feat(jetstream): add mirror sync loop and source coordination with filtering (C9+C10)
This commit is contained in:
@@ -1,22 +1,364 @@
|
||||
using System.Threading.Channels;
|
||||
using NATS.Server.JetStream.Storage;
|
||||
|
||||
namespace NATS.Server.JetStream.MirrorSource;
|
||||
|
||||
public sealed class MirrorCoordinator
|
||||
// Go reference: server/stream.go:2788-2854 (processMirrorMsgs), 3125-3400 (setupMirrorConsumer)
|
||||
// Go reference: server/stream.go:2863-3014 (processInboundMirrorMsg)
|
||||
|
||||
/// <summary>
|
||||
/// Coordinates continuous synchronization from an origin stream to a local mirror.
|
||||
/// Runs a background pull loop that fetches batches of messages from the origin,
|
||||
/// applies them to the local store, and tracks origin-to-current sequence alignment
|
||||
/// for catchup after restarts. Includes exponential backoff retry on failures
|
||||
/// and health reporting via lag calculation.
|
||||
/// </summary>
|
||||
public sealed class MirrorCoordinator : IAsyncDisposable
|
||||
{
|
||||
// Go: sourceHealthCheckInterval = 10 * time.Second
|
||||
private static readonly TimeSpan HealthCheckInterval = TimeSpan.FromSeconds(10);
|
||||
|
||||
// Go: sourceHealthHB = 1 * time.Second
|
||||
private static readonly TimeSpan HeartbeatInterval = TimeSpan.FromSeconds(1);
|
||||
|
||||
private static readonly TimeSpan InitialRetryDelay = TimeSpan.FromMilliseconds(250);
|
||||
private static readonly TimeSpan MaxRetryDelay = TimeSpan.FromSeconds(30);
|
||||
private const int DefaultBatchSize = 256;
|
||||
|
||||
private readonly IStreamStore _targetStore;
|
||||
private readonly Channel<StoredMessage> _inbound;
|
||||
private readonly Lock _gate = new();
|
||||
private CancellationTokenSource? _cts;
|
||||
private Task? _syncLoop;
|
||||
private int _consecutiveFailures;
|
||||
|
||||
/// <summary>Last sequence number successfully applied from the origin stream.</summary>
|
||||
public ulong LastOriginSequence { get; private set; }
|
||||
|
||||
/// <summary>UTC timestamp of the last successful sync operation.</summary>
|
||||
public DateTime LastSyncUtc { get; private set; }
|
||||
|
||||
/// <summary>Number of consecutive sync failures (resets on success).</summary>
|
||||
public int ConsecutiveFailures
|
||||
{
|
||||
get { lock (_gate) return _consecutiveFailures; }
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Whether the background sync loop is actively running.
|
||||
/// </summary>
|
||||
public bool IsRunning
|
||||
{
|
||||
get { lock (_gate) return _syncLoop is not null && !_syncLoop.IsCompleted; }
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Current lag: origin last sequence minus local last sequence.
|
||||
/// Returns 0 when fully caught up or when origin sequence is unknown.
|
||||
/// </summary>
|
||||
public ulong Lag { get; private set; }
|
||||
|
||||
// Go: mirror.sseq — stream sequence tracking for gap detection
|
||||
private ulong _expectedOriginSeq;
|
||||
|
||||
// Go: mirror.dseq — delivery sequence tracking
|
||||
private ulong _deliverySeq;
|
||||
|
||||
public MirrorCoordinator(IStreamStore targetStore)
|
||||
{
|
||||
_targetStore = targetStore;
|
||||
_inbound = Channel.CreateUnbounded<StoredMessage>(new UnboundedChannelOptions
|
||||
{
|
||||
SingleReader = true,
|
||||
SingleWriter = false,
|
||||
});
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Processes a single inbound message from the origin stream.
|
||||
/// This is the direct-call path used when the origin and mirror are in the same process.
|
||||
/// Go reference: server/stream.go:2863-3014 (processInboundMirrorMsg)
|
||||
/// </summary>
|
||||
public async Task OnOriginAppendAsync(StoredMessage message, CancellationToken ct)
|
||||
{
|
||||
// Go: sseq == mset.mirror.sseq+1 — normal in-order delivery
|
||||
if (_expectedOriginSeq > 0 && message.Sequence <= _expectedOriginSeq)
|
||||
{
|
||||
// Ignore older/duplicate messages (Go: sseq <= mset.mirror.sseq)
|
||||
return;
|
||||
}
|
||||
|
||||
// Go: sseq > mset.mirror.sseq+1 and dseq == mset.mirror.dseq+1 — gap in origin (deleted/expired)
|
||||
// For in-process mirrors we skip gap handling since the origin store handles its own deletions.
|
||||
|
||||
await _targetStore.AppendAsync(message.Subject, message.Payload, ct);
|
||||
_expectedOriginSeq = message.Sequence;
|
||||
_deliverySeq++;
|
||||
LastOriginSequence = message.Sequence;
|
||||
LastSyncUtc = DateTime.UtcNow;
|
||||
Lag = 0; // In-process mirror receives messages synchronously, so lag is always zero here.
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Enqueues a message for processing by the background sync loop.
|
||||
/// Used when messages arrive asynchronously (e.g., from a pull consumer on the origin).
|
||||
/// </summary>
|
||||
public bool TryEnqueue(StoredMessage message)
|
||||
{
|
||||
return _inbound.Writer.TryWrite(message);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Starts the background sync loop that drains the inbound channel and applies
|
||||
/// messages to the local store. This models Go's processMirrorMsgs goroutine.
|
||||
/// Go reference: server/stream.go:2788-2854 (processMirrorMsgs)
|
||||
/// </summary>
|
||||
public void StartSyncLoop()
|
||||
{
|
||||
lock (_gate)
|
||||
{
|
||||
if (_syncLoop is not null && !_syncLoop.IsCompleted)
|
||||
return;
|
||||
|
||||
_cts = new CancellationTokenSource();
|
||||
_syncLoop = RunSyncLoopAsync(_cts.Token);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Starts the background sync loop with a pull-based fetch from the origin store.
|
||||
/// This models Go's setupMirrorConsumer + processMirrorMsgs pattern where the mirror
|
||||
/// actively pulls batches from the origin.
|
||||
/// Go reference: server/stream.go:3125-3400 (setupMirrorConsumer)
|
||||
/// </summary>
|
||||
public void StartPullSyncLoop(IStreamStore originStore, int batchSize = DefaultBatchSize)
|
||||
{
|
||||
lock (_gate)
|
||||
{
|
||||
if (_syncLoop is not null && !_syncLoop.IsCompleted)
|
||||
return;
|
||||
|
||||
_cts = new CancellationTokenSource();
|
||||
_syncLoop = RunPullSyncLoopAsync(originStore, batchSize, _cts.Token);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Stops the background sync loop and waits for it to complete.
|
||||
/// Go reference: server/stream.go:3027-3032 (cancelMirrorConsumer)
|
||||
/// </summary>
|
||||
public async Task StopAsync()
|
||||
{
|
||||
CancellationTokenSource? cts;
|
||||
Task? loop;
|
||||
lock (_gate)
|
||||
{
|
||||
cts = _cts;
|
||||
loop = _syncLoop;
|
||||
}
|
||||
|
||||
if (cts is not null)
|
||||
{
|
||||
await cts.CancelAsync();
|
||||
if (loop is not null)
|
||||
{
|
||||
try { await loop; }
|
||||
catch (OperationCanceledException) { }
|
||||
}
|
||||
}
|
||||
|
||||
lock (_gate)
|
||||
{
|
||||
_cts?.Dispose();
|
||||
_cts = null;
|
||||
_syncLoop = null;
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Reports current health state for monitoring.
|
||||
/// Go reference: server/stream.go:2739-2743 (mirrorInfo), 2698-2736 (sourceInfo)
|
||||
/// </summary>
|
||||
public MirrorHealthReport GetHealthReport(ulong? originLastSeq = null)
|
||||
{
|
||||
var lag = originLastSeq.HasValue && originLastSeq.Value > LastOriginSequence
|
||||
? originLastSeq.Value - LastOriginSequence
|
||||
: Lag;
|
||||
|
||||
return new MirrorHealthReport
|
||||
{
|
||||
LastOriginSequence = LastOriginSequence,
|
||||
LastSyncUtc = LastSyncUtc,
|
||||
Lag = lag,
|
||||
ConsecutiveFailures = ConsecutiveFailures,
|
||||
IsRunning = IsRunning,
|
||||
IsStalled = LastSyncUtc != default
|
||||
&& DateTime.UtcNow - LastSyncUtc > HealthCheckInterval,
|
||||
};
|
||||
}
|
||||
|
||||
public async ValueTask DisposeAsync()
|
||||
{
|
||||
await StopAsync();
|
||||
_inbound.Writer.TryComplete();
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------------
|
||||
// Background sync loop: channel-based (inbound messages pushed to us)
|
||||
// Go reference: server/stream.go:2788-2854 (processMirrorMsgs main loop)
|
||||
// -------------------------------------------------------------------------
|
||||
private async Task RunSyncLoopAsync(CancellationToken ct)
|
||||
{
|
||||
// Go: t := time.NewTicker(sourceHealthCheckInterval)
|
||||
using var healthTimer = new PeriodicTimer(HealthCheckInterval);
|
||||
var reader = _inbound.Reader;
|
||||
|
||||
while (!ct.IsCancellationRequested)
|
||||
{
|
||||
try
|
||||
{
|
||||
// Go: select { case <-msgs.ch: ... case <-t.C: ... }
|
||||
// We process all available messages, then wait for more or health check.
|
||||
while (reader.TryRead(out var msg))
|
||||
{
|
||||
await ProcessInboundMessageAsync(msg, ct);
|
||||
}
|
||||
|
||||
// Wait for either a new message or health check tick
|
||||
var readTask = reader.WaitToReadAsync(ct).AsTask();
|
||||
var healthTask = healthTimer.WaitForNextTickAsync(ct).AsTask();
|
||||
await Task.WhenAny(readTask, healthTask);
|
||||
|
||||
if (ct.IsCancellationRequested)
|
||||
break;
|
||||
|
||||
// Drain any messages that arrived
|
||||
while (reader.TryRead(out var msg2))
|
||||
{
|
||||
await ProcessInboundMessageAsync(msg2, ct);
|
||||
}
|
||||
}
|
||||
catch (OperationCanceledException) when (ct.IsCancellationRequested)
|
||||
{
|
||||
break;
|
||||
}
|
||||
catch (Exception)
|
||||
{
|
||||
// Go: mset.retryMirrorConsumer() on errors
|
||||
lock (_gate)
|
||||
{
|
||||
_consecutiveFailures++;
|
||||
}
|
||||
|
||||
var delay = CalculateBackoff(_consecutiveFailures);
|
||||
try { await Task.Delay(delay, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------------
|
||||
// Background sync loop: pull-based (we fetch from origin)
|
||||
// Go reference: server/stream.go:3125-3400 (setupMirrorConsumer creates
|
||||
// ephemeral pull consumer; processMirrorMsgs drains it)
|
||||
// -------------------------------------------------------------------------
|
||||
private async Task RunPullSyncLoopAsync(IStreamStore originStore, int batchSize, CancellationToken ct)
|
||||
{
|
||||
while (!ct.IsCancellationRequested)
|
||||
{
|
||||
try
|
||||
{
|
||||
var messages = await originStore.ListAsync(ct);
|
||||
var applied = 0;
|
||||
|
||||
foreach (var msg in messages)
|
||||
{
|
||||
if (ct.IsCancellationRequested) break;
|
||||
|
||||
// Skip messages we've already synced
|
||||
if (msg.Sequence <= LastOriginSequence)
|
||||
continue;
|
||||
|
||||
await ProcessInboundMessageAsync(msg, ct);
|
||||
applied++;
|
||||
|
||||
if (applied >= batchSize)
|
||||
break;
|
||||
}
|
||||
|
||||
// Update lag based on origin state
|
||||
if (messages.Count > 0)
|
||||
{
|
||||
var originLast = messages[^1].Sequence;
|
||||
Lag = originLast > LastOriginSequence ? originLast - LastOriginSequence : 0;
|
||||
}
|
||||
|
||||
lock (_gate) _consecutiveFailures = 0;
|
||||
|
||||
// Go: If caught up, wait briefly before next poll
|
||||
if (applied == 0)
|
||||
{
|
||||
try { await Task.Delay(HeartbeatInterval, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
catch (OperationCanceledException) when (ct.IsCancellationRequested)
|
||||
{
|
||||
break;
|
||||
}
|
||||
catch (Exception)
|
||||
{
|
||||
lock (_gate) _consecutiveFailures++;
|
||||
var delay = CalculateBackoff(_consecutiveFailures);
|
||||
try { await Task.Delay(delay, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Go reference: server/stream.go:2863-3014 (processInboundMirrorMsg)
|
||||
private async Task ProcessInboundMessageAsync(StoredMessage message, CancellationToken ct)
|
||||
{
|
||||
// Go: sseq <= mset.mirror.sseq — ignore older messages
|
||||
if (_expectedOriginSeq > 0 && message.Sequence <= _expectedOriginSeq)
|
||||
return;
|
||||
|
||||
// Go: dc > 1 — skip redelivered messages
|
||||
if (message.Redelivered)
|
||||
return;
|
||||
|
||||
// Go: sseq == mset.mirror.sseq+1 — normal sequential delivery
|
||||
// Go: else — gap handling (skip sequences if deliver seq matches)
|
||||
await _targetStore.AppendAsync(message.Subject, message.Payload, ct);
|
||||
_expectedOriginSeq = message.Sequence;
|
||||
_deliverySeq++;
|
||||
LastOriginSequence = message.Sequence;
|
||||
LastSyncUtc = DateTime.UtcNow;
|
||||
|
||||
lock (_gate) _consecutiveFailures = 0;
|
||||
}
|
||||
|
||||
// Go reference: server/stream.go:3478-3505 (calculateRetryBackoff in setupSourceConsumer)
|
||||
// Exponential backoff with jitter, capped at MaxRetryDelay.
|
||||
private static TimeSpan CalculateBackoff(int failures)
|
||||
{
|
||||
var baseDelay = InitialRetryDelay.TotalMilliseconds * Math.Pow(2, Math.Min(failures - 1, 10));
|
||||
var capped = Math.Min(baseDelay, MaxRetryDelay.TotalMilliseconds);
|
||||
var jitter = Random.Shared.NextDouble() * 0.2 * capped; // +-20% jitter
|
||||
return TimeSpan.FromMilliseconds(capped + jitter);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Health report for a mirror coordinator, used by monitoring endpoints.
|
||||
/// Go reference: server/stream.go:2698-2736 (sourceInfo/StreamSourceInfo)
|
||||
/// </summary>
|
||||
public sealed record MirrorHealthReport
|
||||
{
|
||||
public ulong LastOriginSequence { get; init; }
|
||||
public DateTime LastSyncUtc { get; init; }
|
||||
public ulong Lag { get; init; }
|
||||
public int ConsecutiveFailures { get; init; }
|
||||
public bool IsRunning { get; init; }
|
||||
public bool IsStalled { get; init; }
|
||||
}
|
||||
|
||||
@@ -1,23 +1,109 @@
|
||||
using NATS.Server.JetStream.Storage;
|
||||
using System.Collections.Concurrent;
|
||||
using System.Threading.Channels;
|
||||
using NATS.Server.JetStream.Models;
|
||||
using NATS.Server.JetStream.Storage;
|
||||
using NATS.Server.Subscriptions;
|
||||
|
||||
namespace NATS.Server.JetStream.MirrorSource;
|
||||
|
||||
public sealed class SourceCoordinator
|
||||
// Go reference: server/stream.go:3860-4007 (processInboundSourceMsg)
|
||||
// Go reference: server/stream.go:3752-3833 (processAllSourceMsgs)
|
||||
// Go reference: server/stream.go:3474-3720 (setupSourceConsumer, trySetupSourceConsumer)
|
||||
|
||||
/// <summary>
|
||||
/// Coordinates consumption from a source stream into a target stream with support for:
|
||||
/// - Subject filtering via FilterSubject (Go: StreamSource.FilterSubject)
|
||||
/// - Subject transform prefix applied before storing (Go: SubjectTransforms)
|
||||
/// - Account isolation via SourceAccount
|
||||
/// - Deduplication via Nats-Msg-Id header with configurable window
|
||||
/// - Lag tracking per source
|
||||
/// - Background sync loop with exponential backoff retry
|
||||
/// </summary>
|
||||
public sealed class SourceCoordinator : IAsyncDisposable
|
||||
{
|
||||
// Go: sourceHealthCheckInterval = 10 * time.Second
|
||||
private static readonly TimeSpan HealthCheckInterval = TimeSpan.FromSeconds(10);
|
||||
|
||||
// Go: sourceHealthHB = 1 * time.Second
|
||||
private static readonly TimeSpan HeartbeatInterval = TimeSpan.FromSeconds(1);
|
||||
|
||||
private static readonly TimeSpan InitialRetryDelay = TimeSpan.FromMilliseconds(250);
|
||||
private static readonly TimeSpan MaxRetryDelay = TimeSpan.FromSeconds(30);
|
||||
private const int DefaultBatchSize = 256;
|
||||
|
||||
private readonly IStreamStore _targetStore;
|
||||
private readonly StreamSourceConfig _sourceConfig;
|
||||
private readonly Channel<StoredMessage> _inbound;
|
||||
private readonly Lock _gate = new();
|
||||
private CancellationTokenSource? _cts;
|
||||
private Task? _syncLoop;
|
||||
private int _consecutiveFailures;
|
||||
|
||||
// Go: si.sseq — last stream sequence from origin
|
||||
private ulong _expectedOriginSeq;
|
||||
|
||||
// Go: si.dseq — delivery sequence tracking
|
||||
private ulong _deliverySeq;
|
||||
|
||||
// Deduplication state: tracks recently seen Nats-Msg-Id values with their timestamps.
|
||||
// Go: server/stream.go doesn't have per-source dedup, but the stream's duplicate window
|
||||
// (DuplicateWindowMs) applies to publishes. We implement source-level dedup here.
|
||||
private readonly ConcurrentDictionary<string, DateTime> _dedupWindow = new(StringComparer.Ordinal);
|
||||
private DateTime _lastDedupPrune = DateTime.UtcNow;
|
||||
|
||||
/// <summary>Last sequence number successfully applied from the origin stream.</summary>
|
||||
public ulong LastOriginSequence { get; private set; }
|
||||
|
||||
/// <summary>UTC timestamp of the last successful sync operation.</summary>
|
||||
public DateTime LastSyncUtc { get; private set; }
|
||||
|
||||
/// <summary>Number of consecutive sync failures (resets on success).</summary>
|
||||
public int ConsecutiveFailures
|
||||
{
|
||||
get { lock (_gate) return _consecutiveFailures; }
|
||||
}
|
||||
|
||||
/// <summary>Whether the background sync loop is actively running.</summary>
|
||||
public bool IsRunning
|
||||
{
|
||||
get { lock (_gate) return _syncLoop is not null && !_syncLoop.IsCompleted; }
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Current lag: origin last sequence minus local last sequence.
|
||||
/// Returns 0 when fully caught up.
|
||||
/// </summary>
|
||||
public ulong Lag { get; private set; }
|
||||
|
||||
/// <summary>Total messages dropped by the subject filter.</summary>
|
||||
public long FilteredOutCount { get; private set; }
|
||||
|
||||
/// <summary>Total messages dropped by deduplication.</summary>
|
||||
public long DeduplicatedCount { get; private set; }
|
||||
|
||||
/// <summary>The source configuration driving this coordinator.</summary>
|
||||
public StreamSourceConfig Config => _sourceConfig;
|
||||
|
||||
public SourceCoordinator(IStreamStore targetStore, StreamSourceConfig sourceConfig)
|
||||
{
|
||||
_targetStore = targetStore;
|
||||
_sourceConfig = sourceConfig;
|
||||
_inbound = Channel.CreateUnbounded<StoredMessage>(new UnboundedChannelOptions
|
||||
{
|
||||
SingleReader = true,
|
||||
SingleWriter = false,
|
||||
});
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Processes a single inbound message from the origin stream.
|
||||
/// This is the direct-call path used when the origin and target are in the same process.
|
||||
/// Go reference: server/stream.go:3860-4007 (processInboundSourceMsg)
|
||||
/// </summary>
|
||||
public async Task OnOriginAppendAsync(StoredMessage message, CancellationToken ct)
|
||||
{
|
||||
// Account isolation: skip messages from different accounts.
|
||||
// Go: This is checked at the subscription level, but we enforce it here for in-process sources.
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.SourceAccount)
|
||||
&& !string.IsNullOrWhiteSpace(message.Account)
|
||||
&& !string.Equals(_sourceConfig.SourceAccount, message.Account, StringComparison.Ordinal))
|
||||
@@ -25,12 +111,360 @@ public sealed class SourceCoordinator
|
||||
return;
|
||||
}
|
||||
|
||||
// Subject filter: only forward messages matching the filter.
|
||||
// Go: server/stream.go:3597-3598 — if ssi.FilterSubject != _EMPTY_ { req.Config.FilterSubject = ssi.FilterSubject }
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.FilterSubject)
|
||||
&& !SubjectMatch.MatchLiteral(message.Subject, _sourceConfig.FilterSubject))
|
||||
{
|
||||
FilteredOutCount++;
|
||||
return;
|
||||
}
|
||||
|
||||
// Deduplication: check Nats-Msg-Id header against the dedup window.
|
||||
if (_sourceConfig.DuplicateWindowMs > 0 && message.MsgId is not null)
|
||||
{
|
||||
if (IsDuplicate(message.MsgId))
|
||||
{
|
||||
DeduplicatedCount++;
|
||||
return;
|
||||
}
|
||||
|
||||
RecordMsgId(message.MsgId);
|
||||
}
|
||||
|
||||
// Go: si.sseq <= current — ignore older/duplicate messages
|
||||
if (_expectedOriginSeq > 0 && message.Sequence <= _expectedOriginSeq)
|
||||
return;
|
||||
|
||||
// Subject transform: apply prefix before storing.
|
||||
// Go: server/stream.go:3943-3956 (subject transform for the source)
|
||||
var subject = message.Subject;
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.SubjectTransformPrefix))
|
||||
subject = $"{_sourceConfig.SubjectTransformPrefix}{subject}";
|
||||
|
||||
await _targetStore.AppendAsync(subject, message.Payload, ct);
|
||||
_expectedOriginSeq = message.Sequence;
|
||||
_deliverySeq++;
|
||||
LastOriginSequence = message.Sequence;
|
||||
LastSyncUtc = DateTime.UtcNow;
|
||||
Lag = 0;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Enqueues a message for processing by the background sync loop.
|
||||
/// </summary>
|
||||
public bool TryEnqueue(StoredMessage message)
|
||||
{
|
||||
return _inbound.Writer.TryWrite(message);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Starts the background sync loop that drains the inbound channel.
|
||||
/// Go reference: server/stream.go:3752-3833 (processAllSourceMsgs)
|
||||
/// </summary>
|
||||
public void StartSyncLoop()
|
||||
{
|
||||
lock (_gate)
|
||||
{
|
||||
if (_syncLoop is not null && !_syncLoop.IsCompleted)
|
||||
return;
|
||||
|
||||
_cts = new CancellationTokenSource();
|
||||
_syncLoop = RunSyncLoopAsync(_cts.Token);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Starts a pull-based sync loop that actively fetches from the origin store.
|
||||
/// Go reference: server/stream.go:3474-3720 (setupSourceConsumer + trySetupSourceConsumer)
|
||||
/// </summary>
|
||||
public void StartPullSyncLoop(IStreamStore originStore, int batchSize = DefaultBatchSize)
|
||||
{
|
||||
lock (_gate)
|
||||
{
|
||||
if (_syncLoop is not null && !_syncLoop.IsCompleted)
|
||||
return;
|
||||
|
||||
_cts = new CancellationTokenSource();
|
||||
_syncLoop = RunPullSyncLoopAsync(originStore, batchSize, _cts.Token);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Stops the background sync loop.
|
||||
/// Go reference: server/stream.go:3438-3469 (cancelSourceConsumer)
|
||||
/// </summary>
|
||||
public async Task StopAsync()
|
||||
{
|
||||
CancellationTokenSource? cts;
|
||||
Task? loop;
|
||||
lock (_gate)
|
||||
{
|
||||
cts = _cts;
|
||||
loop = _syncLoop;
|
||||
}
|
||||
|
||||
if (cts is not null)
|
||||
{
|
||||
await cts.CancelAsync();
|
||||
if (loop is not null)
|
||||
{
|
||||
try { await loop; }
|
||||
catch (OperationCanceledException) { }
|
||||
}
|
||||
}
|
||||
|
||||
lock (_gate)
|
||||
{
|
||||
_cts?.Dispose();
|
||||
_cts = null;
|
||||
_syncLoop = null;
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Reports current health state for monitoring.
|
||||
/// Go reference: server/stream.go:2687-2695 (sourcesInfo)
|
||||
/// </summary>
|
||||
public SourceHealthReport GetHealthReport(ulong? originLastSeq = null)
|
||||
{
|
||||
var lag = originLastSeq.HasValue && originLastSeq.Value > LastOriginSequence
|
||||
? originLastSeq.Value - LastOriginSequence
|
||||
: Lag;
|
||||
|
||||
return new SourceHealthReport
|
||||
{
|
||||
SourceName = _sourceConfig.Name,
|
||||
FilterSubject = _sourceConfig.FilterSubject,
|
||||
LastOriginSequence = LastOriginSequence,
|
||||
LastSyncUtc = LastSyncUtc,
|
||||
Lag = lag,
|
||||
ConsecutiveFailures = ConsecutiveFailures,
|
||||
IsRunning = IsRunning,
|
||||
IsStalled = LastSyncUtc != default
|
||||
&& DateTime.UtcNow - LastSyncUtc > HealthCheckInterval,
|
||||
FilteredOutCount = FilteredOutCount,
|
||||
DeduplicatedCount = DeduplicatedCount,
|
||||
};
|
||||
}
|
||||
|
||||
public async ValueTask DisposeAsync()
|
||||
{
|
||||
await StopAsync();
|
||||
_inbound.Writer.TryComplete();
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------------
|
||||
// Background sync loop: channel-based
|
||||
// Go reference: server/stream.go:3752-3833 (processAllSourceMsgs)
|
||||
// -------------------------------------------------------------------------
|
||||
private async Task RunSyncLoopAsync(CancellationToken ct)
|
||||
{
|
||||
using var healthTimer = new PeriodicTimer(HealthCheckInterval);
|
||||
var reader = _inbound.Reader;
|
||||
|
||||
while (!ct.IsCancellationRequested)
|
||||
{
|
||||
try
|
||||
{
|
||||
while (reader.TryRead(out var msg))
|
||||
{
|
||||
await ProcessInboundMessageAsync(msg, ct);
|
||||
}
|
||||
|
||||
var readTask = reader.WaitToReadAsync(ct).AsTask();
|
||||
var healthTask = healthTimer.WaitForNextTickAsync(ct).AsTask();
|
||||
await Task.WhenAny(readTask, healthTask);
|
||||
|
||||
if (ct.IsCancellationRequested)
|
||||
break;
|
||||
|
||||
while (reader.TryRead(out var msg2))
|
||||
{
|
||||
await ProcessInboundMessageAsync(msg2, ct);
|
||||
}
|
||||
}
|
||||
catch (OperationCanceledException) when (ct.IsCancellationRequested)
|
||||
{
|
||||
break;
|
||||
}
|
||||
catch (Exception)
|
||||
{
|
||||
lock (_gate) _consecutiveFailures++;
|
||||
var delay = CalculateBackoff(_consecutiveFailures);
|
||||
try { await Task.Delay(delay, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------------
|
||||
// Background sync loop: pull-based
|
||||
// Go reference: server/stream.go:3474-3720 (setupSourceConsumer)
|
||||
// -------------------------------------------------------------------------
|
||||
private async Task RunPullSyncLoopAsync(IStreamStore originStore, int batchSize, CancellationToken ct)
|
||||
{
|
||||
while (!ct.IsCancellationRequested)
|
||||
{
|
||||
try
|
||||
{
|
||||
var messages = await originStore.ListAsync(ct);
|
||||
var applied = 0;
|
||||
|
||||
foreach (var msg in messages)
|
||||
{
|
||||
if (ct.IsCancellationRequested) break;
|
||||
|
||||
if (msg.Sequence <= LastOriginSequence)
|
||||
continue;
|
||||
|
||||
await ProcessInboundMessageAsync(msg, ct);
|
||||
applied++;
|
||||
|
||||
if (applied >= batchSize)
|
||||
break;
|
||||
}
|
||||
|
||||
// Update lag
|
||||
if (messages.Count > 0)
|
||||
{
|
||||
var originLast = messages[^1].Sequence;
|
||||
Lag = originLast > LastOriginSequence ? originLast - LastOriginSequence : 0;
|
||||
}
|
||||
|
||||
lock (_gate) _consecutiveFailures = 0;
|
||||
|
||||
if (applied == 0)
|
||||
{
|
||||
try { await Task.Delay(HeartbeatInterval, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
catch (OperationCanceledException) when (ct.IsCancellationRequested)
|
||||
{
|
||||
break;
|
||||
}
|
||||
catch (Exception)
|
||||
{
|
||||
lock (_gate) _consecutiveFailures++;
|
||||
var delay = CalculateBackoff(_consecutiveFailures);
|
||||
try { await Task.Delay(delay, ct); }
|
||||
catch (OperationCanceledException) { break; }
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Go reference: server/stream.go:3860-4007 (processInboundSourceMsg)
|
||||
private async Task ProcessInboundMessageAsync(StoredMessage message, CancellationToken ct)
|
||||
{
|
||||
// Account isolation
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.SourceAccount)
|
||||
&& !string.IsNullOrWhiteSpace(message.Account)
|
||||
&& !string.Equals(_sourceConfig.SourceAccount, message.Account, StringComparison.Ordinal))
|
||||
{
|
||||
return;
|
||||
}
|
||||
|
||||
// Subject filter
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.FilterSubject)
|
||||
&& !SubjectMatch.MatchLiteral(message.Subject, _sourceConfig.FilterSubject))
|
||||
{
|
||||
FilteredOutCount++;
|
||||
return;
|
||||
}
|
||||
|
||||
// Deduplication
|
||||
if (_sourceConfig.DuplicateWindowMs > 0 && message.MsgId is not null)
|
||||
{
|
||||
if (IsDuplicate(message.MsgId))
|
||||
{
|
||||
DeduplicatedCount++;
|
||||
return;
|
||||
}
|
||||
|
||||
RecordMsgId(message.MsgId);
|
||||
}
|
||||
|
||||
// Skip already-seen sequences
|
||||
if (_expectedOriginSeq > 0 && message.Sequence <= _expectedOriginSeq)
|
||||
return;
|
||||
|
||||
// Redelivery check (Go: dc > 1)
|
||||
if (message.Redelivered)
|
||||
return;
|
||||
|
||||
// Subject transform
|
||||
var subject = message.Subject;
|
||||
if (!string.IsNullOrWhiteSpace(_sourceConfig.SubjectTransformPrefix))
|
||||
subject = $"{_sourceConfig.SubjectTransformPrefix}{subject}";
|
||||
|
||||
await _targetStore.AppendAsync(subject, message.Payload, ct);
|
||||
_expectedOriginSeq = message.Sequence;
|
||||
_deliverySeq++;
|
||||
LastOriginSequence = message.Sequence;
|
||||
LastSyncUtc = DateTime.UtcNow;
|
||||
|
||||
lock (_gate) _consecutiveFailures = 0;
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------------
|
||||
// Deduplication helpers
|
||||
// -------------------------------------------------------------------------
|
||||
|
||||
private bool IsDuplicate(string msgId)
|
||||
{
|
||||
PruneDedupWindowIfNeeded();
|
||||
return _dedupWindow.ContainsKey(msgId);
|
||||
}
|
||||
|
||||
private void RecordMsgId(string msgId)
|
||||
{
|
||||
_dedupWindow[msgId] = DateTime.UtcNow;
|
||||
}
|
||||
|
||||
private void PruneDedupWindowIfNeeded()
|
||||
{
|
||||
if (_sourceConfig.DuplicateWindowMs <= 0)
|
||||
return;
|
||||
|
||||
var now = DateTime.UtcNow;
|
||||
// Prune at most once per second to avoid overhead
|
||||
if ((now - _lastDedupPrune).TotalMilliseconds < 1000)
|
||||
return;
|
||||
|
||||
_lastDedupPrune = now;
|
||||
var cutoff = now.AddMilliseconds(-_sourceConfig.DuplicateWindowMs);
|
||||
foreach (var kvp in _dedupWindow)
|
||||
{
|
||||
if (kvp.Value < cutoff)
|
||||
_dedupWindow.TryRemove(kvp.Key, out _);
|
||||
}
|
||||
}
|
||||
|
||||
// Go reference: server/stream.go:3478-3505 (calculateRetryBackoff)
|
||||
private static TimeSpan CalculateBackoff(int failures)
|
||||
{
|
||||
var baseDelay = InitialRetryDelay.TotalMilliseconds * Math.Pow(2, Math.Min(failures - 1, 10));
|
||||
var capped = Math.Min(baseDelay, MaxRetryDelay.TotalMilliseconds);
|
||||
var jitter = Random.Shared.NextDouble() * 0.2 * capped;
|
||||
return TimeSpan.FromMilliseconds(capped + jitter);
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Health report for a source coordinator, used by monitoring endpoints.
|
||||
/// Go reference: server/stream.go:2687-2736 (sourcesInfo, sourceInfo)
|
||||
/// </summary>
|
||||
public sealed record SourceHealthReport
|
||||
{
|
||||
public string SourceName { get; init; } = string.Empty;
|
||||
public string? FilterSubject { get; init; }
|
||||
public ulong LastOriginSequence { get; init; }
|
||||
public DateTime LastSyncUtc { get; init; }
|
||||
public ulong Lag { get; init; }
|
||||
public int ConsecutiveFailures { get; init; }
|
||||
public bool IsRunning { get; init; }
|
||||
public bool IsStalled { get; init; }
|
||||
public long FilteredOutCount { get; init; }
|
||||
public long DeduplicatedCount { get; init; }
|
||||
}
|
||||
|
||||
@@ -35,4 +35,12 @@ public sealed class StreamSourceConfig
|
||||
public string Name { get; set; } = string.Empty;
|
||||
public string? SubjectTransformPrefix { get; set; }
|
||||
public string? SourceAccount { get; set; }
|
||||
|
||||
// Go: StreamSource.FilterSubject — only forward messages matching this subject filter.
|
||||
public string? FilterSubject { get; set; }
|
||||
|
||||
// Deduplication window in milliseconds for Nats-Msg-Id header-based dedup.
|
||||
// Defaults to 0 (disabled). When > 0, duplicate messages with the same Nats-Msg-Id
|
||||
// within this window are silently dropped.
|
||||
public int DuplicateWindowMs { get; set; }
|
||||
}
|
||||
|
||||
@@ -8,4 +8,14 @@ public sealed class StoredMessage
|
||||
public DateTime TimestampUtc { get; init; } = DateTime.UtcNow;
|
||||
public string? Account { get; init; }
|
||||
public bool Redelivered { get; init; }
|
||||
|
||||
/// <summary>
|
||||
/// Optional message headers. Used for deduplication (Nats-Msg-Id) and source tracking.
|
||||
/// </summary>
|
||||
public IReadOnlyDictionary<string, string>? Headers { get; init; }
|
||||
|
||||
/// <summary>
|
||||
/// Convenience accessor for the Nats-Msg-Id header value, used by source deduplication.
|
||||
/// </summary>
|
||||
public string? MsgId => Headers is not null && Headers.TryGetValue("Nats-Msg-Id", out var id) ? id : null;
|
||||
}
|
||||
|
||||
@@ -336,6 +336,8 @@ public sealed class StreamManager
|
||||
Name = s.Name,
|
||||
SubjectTransformPrefix = s.SubjectTransformPrefix,
|
||||
SourceAccount = s.SourceAccount,
|
||||
FilterSubject = s.FilterSubject,
|
||||
DuplicateWindowMs = s.DuplicateWindowMs,
|
||||
})],
|
||||
};
|
||||
|
||||
|
||||
Reference in New Issue
Block a user