| | | 1 | | using AsyncResponse; |
| | | 2 | | using AsyncResponse.DurableFlows.Internal; |
| | | 3 | | using AsyncResponse.DurableFlows.Cosmos; |
| | | 4 | | using Microsoft.Azure.Cosmos; |
| | | 5 | | using Microsoft.Extensions.DependencyInjection.Extensions; |
| | | 6 | | using Microsoft.Extensions.Options; |
| | | 7 | | using Newtonsoft.Json; |
| | | 8 | | using System.Net; |
| | | 9 | | |
| | | 10 | | namespace Microsoft.Extensions.DependencyInjection |
| | | 11 | | { |
| | | 12 | | /// <summary>DI registration for the Azure Cosmos DB durable-flow state store.</summary> |
| | | 13 | | public static class CosmosDurableFlowServiceCollectionExtensions |
| | | 14 | | { |
| | | 15 | | /// <summary> |
| | | 16 | | /// Stores durable-flow state in Azure Cosmos DB. Hosts may either register a |
| | | 17 | | /// <see cref="CosmosClient"/> singleton or set connection options here. |
| | | 18 | | /// </summary> |
| | | 19 | | public static AsyncResponseRegistrationBuilder WithCosmosDurableFlows( |
| | | 20 | | this AsyncResponseRegistrationBuilder builder, |
| | | 21 | | Action<CosmosDurableFlowOptions>? configure = null) |
| | | 22 | | { |
| | | 23 | | // Singleton on purpose: database/container provisioning is cached per store instance |
| | | 24 | | // and Cosmos metadata operations are RU-charged and rate-limited — a scoped store would |
| | | 25 | | // re-issue them on every flow execution. A host-registered CosmosClient is reused when |
| | | 26 | | // present; otherwise the store creates and owns one from ConnectionString. Nothing is |
| | | 27 | | // registered as a bare CosmosClient service, so unrelated resolutions of that type are |
| | | 28 | | // never answered — or broken — by this package. |
| | | 29 | | builder.Services.TryAddSingleton(provider => |
| | | 30 | | { |
| | | 31 | | var options = provider.GetRequiredService<IOptions<CosmosDurableFlowOptions>>(); |
| | | 32 | | |
| | | 33 | | var shared = provider.GetService<CosmosClient>(); |
| | | 34 | | if (shared is not null) |
| | | 35 | | return new CosmosFlowStateStore(shared, options); |
| | | 36 | | |
| | | 37 | | if (string.IsNullOrWhiteSpace(options.Value.ConnectionString)) |
| | | 38 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(CosmosDurableFlowOp |
| | | 39 | | return new CosmosFlowStateStore(new CosmosClient(options.Value.ConnectionString), options, ownsClient: t |
| | | 40 | | }); |
| | | 41 | | return builder.WithDurableFlows<CosmosFlowStateStore, CosmosDurableFlowOptions>(configure); |
| | | 42 | | } |
| | | 43 | | } |
| | | 44 | | } |
| | | 45 | | |
| | | 46 | | namespace AsyncResponse.DurableFlows.Cosmos |
| | | 47 | | { |
| | | 48 | | /// <summary>Options for the Azure Cosmos DB durable-flow state store.</summary> |
| | | 49 | | public sealed class CosmosDurableFlowOptions : DurableFlowOptions |
| | | 50 | | { |
| | | 51 | | /// <summary>Optional Cosmos DB connection string used when no <see cref="CosmosClient"/> is registered.</summary> |
| | | 52 | | public string? ConnectionString { get; set; } |
| | | 53 | | |
| | | 54 | | /// <summary>Cosmos database name. Required.</summary> |
| | | 55 | | public string? DatabaseName { get; set; } |
| | | 56 | | |
| | | 57 | | /// <summary>Container storing one durable-flow ledger document per flow id.</summary> |
| | | 58 | | public string ContainerName { get; set; } = "asyncresponse_flow_state"; |
| | | 59 | | |
| | | 60 | | /// <summary>Partition-key path for the container. Default: <c>/flowId</c>.</summary> |
| | | 61 | | public string PartitionKeyPath { get; set; } = "/flowId"; |
| | | 62 | | |
| | | 63 | | /// <summary>Creates the database and container on first use.</summary> |
| | | 64 | | public bool AutoCreateContainer { get; set; } = true; |
| | | 65 | | |
| | | 66 | | /// <summary>Optional throughput used when auto-creating the container.</summary> |
| | | 67 | | public int? Throughput { get; set; } |
| | | 68 | | |
| | | 69 | | /// <summary> |
| | | 70 | | /// Maximum serialized flow-state size in bytes accepted by writes; oversized ledgers fail fast |
| | | 71 | | /// with an actionable error instead of the raw Cosmos 413 the executor would retry into the |
| | | 72 | | /// dead-letter queue. Default: 1.9 MB (headroom under Cosmos's 2 MB item cap for the sibling |
| | | 73 | | /// fields); <c>null</c> disables the guard. |
| | | 74 | | /// </summary> |
| | | 75 | | public long? MaxStateBytes { get; set; } = 1_900_000; |
| | | 76 | | |
| | | 77 | | /// <summary>Validates option values and throws on misconfiguration.</summary> |
| | | 78 | | public void Validate() |
| | | 79 | | { |
| | | 80 | | if (string.IsNullOrWhiteSpace(DatabaseName)) |
| | | 81 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(DatabaseName)} must be conf |
| | | 82 | | if (string.IsNullOrWhiteSpace(ContainerName)) |
| | | 83 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(ContainerName)} must be con |
| | | 84 | | if (string.IsNullOrWhiteSpace(PartitionKeyPath) || !PartitionKeyPath.StartsWith("/", StringComparison.Ordinal)) |
| | | 85 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(PartitionKeyPath)} must sta |
| | | 86 | | if (Throughput is <= 0) |
| | | 87 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(Throughput)} must be positi |
| | | 88 | | if (MaxStateBytes is <= 0) |
| | | 89 | | throw new InvalidOperationException($"{nameof(CosmosDurableFlowOptions)}.{nameof(MaxStateBytes)} must be pos |
| | | 90 | | } |
| | | 91 | | } |
| | | 92 | | |
| | | 93 | | /// <summary>Azure Cosmos DB implementation of <see cref="IFlowStateStore"/>.</summary> |
| | | 94 | | public sealed class CosmosFlowStateStore : IFlowStateStore, IDisposable |
| | | 95 | | { |
| | | 96 | | // Time authority: this store keeps the app clock (DateTime.UtcNow) for expiry and lease |
| | | 97 | | // comparisons. Cosmos conditional writes (ETag preconditions) evaluate client-supplied |
| | | 98 | | // values only — there is no server-clock expression usable inside a point write — so the |
| | | 99 | | // read-check-replace cycles below compare against the app clock and rely on the ETag fence |
| | | 100 | | // for atomicity. Multi-node deployments should keep worker clocks synchronized well inside |
| | | 101 | | // the lease window. (The server's own TTL sweep, by contrast, runs on the service clock.) |
| | | 102 | | private readonly CosmosClient _client; |
| | | 103 | | private readonly CosmosDurableFlowOptions _options; |
| | 3 | 104 | | private readonly SemaphoreSlim _ensureGate = new(1, 1); |
| | | 105 | | private readonly bool _ownsClient; |
| | | 106 | | private bool _created; |
| | | 107 | | |
| | 3 | 108 | | public CosmosFlowStateStore(CosmosClient client, IOptions<CosmosDurableFlowOptions> options, bool ownsClient = false |
| | | 109 | | { |
| | 3 | 110 | | _client = client; |
| | 3 | 111 | | _options = options.Value; |
| | 3 | 112 | | _options.Validate(); |
| | 3 | 113 | | _ownsClient = ownsClient; |
| | 3 | 114 | | } |
| | | 115 | | |
| | | 116 | | public async Task<FlowState?> LoadAsync(string flowId, CancellationToken cancellationToken = default) |
| | | 117 | | { |
| | 3 | 118 | | ArgumentException.ThrowIfNullOrWhiteSpace(flowId); |
| | 3 | 119 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | | 120 | | |
| | | 121 | | try |
| | | 122 | | { |
| | 3 | 123 | | var response = await container.ReadItemAsync<CosmosFlowStateDocument>( |
| | 3 | 124 | | flowId, |
| | 3 | 125 | | new PartitionKey(flowId), |
| | 3 | 126 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 127 | | var document = response.Resource; |
| | | 128 | | // Point reads have no predicate, so the expiry check happens client-side on the app |
| | | 129 | | // clock — see the time-authority note on this class. |
| | 3 | 130 | | if (document.ExpiresAtUtc <= DateTime.UtcNow) |
| | 3 | 131 | | return null; |
| | | 132 | | |
| | 3 | 133 | | return document.Revision is { } revision |
| | 3 | 134 | | ? DurableFlowStoreShared.ReadState(flowId, document.StateJson, revision) |
| | 3 | 135 | | : null; |
| | | 136 | | } |
| | 3 | 137 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.NotFound) |
| | | 138 | | { |
| | 3 | 139 | | return null; |
| | | 140 | | } |
| | 3 | 141 | | } |
| | | 142 | | |
| | | 143 | | public async Task<bool> TryCreateAsync(string flowId, FlowState state, TimeSpan ttl, CancellationToken cancellationT |
| | | 144 | | { |
| | 3 | 145 | | DurableFlowStoreShared.ValidateCreate(flowId, state, ttl); |
| | 3 | 146 | | var stateJson = DurableFlowStoreShared.SerializeBounded(flowId, state, _options.MaxStateBytes, "Cosmos DB"); |
| | 3 | 147 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | | 148 | | |
| | | 149 | | // Bounded retry: documents carry native Cosmos TTL alongside the logical ExpiresAtUtc, so |
| | | 150 | | // the server's TTL sweep can physically purge an expired document between our create's |
| | | 151 | | // 409 and the follow-up read (or the conditional replace). A vanished conflicting |
| | | 152 | | // document means the slot is free — create again — not that a live competitor won. |
| | | 153 | | // The short growing delay lets one purge cycle finish instead of burning all attempts |
| | | 154 | | // inside the same inconsistency window (the Linux emulator is markedly worse here than |
| | | 155 | | // the service, but the race itself is real on both). |
| | 3 | 156 | | for (var attempt = 0; attempt < 4; attempt++) |
| | | 157 | | { |
| | 3 | 158 | | if (attempt > 0) |
| | 2 | 159 | | await Task.Delay(TimeSpan.FromMilliseconds(50 * attempt), cancellationToken).ConfigureAwait(false); |
| | | 160 | | |
| | 3 | 161 | | var now = DateTime.UtcNow; |
| | 3 | 162 | | var document = CreateDocument(flowId, stateJson, state.Revision, ttl, now); |
| | | 163 | | try |
| | | 164 | | { |
| | 3 | 165 | | await container.CreateItemAsync(document, new PartitionKey(flowId), cancellationToken: cancellationToken |
| | 3 | 166 | | return true; |
| | | 167 | | } |
| | 3 | 168 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.Conflict) |
| | | 169 | | { |
| | | 170 | | try |
| | | 171 | | { |
| | 3 | 172 | | var current = await container.ReadItemAsync<CosmosFlowStateDocument>( |
| | 3 | 173 | | flowId, |
| | 3 | 174 | | new PartitionKey(flowId), |
| | 3 | 175 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 176 | | if (current.Resource.ExpiresAtUtc > now) |
| | 3 | 177 | | return false; |
| | | 178 | | |
| | 3 | 179 | | await container.ReplaceItemAsync( |
| | 3 | 180 | | document, |
| | 3 | 181 | | flowId, |
| | 3 | 182 | | new PartitionKey(flowId), |
| | 3 | 183 | | new ItemRequestOptions { IfMatchEtag = current.ETag }, |
| | 3 | 184 | | cancellationToken).ConfigureAwait(false); |
| | 3 | 185 | | return true; |
| | | 186 | | } |
| | 2 | 187 | | catch (CosmosException retryEx) when (retryEx.StatusCode == HttpStatusCode.PreconditionFailed) |
| | | 188 | | { |
| | | 189 | | // The ETag moved under us. That is either a live writer winning the id or the |
| | | 190 | | // TTL purge touching the expired document; only the next read can tell them |
| | | 191 | | // apart, so retry instead of conceding — a live occupant surfaces as an |
| | | 192 | | // unexpired read (false) on the next attempt, a purged one as a clean create. |
| | 2 | 193 | | continue; |
| | | 194 | | } |
| | 0 | 195 | | catch (CosmosException retryEx) when (retryEx.StatusCode == HttpStatusCode.NotFound) |
| | | 196 | | { |
| | | 197 | | // The expired document was TTL-purged after our 409 — the id is free again. |
| | 0 | 198 | | continue; |
| | | 199 | | } |
| | | 200 | | } |
| | 0 | 201 | | } |
| | | 202 | | |
| | 2 | 203 | | return false; |
| | 3 | 204 | | } |
| | | 205 | | |
| | | 206 | | public async Task<bool> TryUpdateAsync( |
| | | 207 | | string flowId, |
| | | 208 | | FlowState state, |
| | | 209 | | long expectedRevision, |
| | | 210 | | TimeSpan ttl, |
| | | 211 | | string? leaseId = null, |
| | | 212 | | CancellationToken cancellationToken = default) |
| | | 213 | | { |
| | 3 | 214 | | DurableFlowStoreShared.ValidateUpdate(flowId, state, expectedRevision, ttl); |
| | 3 | 215 | | var stateJson = DurableFlowStoreShared.SerializeBounded(flowId, state, _options.MaxStateBytes, "Cosmos DB"); |
| | 3 | 216 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | 3 | 217 | | for (var attempt = 0; attempt < 4; attempt++) |
| | | 218 | | { |
| | 3 | 219 | | var now = DateTime.UtcNow; |
| | | 220 | | try |
| | | 221 | | { |
| | 3 | 222 | | var current = await container.ReadItemAsync<CosmosFlowStateDocument>( |
| | 3 | 223 | | flowId, |
| | 3 | 224 | | new PartitionKey(flowId), |
| | 3 | 225 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 226 | | var document = current.Resource; |
| | 3 | 227 | | if (document.ExpiresAtUtc <= now || document.Revision != expectedRevision) |
| | 3 | 228 | | return false; |
| | 3 | 229 | | if (leaseId is not null && (document.LeaseId != leaseId || document.LeaseExpiresAtUtc <= now)) |
| | 3 | 230 | | return false; |
| | | 231 | | |
| | 3 | 232 | | document.StateJson = stateJson; |
| | 3 | 233 | | document.ExpiresAtUtc = DurableFlowStoreShared.AddSaturating(now, ttl); |
| | 3 | 234 | | document.UpdatedAtUtc = now; |
| | 3 | 235 | | document.Revision = state.Revision; |
| | 3 | 236 | | document.Ttl = CosmosTtlSeconds(ttl); |
| | 3 | 237 | | await container.ReplaceItemAsync( |
| | 3 | 238 | | document, |
| | 3 | 239 | | flowId, |
| | 3 | 240 | | new PartitionKey(flowId), |
| | 3 | 241 | | new ItemRequestOptions { IfMatchEtag = current.ETag }, |
| | 3 | 242 | | cancellationToken).ConfigureAwait(false); |
| | 3 | 243 | | return true; |
| | | 244 | | } |
| | 2 | 245 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.NotFound) |
| | | 246 | | { |
| | 2 | 247 | | return false; |
| | | 248 | | } |
| | 2 | 249 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.PreconditionFailed) |
| | | 250 | | { |
| | | 251 | | // Lease renewal also replaces the document and changes its ETag without changing |
| | | 252 | | // the ledger revision. Re-read and retry so that benign race is not reported as a |
| | | 253 | | // lost execution lease; a real state race fails the revision check above. |
| | 2 | 254 | | } |
| | | 255 | | } |
| | | 256 | | |
| | 2 | 257 | | return false; |
| | 3 | 258 | | } |
| | | 259 | | |
| | | 260 | | public Task<bool> TryAcquireLeaseAsync(string flowId, string leaseId, TimeSpan leaseDuration, CancellationToken canc |
| | 3 | 261 | | => UpdateLeaseAsync(flowId, leaseId, leaseDuration, acquire: true, cancellationToken); |
| | | 262 | | |
| | | 263 | | public Task<bool> TryRenewLeaseAsync(string flowId, string leaseId, TimeSpan leaseDuration, CancellationToken cancel |
| | 3 | 264 | | => UpdateLeaseAsync(flowId, leaseId, leaseDuration, acquire: false, cancellationToken); |
| | | 265 | | |
| | | 266 | | public async Task ReleaseLeaseAsync(string flowId, string leaseId, CancellationToken cancellationToken = default) |
| | | 267 | | { |
| | 3 | 268 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | 3 | 269 | | for (var attempt = 0; attempt < 4; attempt++) |
| | | 270 | | { |
| | | 271 | | try |
| | | 272 | | { |
| | 3 | 273 | | var current = await container.ReadItemAsync<CosmosFlowStateDocument>( |
| | 3 | 274 | | flowId, |
| | 3 | 275 | | new PartitionKey(flowId), |
| | 3 | 276 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 277 | | if (current.Resource.LeaseId != leaseId) |
| | 3 | 278 | | return; |
| | | 279 | | |
| | 3 | 280 | | current.Resource.LeaseId = null; |
| | 3 | 281 | | current.Resource.LeaseExpiresAtUtc = null; |
| | | 282 | | // Every replace refreshes _ts — the anchor the server-side TTL counts from — so |
| | | 283 | | // re-persisting the stored full-window ttl would restart the physical-retention |
| | | 284 | | // countdown and decouple it from the logical ExpiresAtUtc. Rewrite it from the |
| | | 285 | | // remaining logical window instead. (Checkpoints recompute both together in |
| | | 286 | | // TryUpdateAsync; only the lease paths replace without moving ExpiresAtUtc.) |
| | 3 | 287 | | current.Resource.Ttl = CosmosTtlSeconds(current.Resource.ExpiresAtUtc, DateTime.UtcNow); |
| | 3 | 288 | | await container.ReplaceItemAsync( |
| | 3 | 289 | | current.Resource, |
| | 3 | 290 | | flowId, |
| | 3 | 291 | | new PartitionKey(flowId), |
| | 3 | 292 | | new ItemRequestOptions { IfMatchEtag = current.ETag }, |
| | 3 | 293 | | cancellationToken).ConfigureAwait(false); |
| | 3 | 294 | | return; |
| | | 295 | | } |
| | 2 | 296 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.NotFound) |
| | | 297 | | { |
| | 2 | 298 | | return; |
| | | 299 | | } |
| | 2 | 300 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.PreconditionFailed) |
| | | 301 | | { |
| | 2 | 302 | | } |
| | | 303 | | } |
| | 3 | 304 | | } |
| | | 305 | | |
| | | 306 | | public async Task<bool> TryDeleteAsync(string flowId, CancellationToken cancellationToken = default) |
| | | 307 | | { |
| | 3 | 308 | | ArgumentException.ThrowIfNullOrWhiteSpace(flowId); |
| | 3 | 309 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | | 310 | | |
| | | 311 | | try |
| | | 312 | | { |
| | 3 | 313 | | await container.DeleteItemAsync<CosmosFlowStateDocument>( |
| | 3 | 314 | | flowId, |
| | 3 | 315 | | new PartitionKey(flowId), |
| | 3 | 316 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 317 | | return true; |
| | | 318 | | } |
| | 3 | 319 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.NotFound) |
| | | 320 | | { |
| | 3 | 321 | | return false; |
| | | 322 | | } |
| | 3 | 323 | | } |
| | | 324 | | |
| | | 325 | | private async Task<Container> GetContainerAsync(CancellationToken cancellationToken) |
| | | 326 | | { |
| | 3 | 327 | | if (!_created) |
| | 3 | 328 | | await EnsureCreatedAsync(cancellationToken).ConfigureAwait(false); |
| | | 329 | | |
| | 3 | 330 | | return _client.GetContainer(_options.DatabaseName, _options.ContainerName); |
| | 3 | 331 | | } |
| | | 332 | | |
| | | 333 | | private async Task EnsureCreatedAsync(CancellationToken cancellationToken) |
| | | 334 | | { |
| | 3 | 335 | | if (_created) |
| | 2 | 336 | | return; |
| | | 337 | | |
| | 3 | 338 | | await _ensureGate.WaitAsync(cancellationToken).ConfigureAwait(false); |
| | | 339 | | try |
| | | 340 | | { |
| | 3 | 341 | | if (_created) |
| | 2 | 342 | | return; |
| | | 343 | | |
| | | 344 | | ContainerResponse container; |
| | 3 | 345 | | if (_options.AutoCreateContainer) |
| | | 346 | | { |
| | 3 | 347 | | var database = await _client.CreateDatabaseIfNotExistsAsync( |
| | 3 | 348 | | _options.DatabaseName, |
| | 3 | 349 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | | 350 | | |
| | | 351 | | // DefaultTimeToLive = -1 enables per-item TTL without a container-wide default. |
| | 3 | 352 | | var properties = new ContainerProperties(_options.ContainerName, _options.PartitionKeyPath) |
| | 3 | 353 | | { |
| | 3 | 354 | | DefaultTimeToLive = -1 |
| | 3 | 355 | | }; |
| | 3 | 356 | | container = await database.Database.CreateContainerIfNotExistsAsync( |
| | 3 | 357 | | properties, |
| | 3 | 358 | | _options.Throughput, |
| | 3 | 359 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | | 360 | | } |
| | | 361 | | else |
| | | 362 | | { |
| | 3 | 363 | | container = await _client |
| | 3 | 364 | | .GetContainer(_options.DatabaseName, _options.ContainerName) |
| | 3 | 365 | | .ReadContainerAsync(cancellationToken: cancellationToken) |
| | 3 | 366 | | .ConfigureAwait(false); |
| | | 367 | | } |
| | | 368 | | |
| | 3 | 369 | | if (!string.Equals(container.Resource.PartitionKeyPath, _options.PartitionKeyPath, StringComparison.Ordinal) |
| | 2 | 370 | | throw new InvalidOperationException( |
| | 2 | 371 | | $"Cosmos container '{_options.ContainerName}' uses partition key '{container.Resource.PartitionKeyPa |
| | 2 | 372 | | $"but '{_options.PartitionKeyPath}' is required."); |
| | 3 | 373 | | if (container.Resource.DefaultTimeToLive is null) |
| | 3 | 374 | | throw new InvalidOperationException( |
| | 3 | 375 | | $"Cosmos container '{_options.ContainerName}' does not have TTL enabled. " + |
| | 3 | 376 | | "Enable container TTL (DefaultTimeToLive = -1) before using it for durable flows."); |
| | | 377 | | |
| | 3 | 378 | | ValidateHostSerializer(); |
| | 3 | 379 | | _created = true; |
| | 3 | 380 | | } |
| | | 381 | | finally |
| | | 382 | | { |
| | 3 | 383 | | _ensureGate.Release(); |
| | | 384 | | } |
| | 3 | 385 | | } |
| | | 386 | | |
| | | 387 | | /// <summary> |
| | | 388 | | /// Fails provisioning fast when a host-registered <see cref="CosmosClient"/> carries a custom |
| | | 389 | | /// serializer that does not honor the flow-state document's JSON property names. |
| | | 390 | | /// <see cref="CosmosFlowStateDocument"/> is attributed for both Newtonsoft.Json |
| | | 391 | | /// (<c>[JsonProperty]</c>) and System.Text.Json (<c>[JsonPropertyName]</c>), so the SDK's |
| | | 392 | | /// default serializer and STJ-based serializers both map correctly; a serializer honoring |
| | | 393 | | /// neither would write documents whose <c>id</c> Cosmos rejects — or, worse, whose fields |
| | | 394 | | /// silently round-trip as nulls. The probe serializes a sentinel document through the host's |
| | | 395 | | /// serializer and verifies the wire property names survive. |
| | | 396 | | /// </summary> |
| | | 397 | | private void ValidateHostSerializer() |
| | | 398 | | { |
| | 3 | 399 | | if (_client.ClientOptions?.Serializer is not { } serializer) |
| | 2 | 400 | | return; // The SDK default serializer honors [JsonProperty]; nothing to probe. |
| | | 401 | | |
| | 1 | 402 | | var now = DateTime.UtcNow; |
| | 1 | 403 | | using var stream = serializer.ToStream(new CosmosFlowStateDocument |
| | 1 | 404 | | { |
| | 1 | 405 | | Id = "asyncresponse-serializer-probe", |
| | 1 | 406 | | FlowId = "asyncresponse-serializer-probe", |
| | 1 | 407 | | StateJson = "{}", |
| | 1 | 408 | | ExpiresAtUtc = now, |
| | 1 | 409 | | UpdatedAtUtc = now, |
| | 1 | 410 | | Revision = 0, |
| | 1 | 411 | | LeaseId = "probe", |
| | 1 | 412 | | LeaseExpiresAtUtc = now, |
| | 1 | 413 | | Ttl = 1 |
| | 1 | 414 | | }); |
| | 1 | 415 | | using var probe = System.Text.Json.JsonDocument.Parse(stream); |
| | 1 | 416 | | if (!probe.RootElement.TryGetProperty("id", out _) || !probe.RootElement.TryGetProperty("leaseExpiresAtUtc", out |
| | | 417 | | { |
| | 0 | 418 | | throw new InvalidOperationException( |
| | 0 | 419 | | $"The registered CosmosClient's serializer ({serializer.GetType().Name}) does not honor the durable-flow |
| | 0 | 420 | | "JSON property names ('id', 'flowId', 'leaseExpiresAtUtc', ...). Flow-state documents would be written w |
| | 0 | 421 | | "property names and could not be read back. Configure the serializer to honor Newtonsoft.Json [JsonPrope |
| | 0 | 422 | | "System.Text.Json [JsonPropertyName] attributes, or let the SDK use its default serializer."); |
| | | 423 | | } |
| | 1 | 424 | | } |
| | | 425 | | |
| | | 426 | | private async Task<bool> UpdateLeaseAsync( |
| | | 427 | | string flowId, |
| | | 428 | | string leaseId, |
| | | 429 | | TimeSpan leaseDuration, |
| | | 430 | | bool acquire, |
| | | 431 | | CancellationToken cancellationToken) |
| | | 432 | | { |
| | 3 | 433 | | ArgumentException.ThrowIfNullOrWhiteSpace(flowId); |
| | 3 | 434 | | ArgumentException.ThrowIfNullOrWhiteSpace(leaseId); |
| | 3 | 435 | | if (leaseDuration <= TimeSpan.Zero) |
| | 2 | 436 | | throw new ArgumentOutOfRangeException(nameof(leaseDuration)); |
| | | 437 | | |
| | 3 | 438 | | var container = await GetContainerAsync(cancellationToken).ConfigureAwait(false); |
| | 3 | 439 | | for (var attempt = 0; attempt < 4; attempt++) |
| | | 440 | | { |
| | 3 | 441 | | var now = DateTime.UtcNow; |
| | | 442 | | try |
| | | 443 | | { |
| | 3 | 444 | | var current = await container.ReadItemAsync<CosmosFlowStateDocument>( |
| | 3 | 445 | | flowId, |
| | 3 | 446 | | new PartitionKey(flowId), |
| | 3 | 447 | | cancellationToken: cancellationToken).ConfigureAwait(false); |
| | 3 | 448 | | var document = current.Resource; |
| | 3 | 449 | | if (document.ExpiresAtUtc <= now || document.Revision is null) |
| | 2 | 450 | | return false; |
| | 3 | 451 | | if (acquire) |
| | | 452 | | { |
| | 3 | 453 | | if (document.LeaseId is not null && document.LeaseId != leaseId && document.LeaseExpiresAtUtc > now) |
| | 3 | 454 | | return false; |
| | | 455 | | } |
| | 3 | 456 | | else if (document.LeaseId != leaseId || document.LeaseExpiresAtUtc <= now) |
| | | 457 | | { |
| | 3 | 458 | | return false; |
| | | 459 | | } |
| | | 460 | | |
| | 3 | 461 | | document.LeaseId = leaseId; |
| | 3 | 462 | | document.LeaseExpiresAtUtc = DurableFlowStoreShared.AddSaturating(now, leaseDuration); |
| | | 463 | | // Same _ts realignment as ReleaseLeaseAsync: a lease heartbeat replaces the |
| | | 464 | | // document without moving ExpiresAtUtc, so it must not restart the server TTL's |
| | | 465 | | // full retention window. |
| | 3 | 466 | | document.Ttl = CosmosTtlSeconds(document.ExpiresAtUtc, now); |
| | 3 | 467 | | await container.ReplaceItemAsync( |
| | 3 | 468 | | document, |
| | 3 | 469 | | flowId, |
| | 3 | 470 | | new PartitionKey(flowId), |
| | 3 | 471 | | new ItemRequestOptions { IfMatchEtag = current.ETag }, |
| | 3 | 472 | | cancellationToken).ConfigureAwait(false); |
| | 3 | 473 | | return true; |
| | | 474 | | } |
| | 2 | 475 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.NotFound) |
| | | 476 | | { |
| | 2 | 477 | | return false; |
| | | 478 | | } |
| | 2 | 479 | | catch (CosmosException ex) when (ex.StatusCode == HttpStatusCode.PreconditionFailed) |
| | | 480 | | { |
| | 2 | 481 | | } |
| | | 482 | | } |
| | | 483 | | |
| | 2 | 484 | | return false; |
| | 3 | 485 | | } |
| | | 486 | | |
| | | 487 | | private static CosmosFlowStateDocument CreateDocument(string flowId, string stateJson, long revision, TimeSpan ttl, |
| | 3 | 488 | | => new() |
| | 3 | 489 | | { |
| | 3 | 490 | | Id = flowId, |
| | 3 | 491 | | FlowId = flowId, |
| | 3 | 492 | | StateJson = stateJson, |
| | 3 | 493 | | ExpiresAtUtc = DurableFlowStoreShared.AddSaturating(now, ttl), |
| | 3 | 494 | | UpdatedAtUtc = now, |
| | 3 | 495 | | Revision = revision, |
| | 3 | 496 | | // Cosmos reaps the item itself once container TTL is enabled. Ceiling keeps the |
| | 3 | 497 | | // server-side TTL from being shorter than the requested duration. |
| | 3 | 498 | | Ttl = CosmosTtlSeconds(ttl) |
| | 3 | 499 | | }; |
| | | 500 | | |
| | | 501 | | /// <summary>Per-item TTL in whole seconds, rounded up and saturated at int.MaxValue (~68 years) for absurd expiries |
| | | 502 | | private static int CosmosTtlSeconds(TimeSpan ttl) |
| | 3 | 503 | | => (int)Math.Min(Math.Ceiling(ttl.TotalSeconds), int.MaxValue); |
| | | 504 | | |
| | | 505 | | /// <summary> |
| | | 506 | | /// Remaining per-item TTL in whole seconds until <paramref name="expiresAtUtc"/>, rounded up |
| | | 507 | | /// and floored at 1 (Cosmos rejects 0). Used by replaces that keep the logical expiry in place: |
| | | 508 | | /// an already-due document collapses to the shortest legal TTL so the next sweep purges it |
| | | 509 | | /// instead of the replace granting it a fresh retention window. |
| | | 510 | | /// </summary> |
| | | 511 | | private static int CosmosTtlSeconds(DateTime expiresAtUtc, DateTime now) |
| | 3 | 512 | | => (int)Math.Min(Math.Max(Math.Ceiling((expiresAtUtc - now).TotalSeconds), 1), int.MaxValue); |
| | | 513 | | |
| | | 514 | | /// <summary>Disposes the Cosmos client when the store created (and therefore owns) it.</summary> |
| | | 515 | | public void Dispose() |
| | | 516 | | { |
| | 2 | 517 | | _ensureGate.Dispose(); |
| | 2 | 518 | | if (_ownsClient) |
| | 2 | 519 | | _client.Dispose(); |
| | 2 | 520 | | } |
| | | 521 | | } |
| | | 522 | | |
| | | 523 | | /// <summary> |
| | | 524 | | /// One durable-flow ledger document. Attributed for BOTH Newtonsoft.Json and System.Text.Json: |
| | | 525 | | /// the Cosmos SDK's default serializer is Newtonsoft-based, but hosts may register a |
| | | 526 | | /// <see cref="CosmosClient"/> with an STJ-based serializer — with single-stack attributes such a |
| | | 527 | | /// client would silently write PascalCase property names (breaking <c>id</c> and every read |
| | | 528 | | /// back). <see cref="CosmosFlowStateStore"/> additionally probes custom serializers at |
| | | 529 | | /// provisioning time and fails fast when neither attribute set is honored. |
| | | 530 | | /// </summary> |
| | | 531 | | internal sealed class CosmosFlowStateDocument |
| | | 532 | | { |
| | | 533 | | [JsonProperty("id")] |
| | | 534 | | [System.Text.Json.Serialization.JsonPropertyName("id")] |
| | | 535 | | public string Id { get; set; } = ""; |
| | | 536 | | |
| | | 537 | | [JsonProperty("flowId")] |
| | | 538 | | [System.Text.Json.Serialization.JsonPropertyName("flowId")] |
| | | 539 | | public string FlowId { get; set; } = ""; |
| | | 540 | | |
| | | 541 | | [JsonProperty("stateJson")] |
| | | 542 | | [System.Text.Json.Serialization.JsonPropertyName("stateJson")] |
| | | 543 | | public string StateJson { get; set; } = ""; |
| | | 544 | | |
| | | 545 | | [JsonProperty("expiresAtUtc")] |
| | | 546 | | [System.Text.Json.Serialization.JsonPropertyName("expiresAtUtc")] |
| | | 547 | | public DateTime ExpiresAtUtc { get; set; } |
| | | 548 | | |
| | | 549 | | [JsonProperty("updatedAtUtc")] |
| | | 550 | | [System.Text.Json.Serialization.JsonPropertyName("updatedAtUtc")] |
| | | 551 | | public DateTime UpdatedAtUtc { get; set; } |
| | | 552 | | |
| | | 553 | | [JsonProperty("revision")] |
| | | 554 | | [System.Text.Json.Serialization.JsonPropertyName("revision")] |
| | | 555 | | public long? Revision { get; set; } |
| | | 556 | | |
| | | 557 | | [JsonProperty("leaseId", NullValueHandling = NullValueHandling.Ignore)] |
| | | 558 | | [System.Text.Json.Serialization.JsonPropertyName("leaseId")] |
| | | 559 | | [System.Text.Json.Serialization.JsonIgnore(Condition = System.Text.Json.Serialization.JsonIgnoreCondition.WhenWritin |
| | | 560 | | public string? LeaseId { get; set; } |
| | | 561 | | |
| | | 562 | | [JsonProperty("leaseExpiresAtUtc", NullValueHandling = NullValueHandling.Ignore)] |
| | | 563 | | [System.Text.Json.Serialization.JsonPropertyName("leaseExpiresAtUtc")] |
| | | 564 | | [System.Text.Json.Serialization.JsonIgnore(Condition = System.Text.Json.Serialization.JsonIgnoreCondition.WhenWritin |
| | | 565 | | public DateTime? LeaseExpiresAtUtc { get; set; } |
| | | 566 | | |
| | | 567 | | /// <summary>Cosmos per-item TTL in seconds; honored once the container enables TTL.</summary> |
| | | 568 | | [JsonProperty("ttl", NullValueHandling = NullValueHandling.Ignore)] |
| | | 569 | | [System.Text.Json.Serialization.JsonPropertyName("ttl")] |
| | | 570 | | [System.Text.Json.Serialization.JsonIgnore(Condition = System.Text.Json.Serialization.JsonIgnoreCondition.WhenWritin |
| | | 571 | | public int? Ttl { get; set; } |
| | | 572 | | } |
| | | 573 | | } |