| | | 1 | | using Microsoft.Extensions.Logging; |
| | | 2 | | |
| | | 3 | | namespace AsyncResponse; |
| | | 4 | | |
| | | 5 | | /// <summary>Coordinates atomic flow creation, optimistic updates, and one active executor per flow id.</summary> |
| | | 6 | | internal static class FlowStateConcurrency |
| | | 7 | | { |
| | | 8 | | private const int MaxUpdateAttempts = 8; |
| | | 9 | | |
| | | 10 | | public static Task<bool> TryCreateAsync( |
| | | 11 | | IFlowStateStore store, |
| | | 12 | | string flowId, |
| | | 13 | | FlowState state, |
| | | 14 | | TimeSpan ttl, |
| | | 15 | | CancellationToken cancellationToken = default) |
| | | 16 | | { |
| | 3 | 17 | | state.Revision = 0; |
| | 3 | 18 | | return store.TryCreateAsync(flowId, state, ttl, cancellationToken); |
| | | 19 | | } |
| | | 20 | | |
| | | 21 | | public static async Task<FlowExecutionLease?> TryAcquireExecutionLeaseAsync( |
| | | 22 | | IFlowStateStore store, |
| | | 23 | | string flowId, |
| | | 24 | | DurableFlowOptions options, |
| | | 25 | | ILogger logger, |
| | | 26 | | CancellationToken cancellationToken = default) |
| | | 27 | | { |
| | 3 | 28 | | ValidateOptions(options); |
| | | 29 | | |
| | 3 | 30 | | var leaseId = Guid.NewGuid().ToString("N"); |
| | 3 | 31 | | if (!await store.TryAcquireLeaseAsync( |
| | 3 | 32 | | flowId, |
| | 3 | 33 | | leaseId, |
| | 3 | 34 | | options.ExecutionLeaseDuration, |
| | 3 | 35 | | cancellationToken).ConfigureAwait(false)) |
| | 2 | 36 | | return null; |
| | | 37 | | |
| | 3 | 38 | | return new FlowExecutionLease(store, flowId, leaseId, options, logger); |
| | 3 | 39 | | } |
| | | 40 | | |
| | | 41 | | public static async Task<bool> MutateAsync( |
| | | 42 | | IFlowStateStore store, |
| | | 43 | | string flowId, |
| | | 44 | | TimeSpan ttl, |
| | | 45 | | Func<FlowState, bool> mutate, |
| | | 46 | | CancellationToken cancellationToken = default) |
| | | 47 | | { |
| | 2 | 48 | | for (var attempt = 0; attempt < MaxUpdateAttempts; attempt++) |
| | | 49 | | { |
| | 2 | 50 | | var state = await store.LoadAsync(flowId, cancellationToken).ConfigureAwait(false); |
| | 2 | 51 | | if (state is null) |
| | 2 | 52 | | return false; |
| | | 53 | | |
| | 2 | 54 | | if (!mutate(state)) |
| | 2 | 55 | | return true; |
| | | 56 | | |
| | 2 | 57 | | var expectedRevision = state.Revision; |
| | 2 | 58 | | state.Revision = checked(expectedRevision + 1); |
| | 2 | 59 | | state.UpdatedAtUtc = DateTime.UtcNow; |
| | 2 | 60 | | if (await store.TryUpdateAsync( |
| | 2 | 61 | | flowId, |
| | 2 | 62 | | state, |
| | 2 | 63 | | expectedRevision, |
| | 2 | 64 | | ttl, |
| | 2 | 65 | | leaseId: null, |
| | 2 | 66 | | cancellationToken).ConfigureAwait(false)) |
| | 2 | 67 | | return true; |
| | | 68 | | } |
| | | 69 | | |
| | 2 | 70 | | throw new InvalidOperationException( |
| | 2 | 71 | | $"Durable flow '{flowId}' changed repeatedly while applying a recovery update; retry the operation."); |
| | 2 | 72 | | } |
| | | 73 | | |
| | | 74 | | internal static void ValidateOptions(DurableFlowOptions options) |
| | | 75 | | { |
| | 3 | 76 | | if (options.StateExpiry <= TimeSpan.Zero) |
| | 2 | 77 | | throw new InvalidOperationException($"{nameof(DurableFlowOptions)}.{nameof(options.StateExpiry)} must be pos |
| | 3 | 78 | | if (options.DefaultStepTimeout is { } defaultStepTimeout && defaultStepTimeout <= TimeSpan.Zero) |
| | 2 | 79 | | throw new InvalidOperationException($"{nameof(DurableFlowOptions)}.{nameof(options.DefaultStepTimeout)} must |
| | 3 | 80 | | if (options.ExecutionLeaseDuration <= TimeSpan.Zero) |
| | 2 | 81 | | throw new InvalidOperationException($"{nameof(DurableFlowOptions)}.{nameof(options.ExecutionLeaseDuration)} |
| | 3 | 82 | | if (options.ExecutionLeaseRenewInterval <= TimeSpan.Zero |
| | 3 | 83 | | || options.ExecutionLeaseRenewInterval >= options.ExecutionLeaseDuration) |
| | | 84 | | { |
| | 2 | 85 | | throw new InvalidOperationException( |
| | 2 | 86 | | $"{nameof(DurableFlowOptions)}.{nameof(options.ExecutionLeaseRenewInterval)} must be positive and shorte |
| | 2 | 87 | | $"{nameof(DurableFlowOptions.ExecutionLeaseDuration)}."); |
| | | 88 | | } |
| | 3 | 89 | | if (options.ProgressPersistenceInterval < TimeSpan.Zero) |
| | 2 | 90 | | throw new InvalidOperationException($"{nameof(DurableFlowOptions)}.{nameof(options.ProgressPersistenceInterv |
| | 3 | 91 | | } |
| | | 92 | | } |
| | | 93 | | |
| | | 94 | | /// <summary>One distributed durable-flow execution lease.</summary> |
| | | 95 | | internal sealed class FlowExecutionLease : IAsyncDisposable |
| | | 96 | | { |
| | | 97 | | private readonly IFlowStateStore _store; |
| | | 98 | | private readonly string _flowId; |
| | | 99 | | private readonly string _leaseId; |
| | | 100 | | private readonly DurableFlowOptions _options; |
| | | 101 | | private readonly ILogger _logger; |
| | | 102 | | private readonly CancellationTokenSource _stop = new(); |
| | | 103 | | private readonly CancellationTokenSource _lost = new(); |
| | | 104 | | private readonly Task _renewal; |
| | | 105 | | private DateTime _validUntilUtc; |
| | | 106 | | private int _disposed; |
| | | 107 | | |
| | | 108 | | public FlowExecutionLease( |
| | | 109 | | IFlowStateStore store, |
| | | 110 | | string flowId, |
| | | 111 | | string leaseId, |
| | | 112 | | DurableFlowOptions options, |
| | | 113 | | ILogger logger) |
| | | 114 | | { |
| | | 115 | | _store = store; |
| | | 116 | | _flowId = flowId; |
| | | 117 | | _leaseId = leaseId; |
| | | 118 | | _options = options; |
| | | 119 | | _logger = logger; |
| | | 120 | | _validUntilUtc = DateTime.UtcNow.Add(options.ExecutionLeaseDuration); |
| | | 121 | | _renewal = RenewLoopAsync(); |
| | | 122 | | } |
| | | 123 | | |
| | | 124 | | public CancellationToken LostToken => _lost.Token; |
| | | 125 | | |
| | | 126 | | /// <summary> |
| | | 127 | | /// Throws when the lease is lost. <paramref name="cause"/> (e.g. the exception that made the |
| | | 128 | | /// caller check) is attached as the inner exception so the real failure is not discarded. |
| | | 129 | | /// </summary> |
| | | 130 | | public void ThrowIfLost(Exception? cause = null) |
| | | 131 | | { |
| | | 132 | | if (_lost.IsCancellationRequested) |
| | | 133 | | throw new InvalidOperationException($"Durable flow '{_flowId}' lost its execution lease; the worker will ret |
| | | 134 | | } |
| | | 135 | | |
| | | 136 | | public async Task SaveAsync(FlowState state, TimeSpan ttl, CancellationToken cancellationToken = default, Exception? |
| | | 137 | | { |
| | | 138 | | ThrowIfLost(cause); |
| | | 139 | | var expectedRevision = state.Revision; |
| | | 140 | | state.Revision = checked(expectedRevision + 1); |
| | | 141 | | state.UpdatedAtUtc = DateTime.UtcNow; |
| | | 142 | | |
| | | 143 | | try |
| | | 144 | | { |
| | | 145 | | if (await _store.TryUpdateAsync( |
| | | 146 | | _flowId, |
| | | 147 | | state, |
| | | 148 | | expectedRevision, |
| | | 149 | | ttl, |
| | | 150 | | _leaseId, |
| | | 151 | | cancellationToken).ConfigureAwait(false)) |
| | | 152 | | return; |
| | | 153 | | } |
| | | 154 | | catch |
| | | 155 | | { |
| | | 156 | | state.Revision = expectedRevision; |
| | | 157 | | MarkLost(); |
| | | 158 | | |
| | | 159 | | // The store exception propagates; keep the failure this save was recording from |
| | | 160 | | // vanishing with it. |
| | | 161 | | if (cause is not null) |
| | | 162 | | _logger.LogWarning(cause, "Durable flow '{FlowId}' failed to checkpoint; the failure it was recording is |
| | | 163 | | throw; |
| | | 164 | | } |
| | | 165 | | |
| | | 166 | | state.Revision = expectedRevision; |
| | | 167 | | MarkLost(); |
| | | 168 | | throw await CreateSaveRejectedExceptionAsync(expectedRevision, cause, cancellationToken).ConfigureAwait(false); |
| | | 169 | | } |
| | | 170 | | |
| | | 171 | | /// <summary> |
| | | 172 | | /// Builds the exception for a rejected checkpoint write. The store's compare-and-swap only |
| | | 173 | | /// returns <c>false</c>, so the reason is diagnosed with a best-effort re-read: a revision |
| | | 174 | | /// conflict — a concurrent lease-bypassing writer such as <c>RecoverAsync</c>, <c>FailAsync</c>, |
| | | 175 | | /// or an operator parking the run — is reported as such instead of as a lost lease, which sent |
| | | 176 | | /// operators hunting phantom lease problems. Behavior is unchanged either way: the lease is |
| | | 177 | | /// abandoned (<see cref="MarkLost"/> already ran) and the delivery retries from the last |
| | | 178 | | /// checkpoint; <paramref name="cause"/> rides along as the inner exception so the failure that |
| | | 179 | | /// triggered the save is not discarded. |
| | | 180 | | /// </summary> |
| | | 181 | | private async Task<InvalidOperationException> CreateSaveRejectedExceptionAsync( |
| | | 182 | | long expectedRevision, |
| | | 183 | | Exception? cause, |
| | | 184 | | CancellationToken cancellationToken) |
| | | 185 | | { |
| | | 186 | | var reason = "its execution lease was no longer held (expired or taken over)"; |
| | | 187 | | try |
| | | 188 | | { |
| | | 189 | | var current = await _store.LoadAsync(_flowId, cancellationToken).ConfigureAwait(false); |
| | | 190 | | if (current is null) |
| | | 191 | | reason = "its ledger entry is gone (expired or deleted)"; |
| | | 192 | | else if (current.Revision != expectedRevision) |
| | | 193 | | reason = $"a concurrent write advanced the ledger (revision {expectedRevision} -> {current.Revision}: a |
| | | 194 | | } |
| | | 195 | | catch |
| | | 196 | | { |
| | | 197 | | // Best-effort diagnosis only — the rejection itself is what matters. |
| | | 198 | | } |
| | | 199 | | |
| | | 200 | | return new InvalidOperationException( |
| | | 201 | | $"Durable flow '{_flowId}' could not checkpoint because {reason}; the worker abandons this execution and the |
| | | 202 | | cause); |
| | | 203 | | } |
| | | 204 | | |
| | | 205 | | private async Task RenewLoopAsync() |
| | | 206 | | { |
| | | 207 | | while (!_stop.IsCancellationRequested) |
| | | 208 | | { |
| | | 209 | | try |
| | | 210 | | { |
| | | 211 | | await Task.Delay(_options.ExecutionLeaseRenewInterval, _stop.Token).ConfigureAwait(false); |
| | | 212 | | if (!await _store.TryRenewLeaseAsync( |
| | | 213 | | _flowId, |
| | | 214 | | _leaseId, |
| | | 215 | | _options.ExecutionLeaseDuration, |
| | | 216 | | _stop.Token).ConfigureAwait(false)) |
| | | 217 | | { |
| | | 218 | | MarkLost(); |
| | | 219 | | return; |
| | | 220 | | } |
| | | 221 | | |
| | | 222 | | _validUntilUtc = DateTime.UtcNow.Add(_options.ExecutionLeaseDuration); |
| | | 223 | | } |
| | | 224 | | catch (OperationCanceledException) when (_stop.IsCancellationRequested) |
| | | 225 | | { |
| | | 226 | | return; |
| | | 227 | | } |
| | | 228 | | catch (Exception ex) |
| | | 229 | | { |
| | | 230 | | _logger.LogWarning(ex, "Failed to renew durable flow {FlowId} execution lease; retrying before expiry.", |
| | | 231 | | if (DateTime.UtcNow >= _validUntilUtc) |
| | | 232 | | { |
| | | 233 | | MarkLost(); |
| | | 234 | | return; |
| | | 235 | | } |
| | | 236 | | } |
| | | 237 | | } |
| | | 238 | | } |
| | | 239 | | |
| | | 240 | | private void MarkLost() |
| | | 241 | | { |
| | | 242 | | try |
| | | 243 | | { |
| | | 244 | | _lost.Cancel(); |
| | | 245 | | } |
| | | 246 | | catch (ObjectDisposedException) |
| | | 247 | | { |
| | | 248 | | // Disposal won the race. |
| | | 249 | | } |
| | | 250 | | } |
| | | 251 | | |
| | | 252 | | public async ValueTask DisposeAsync() |
| | | 253 | | { |
| | | 254 | | if (Interlocked.Exchange(ref _disposed, 1) != 0) |
| | | 255 | | return; |
| | | 256 | | |
| | | 257 | | _stop.Cancel(); |
| | | 258 | | await _renewal.ConfigureAwait(false); |
| | | 259 | | |
| | | 260 | | try |
| | | 261 | | { |
| | | 262 | | await _store.ReleaseLeaseAsync(_flowId, _leaseId, CancellationToken.None).ConfigureAwait(false); |
| | | 263 | | } |
| | | 264 | | catch (Exception ex) |
| | | 265 | | { |
| | | 266 | | _logger.LogWarning(ex, "Failed to release durable flow {FlowId} execution lease; it will expire.", _flowId); |
| | | 267 | | } |
| | | 268 | | |
| | | 269 | | _stop.Dispose(); |
| | | 270 | | _lost.Dispose(); |
| | | 271 | | } |
| | | 272 | | } |