Replace eager Dictionary<ulong, StoredMessage> with lightweight
Dictionary<ulong, MessageMeta> to eliminate ~200B StoredMessage
allocation per message on the write path.
- Add MessageMeta struct (BlockId, Subject, PayloadLength, HeaderLength,
TimestampNs) — ~40B vs ~200B for StoredMessage
- Add MaterializeMessage(seq) for on-demand reconstruction from blocks
- Update all ~60 _messages references to use _meta
- Methods needing full payload (LoadAsync, ListAsync, etc.) call
MaterializeMessage; metadata-only paths use _meta directly
- Fix MsgBlock.WriteAt to clear stale delete markers on re-write
- Add cached state properties (LastSeq, MessageCount, TotalBytes, FirstSeq)
to IStreamStore/FileStore/MemStore — eliminates GetStateAsync on publish path
- Add Capture(StreamHandle, ...) overload to StreamManager — eliminates
double FindBySubject lookup (once in JetStreamPublisher, once in Capture)
- Remove _messageIndexes dictionary from FileStore write path — all lookups
now use _messages directly, saving ~48B allocation per message
- Add JetStreamPubAckFormatter for hand-rolled UTF-8 success ack formatting —
avoids JsonSerializer overhead on the hot publish path
- Switch flush loop to exponential backoff (1→2→4→8ms) matching Go server
Replace per-message DeliverMessage/flush in DeliverPullFetchMessagesAsync
with SendMessageNoFlush + batch flush every 64 messages. Add signal-based
wakeup (StreamHandle.NotifyPublish/WaitForPublishAsync) to replace 5ms
Task.Delay polling in both DeliverPullFetchMessagesAsync and
PullConsumerEngine.WaitForMessageAsync. Publishers signal waiting
consumers immediately after store append.
Implement Go's pcd (per-client deferred flush) pattern to reduce write-loop
wakeups during fan-out delivery, optimize ack reply string construction with
stack-based formatting, cache CompiledFilter on ConsumerHandle, and pool
fetch message lists. Durable consumer fetch improves from 0.60x to 0.74x Go.
Pub/sub 1:1 (16B) improved from 0.18x to 0.50x, fan-out from 0.18x to 0.44x,
and JetStream durable fetch from 0.13x to 0.64x vs Go. Key changes: replace
.ToArray() copy in SendMessage with pooled buffer handoff, batch multiple small
writes into single WriteAsync via 64KB coalesce buffer in write loop, and remove
profiling Stopwatch instrumentation from ProcessMessage/StreamManager hot paths.
Implement Go-parity background flush loop (coalesce 16KB/8ms) in MsgBlock/FileStore,
replace O(n) GetStateAsync with incremental counters, skip PruneExpired/LoadAsync/
PrunePerSubject when not needed, and bypass RAFT for single-replica streams. Fix counter
tracking bugs in RemoveMsg/EraseMsg/TTL expiry and ObjectDisposedException races in
flush loop disposal. FileStore optimizations verified with 3112/3112 JetStream tests
passing; async publish benchmark remains at ~174 msg/s due to E2E protocol path bottleneck.
Add JetStream stream/consumer config and data replication across cluster
peers via $JS.INTERNAL.* subjects with BroadcastRoutedMessageAsync (sends
to all peers, bypassing pool routing). Capture routed data messages into
local JetStream stores in DeliverRemoteMessage. Fix leaf node solicited
reconnect by re-launching the retry loop in WatchConnectionAsync after
disconnect.
Unskips 4 of 5 E2E cluster tests (LeaderDies_NewLeaderElected,
R3Stream_NodeDies_PublishContinues, Consumer_NodeDies_PullContinuesOnSurvivor,
Leaf_HubRestart_LeafReconnects). The 5th (LeaderRestart_RejoinsAsFollower)
requires RAFT log catchup which is a separate feature.
Root cause: StreamManager.CreateStore() used a hardcoded temp path for
FileStore instead of the configured store_dir from JetStream config.
This caused stream data to accumulate across test runs in a shared
directory, producing wrong message counts (e.g., expected 5 but got 80).
Server fix:
- Pass storeDir from JetStream config through to StreamManager
- CreateStore() now uses the configured store_dir for FileStore paths
Test fixes for tests that now pass (3):
- R3Stream_CreateAndPublish_ReplicatedAcrossNodes: delete stream before
test, verify only on publishing node (no cross-node replication yet)
- R3Stream_Purge_ReplicatedAcrossNodes: same pattern
- LogReplication_AllReplicasHaveData: same pattern
Tests skipped pending RAFT implementation (5):
- LeaderDies_NewLeaderElected: requires RAFT leader re-election
- LeaderRestart_RejoinsAsFollower: requires RAFT log catchup
- R3Stream_NodeDies_PublishContinues: requires cross-node replication
- Consumer_NodeDies_PullContinuesOnSurvivor: requires replicated state
- Leaf_HubRestart_LeafReconnects: leaf reconnection after hub restart
Adds 8 new E2E tests to JetStreamTests.cs (tests 11-18) covering push
consumer config, AckNone/AckAll policies, Interest/WorkQueue retention,
ordered consumers, mirror streams, and source streams.
Fixes three server gaps exposed by the new tests: mirror JSON parsing
(deliver_subject and mirror object fields were silently ignored in stream
and consumer API handlers), and deliver_subject omitted from consumer
info wire format. Also fixes ShutdownDrainTests to use TaskCompletionSource
on ConnectionDisconnected instead of a Task.Delay poll loop.
Add GetMirrorInfo/GetSourceInfo methods to MirrorCoordinator and
SourceCoordinator returning structured MirrorInfoResponse/SourceInfoResponse
records (Name, Lag, Active ms, FilterSubject, Error). Wire both into
StreamManager.GetMirrorInfo and GetSourceInfos for the monitoring API path.
Add ValidateConfigUpdate to StreamManager with immutability rules for storage type,
mirror, sources, and retention policy; sealed stream guard; MaxConsumers decrease
prevention; even-replica rejection; and subject overlap detection against peer streams.
Wire the check into CreateOrUpdate for all update paths. 12 new tests in
ConfigUpdateValidationTests.cs cover all rules including the StreamManager integration test.
Enhance StreamSnapshotService with CreateTarSnapshotAsync / RestoreTarSnapshotAsync
methods that produce a Snappy-compressed TAR archive (stream.json + messages/*.json).
Add CreateTarSnapshotWithDeadlineAsync for deadline-bounded snapshots, and a
SnapshotRestoreResult record carrying stats. 10 new unit tests in
JetStream/Snapshots/StreamSnapshotTests.cs exercise the full create/restore
round-trip, compression format, empty-stream edge case, and deadline enforcement.
Adds ProposeWaitingRequest, RegisterClusterPending, RemoveClusterPending,
GetClusterPendingRequests, and ClusterPendingCount to PullConsumerEngine,
backed by a ConcurrentDictionary keyed by reply subject. Includes 10 xUnit
tests covering quorum checks, pending tracking, and concurrent access patterns.
Add ResetToSequence to ConsumerManager that updates NextSequence,
clears AckProcessor state via new ClearAll(), and zeroes PendingBytes.
Add AckProcessor.SetAckFloor() that prunes pending entries below the
new floor. Go reference: consumer.go:4241 processResetReq.
Implements DeliveryInterestTracker (consumer.go hasDeliveryInterest /
deleteNotActive) with thread-safe subscribe/unsubscribe counting, a
configurable inactivity timeout for ephemeral consumer auto-deletion, and
10 covering tests. Also adds SlopwatchSuppress to a pre-existing
Task.Delay in MaxDeliveriesTests.cs that was already testing time-based expiry.
Add FilterSkipTracker using SubjectMatch.MatchLiteral() for NATS token-based
filter matching with sorted-set skip sequence gap tracking. Includes 14 tests
covering exact match, star/gt wildcards, multi-filter, counters, RecordSkip,
NextUnskippedSequence, PurgeBelow, and Reset.
Adds MaxDeliver property, exceeded sequence tracking, DrainExceeded,
DeliveryExceededPolicy enum, and ScheduleRedelivery enforcement to
AckProcessor; 10 new tests in MaxDeliveriesTests.cs.
Heartbeat frames now include Nats-Pending-Messages and Nats-Pending-Bytes
headers populated from the ConsumerHandle. Flow control frames increment
FlowControlPendingCount; AcknowledgeFlowControl() decrements it. IsFlowControlStalled
returns true when pending count reaches MaxFlowControlPending (2).
Go reference: consumer.go:5222 (sendIdleHeartbeat), consumer.go:5495 (sendFlowControl).
Enhances HandlePause to parse pause_until (RFC3339) for time-bounded
pauses and returns current pause state in the response body. Adds
Paused/PauseUntil properties and PauseResponse factory to
JetStreamApiResponse. Covers 10 new parity tests.
Add JetStream advisory subject constants to EventSubjects and a lightweight
AdvisoryPublisher that publishes stream/consumer lifecycle events to
$JS.EVENT.ADVISORY.* subjects without depending on InternalEventSystem
directly (testable via Action delegate injection).
Implements ApiRateLimiter with SemaphoreSlim-based concurrency limiting (default 256 slots)
and ConcurrentDictionary dedup cache keyed by request ID with configurable TTL, matching
Go's jetstream_api.go maxConcurrentRequests semaphore and dedup window. Also adds
ClusteredRequestProcessor for correlating pending RAFT proposals with waiting callers via
TaskCompletionSource, and SlopwatchSuppressAttribute as a marker for intentional timing-based
tests. 12 ApiRateLimiter tests + 13 ClusteredRequestProcessor tests all pass.
Implement HandleClusteredCreateAsync, HandleClusteredUpdateAsync, and
HandleClusteredDeleteAsync on StreamApiHandlers, and HandleClusteredCreateAsync
and HandleClusteredDeleteAsync on ConsumerApiHandlers. These handlers propose
operations to the meta RAFT group (JetStreamMetaGroup) instead of operating on
the local StreamManager/ConsumerManager, matching the Go jsClusteredStreamRequest
and jsClusteredConsumerRequest patterns (jetstream_cluster.go:7620-8265).
Ten tests in ClusteredApiTests.cs verify: stream create proposes to meta group,
duplicate-stream error, not-leader error (code 10003), stream update, stream
delete, not-found-on-delete, consumer create on stream, consumer-on-missing-stream
error, consumer delete, and not-found consumer delete.
Add ILeaderForwarder interface and DefaultLeaderForwarder, update
JetStreamApiRouter with RouteAsync + ForwardedCount so non-leader nodes
can attempt to forward mutating requests to the meta-group leader before
falling back to a NotLeader error response.
Implements AssignmentCodec with JSON serialization for StreamAssignment
and ConsumerAssignment, plus Snappy compression helpers for large payloads.
Adds 12 tests covering round-trips, error handling, compression, and a
golden fixture for format stability.
Add Version property to StreamAssignment and ConsumerAssignment so future-version
entries from newer cluster peers are gracefully skipped rather than applied incorrectly.
ProcessStreamAssignment and ProcessConsumerAssignment now reject entries with
Version > CurrentVersion (1), incrementing SkippedUnsupportedEntries for operator
visibility. Add MetaEntryType.Unknown with a no-op ApplyEntry handler so unrecognised
RAFT entry types never crash the cluster. Version 0 is treated as current for
backward compatibility with pre-versioned assignments.
Implement IsMember, SetPreferred, RemovePeer, AddPeer, CreateRaftGroup
factory, IsUnderReplicated, and IsOverReplicated on RaftGroup. Add 22
RaftGroupLifecycleTests covering all helpers and quorum size calculation.
Extends PlacementEngine.SelectPeerGroup with three new capabilities ported
from jetstream_cluster.go:7212 selectPeerGroup:
- JetStreamUniqueTag enforcement: PlacementPolicy.UniqueTag (e.g. "az")
ensures no two replicas share the same value for a tag with that prefix,
matching Go's uniqueTagPrefix / checkUniqueTag logic.
- MaxAssetsPerPeer HA limit: peers at or above their asset ceiling are
deprioritised (moved to fallback), not hard-excluded, so selection still
succeeds when no preferred peers remain.
- Weighted scoring: candidates sorted by
score = AvailableStorage - (CurrentAssets * assetCostWeight)
(DefaultAssetCostWeight = 1 GiB) replacing the raw-storage sort, with a
custom weight parameter for testing.
10 new tests in TopologyPlacementTests.cs cover all three features and their
edge cases. All 30 PlacementEngine tests continue to pass.
Add ProcessAddPeer/ProcessRemovePeer to JetStreamMetaGroup for
peer-driven stream reassignment. Includes AddKnownPeer/RemoveKnownPeer
tracked in a HashSet, RemovePeerFromStream, RemapStreamAssignment with
replacement-peer selection from the known pool, and DesiredReplicas on
RaftGroup for under-replication detection. Go ref: jetstream_cluster.go:2290-2439.
Extend ApplyEntry in JetStreamMetaGroup with PeerAdd/PeerRemove dispatch
to the existing Task 12 peer management methods. Add StreamMsgOp (Store,
Remove, Purge) and ConsumerOp (Ack, Nak, Deliver, Term, Progress) enums
plus ApplyStreamMsgOp and ApplyConsumerEntry methods to StreamReplicaGroup.
Extend ApplyCommittedEntriesAsync to parse smsg:/centry: command prefixes
and route to the new apply methods. Add MetaEntryType.PeerAdd/PeerRemove
enum values. 35 new tests in EntryApplicationTests.cs, all passing.
Replace HashSet<ulong> _deleted in MsgBlock with SequenceSet — a sorted-range
list that compresses contiguous deletions into (Start, End) intervals. Adds
O(log n) Contains/Add via binary search on range count, matching Go's avl.SequenceSet
semantics with a simpler implementation.
- Add SequenceSet.cs: sorted-range compressed set with Add/Remove/Contains/Count/Clear
and IEnumerable<ulong> in ascending order. Binary search for all O(log n) ops.
- Replace HashSet<ulong> _deleted and _skipSequences in MsgBlock with SequenceSet.
- Add secureErase parameter (default false) to MsgBlock.Delete(): when true, payload
bytes are overwritten with RandomNumberGenerator.Fill() before the delete record is
written, making original content unrecoverable on disk.
- Update FileStore.DeleteInBlock() to propagate secureErase flag.
- Update FileStore.EraseMsg() to use secureErase: true via block layer instead of
delegating to RemoveMsg().
- Add SequenceSetTests.cs: 25 tests covering Add, Remove, Contains, Count, range
compression, gap filling, bridge merges, enumeration, boundary values, round-trip.
- Add FileStoreTombstoneTrackingTests.cs: 12 tests covering SequenceSet tracking in
MsgBlock, tombstone persistence through RebuildIndex recovery, secure erase
payload overwrite verification, and FileStore.EraseMsg integration.
Go reference: filestore.go:5267 (removeMsg), filestore.go:5890 (eraseMsg),
avl/seqset.go (SequenceSet).
- Add AtomicFileWriter static helper: writes to {path}.{random}.tmp, flushes,
then File.Move(overwrite:true) — concurrent-safe via unique temp path per call
- Add _stateWriteLock (SemaphoreSlim 1,1) to FileStore; dispose in both Dispose
and DisposeAsync paths
- Promote WriteStreamState to async WriteStreamStateAsync using AtomicFileWriter
under the write lock; FlushAllPending now returns Task
- Update IStreamStore.FlushAllPending signature to Task; fix MemStore no-op impl
- Fix FileStoreCrashRecoveryTests to await FlushAllPending (3 sync→async tests)
- Add 9 AtomicFileWriterTests covering create, no-tmp-remains, overwrite,
concurrent safety, memory overload, empty data, and large payload
Add _lastChecksum field and LastChecksum property to MsgBlock tracking
the XxHash64 checksum of the last written record (Go: msgBlock.lchk,
filestore.go:2204). Capture the checksum from the encoded record trailer
on every Write/WriteAt/WriteSkip call. Read-path validation happens
naturally through the existing MessageRecord.Decode checksum check.
Exposes public RecordFailure/RecordSuccess/GetRetryDelay (no-jitter, deterministic)
on MirrorCoordinator, plus RecordSourceSeq with HasGap/GapStart/GapEnd properties
and SetError/ClearError/HasError/ErrorMessage for error state. Makes IsDuplicate
and RecordMsgId public on SourceCoordinator and adds PruneDedupWindow(DateTimeOffset)
for explicit-cutoff dedup window pruning. Adds 5 unit tests in MirrorSourceRetryTests.
Implements Interest retention policy logic that tracks which consumers
are interested in each message subject and whether they have acknowledged
delivery, retaining messages until all interested consumers have acked.
Go reference: stream.go checkInterestState/noInterest.
Add AssignPinId, ValidatePinId, and UnassignPinId to PriorityGroupManager,
plus CurrentPinId tracking on PriorityGroup, porting Go consumer.go
(setPinnedTimer, assignNewPinId) pin ID semantics. Covered by 7 new tests.
Adds PauseUntilUtc to ConsumerHandle, a new Pause(DateTime) overload,
Resume, IsPaused, and GetPauseUntil to ConsumerManager. A System.Threading.Timer
fires when the deadline passes and calls AutoResume, raising OnAutoResumed so
tests can synchronise via SemaphoreSlim instead of Task.Delay. ConsumerManager
now implements IDisposable to clean up outstanding timers. Timer is also
cancelled on explicit Resume and Delete.
Go reference: consumer.go (pauseConsumer / resumeConsumer / isPaused).
Implements a FIFO pull-request queue (WaitingRequestQueue) with per-request
mutable batch countdown and byte budget tracking, plus RemoveExpired cleanup.
Complements the existing priority-based PullRequestWaitQueue for pull consumer delivery.
Go reference: consumer.go waitQueue / processNextMsgRequest (lines 4276-4450).
Add RedeliveryTracker constructor overload, Register(ulong, string) for
tracker-based ack-wait, ProcessAck(ulong) payload-free overload, GetDeadline,
CanRegister for maxAckPending enforcement, ParseAckType static parser, and
AckType enum. All existing API signatures are preserved; 9 new tests added in
AckProcessorEnhancedTests.cs with no regressions to existing 10 tests.
Add new constructor, Schedule(DateTimeOffset), GetDue(DateTimeOffset),
IncrementDeliveryCount, IsMaxDeliveries(), and GetBackoffDelay() to
RedeliveryTracker without breaking existing API. Uses PriorityQueue<ulong,
DateTimeOffset> for deadline-ordered dispatch mirroring Go consumer.go rdq.
Add ProcessLeaderChange(bool) method and OnLeaderChange event to JetStreamMetaGroup.
Refactor StepDown() to delegate inflight clearing through ProcessLeaderChange,
enabling subscribers to react to leadership transitions.
Go reference: jetstream_cluster.go:7001-7074 processLeaderChange.
Joseph Doherty