeys) == 0 { | ||
| return nil | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,74 @@ | ||
| package kv | ||
|
|
||
| import ( | ||
| "bytes" | ||
| "context" | ||
| "path/filepath" | ||
| "testing" | ||
|
|
||
| "github.com/bootjp/elastickv/internal/encryption" | ||
| pb "github.com/bootjp/elastickv/proto" | ||
| "github.com/bootjp/elastickv/store" | ||
| "github.com/stretchr/testify/require" | ||
| "google.golang.org/protobuf/proto" | ||
| ) | ||
|
|
||
| func TestDynamoAppendDistinctRaftEntryCannotAcknowledgeTimestampCollision(t *testing.T) { | ||
| for _, encrypted := range []bool{false, true} { | ||
| t.Run(map[bool]string{false: "plaintext", true: "encrypted"}[encrypted], func(t *testing.T) { | ||
| ctx := context.Background() | ||
| var opts []store.PebbleStoreOption | ||
| if encrypted { | ||
| keys := encryption.NewKeystore() | ||
| require.NoError(t, keys.Set(1, bytes.Repeat([]byte{0x37}, encryption.KeySize))) | ||
| cipher, err := encryption.NewCipher(keys) | ||
| require.NoError(t, err) | ||
| opts = append(opts, store.WithEncryption(cipher, store.NewCounterNonceFactory(1, 1), func() (uint32, bool) { return 1, true })) | ||
| } | ||
| st, err := store.NewPebbleStore(filepath.Join(t.TempDir(), "fsm.db"), opts...) | ||
| require.NoError(t, err) | ||
| t.Cleanup(func() { require.NoError(t, st.Close()) }) | ||
| fsm, ok := NewKvFSMWithHLC(st, NewHLC()).(*kvFSM) | ||
| require.True(t, ok) | ||
| key := []byte(DynamoItemPrefix + "amVwc2VuX2FwcGVuZA|2|16\x00\x01") | ||
| // Minimized from the same committed WAL that produced the missing | ||
| // acknowledged append37/key16: distinct entries371/372 in term5 | ||
| // carried different item values at this identical commit timestamp. | ||
| const collisionTS uint64 = 117485124856709124 | ||
| first := dynamoCollisionItem(t, "43") | ||
| lost := dynamoCollisionItem(t, "37") | ||
| fsm.SetApplyIndex(371) | ||
| require.NoError(t, applyFSMRequest(t, fsm, onePhaseReq(collisionTS-1, collisionTS, 0, key, first))) | ||
| // An actual replay of entry371 remains idempotent. | ||
| require.NoError(t, applyFSMRequest(t, fsm, onePhaseReq(collisionTS-1, collisionTS, 0, key, first))) | ||
| fsm.SetApplyIndex(372) | ||
| err = applyFSMRequest(t, fsm, onePhaseReq(collisionTS-2, collisionTS, 0, key, lost)) | ||
| require.ErrorIs(t, err, store.ErrWriteConflict, "another entry must not return success without writing its value") | ||
| got, err := st.GetAt(ctx, key, collisionTS) | ||
| require.NoError(t, err) | ||
| require.Equal(t, first, got) | ||
| // Re-read/recompute at a fresh timestamp preserves both appends. | ||
| both := dynamoCollisionItem(t, "43", "37") | ||
| fsm.SetApplyIndex(373) | ||
| require.NoError(t, applyFSMRequest(t, fsm, onePhaseReq(collisionTS, collisionTS+1, 0, key, both))) | ||
| got, err = st.GetAt(ctx, key, collisionTS+1) | ||
| require.NoError(t, err) | ||
| require.Equal(t, both, got) | ||
| }) | ||
| } | ||
| } | ||
|
|
||
| func dynamoCollisionItem(t *testing.T, values ...string) []byte { | ||
| t.Helper() | ||
| list := &pb.DynamoAttributeValueList{} | ||
| for _, value := range values { | ||
| list.Values = append(list.Values, &pb.DynamoAttributeValue{Value: &pb.DynamoAttributeValue_N{N: value}}) | ||
| } | ||
| item := &pb.DynamoItem{Attributes: map[string]*pb.DynamoAttributeValue{ | ||
| "pk": {Value: &pb.DynamoAttributeValue_S{S: "16"}}, | ||
| "val": {Value: &pb.DynamoAttributeValue_L{L: list}}, | ||
| }} | ||
| body, err := (proto.MarshalOptions{Deterministic: true}).Marshal(item) | ||
| require.NoError(t, err) | ||
| return append([]byte{0, 'D', 'I', 1}, body...) | ||
| } |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,75 @@ | ||
| package store | ||
|
|
||
| import ( | ||
| "context" | ||
| "testing" | ||
|
|
||
| "github.com/stretchr/testify/require" | ||
| ) | ||
|
|
||
| func TestOnePhaseApplyRejectsTimestampAtAppliedWatermark(t *testing.T) { | ||
| for _, tc := range []struct { | ||
| name string | ||
| new func(*testing.T) MVCCStore | ||
| }{ | ||
| {"memory", func(*testing.T) MVCCStore { return NewMVCCStore() }}, | ||
| {"pebble", newApplyIndexPebbleStore}, | ||
| } { | ||
| t.Run(tc.name, func(t *testing.T) { | ||
| ctx := context.Background() | ||
| st := tc.new(t) | ||
| put := func(key, val string) []*KVPairMutation { | ||
| return []*KVPairMutation{{Op: OpTypePut, Key: []byte(key), Value: []byte(val)}} | ||
| } | ||
| require.NoError(t, st.ApplyMutationsRaftAtFenced(ctx, put("first", "45"), nil, 99, 100, 10)) | ||
| for _, ts := range []uint64{100, 99} { | ||
| err := st.ApplyMutationsRaftAtFenced(ctx, put("later", "44"), nil, 98, ts, 11) | ||
| require.ErrorIs(t, err, ErrWriteConflict) | ||
| } | ||
| _, err := st.GetAt(ctx, []byte("later"), 100) | ||
| require.ErrorIs(t, err, ErrKeyNotFound) | ||
| require.Equal(t, uint64(100), st.LastCommitTS()) | ||
| require.NoError(t, st.ApplyMutationsRaftAtFenced(ctx, put("later", "44"), nil, 100, 101, 11)) | ||
| }) | ||
| } | ||
| } | ||
|
|
||
| func TestOnePhaseRaftEntryRejectsCollidingAppend(t *testing.T) { | ||
| ctx := context.Background() | ||
| st := newApplyIndexPebbleStore(t) | ||
| applier := raftEntryApplier(t, st) | ||
| key := ListItemKey([]byte("jepsen-list"), 43) | ||
| put := func(value string) []*KVPairMutation { | ||
| return []*KVPairMutation{{Op: OpTypePut, Key: key, Value: []byte(value)}} | ||
| } | ||
| require.NoError(t, applier.ApplyRaftEntry(ctx, 10, func(ctx context.Context) error { | ||
| return st.ApplyMutationsRaftAtFenced(ctx, put("45"), nil, 99, 100, 10) | ||
| })) | ||
| require.NoError(t, applier.ApplyRaftEntry(ctx, 11, func(ctx context.Context) error { | ||
| err := st.ApplyMutationsRaftAtFenced(ctx, put("44"), nil, 99, 100, 11) | ||
| require.ErrorIs(t, err, ErrWriteConflict) | ||
| return nil | ||
| })) | ||
| got, err := st.GetAt(ctx, key, 100) | ||
| require.NoError(t, err) | ||
| require.Equal(t, []byte("45"), got) | ||
| } | ||
|
|
||
| func TestOnePhaseRaftEntryRejectsCollidingWriteInSameBatch(t *testing.T) { | ||
| ctx := context.Background() | ||
| st := newApplyIndexPebbleStore(t) | ||
| applier := raftEntryApplier(t, st) | ||
| key := ListItemKey([]byte("jepsen-list"), 43) | ||
| put := func(value string) []*KVPairMutation { | ||
| return []*KVPairMutation{{Op: OpTypePut, Key: key, Value: []byte(value)}} | ||
| } | ||
| require.NoError(t, applier.ApplyRaftEntry(ctx, 10, func(ctx context.Context) error { | ||
| require.NoError(t, st.ApplyMutationsRaftAtFenced(ctx, put("45"), nil, 99, 100, 10)) | ||
| err := st.ApplyMutationsRaftAtFenced(ctx, put("44"), nil, 99, 100, 10) | ||
| require.ErrorIs(t, err, ErrWriteConflict) | ||
| return nil | ||
| })) | ||
| got, err := st.GetAt(ctx, key, 100) | ||
| require.NoError(t, err) | ||
| require.Equal(t, []byte("45"), got) | ||
| } |
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
When an option-2 retry contains a delete and another transaction committed any version of its primary key at the colliding
PrevCommitTS, this loop skips the delete entirely; for a delete-only write, or when all PUTs happen to match, the function therefore returnstruewithout verifying that a tombstone exists at that timestamp. The FSM then no-ops the whole Redis EXEC or DynamoDB retry and can acknowledge a deletion that never occurred. Compare the exact operation/tombstone state for DEL mutations before declaring the prior attempt a match.Useful? React with 👍 / 👎.