azure-service-fabric - Service Fabric 有状态服务因 FailFast 而失败

Question

我在 Azure 集群中运行 Service Fabric 应用程序。该应用程序可以正常运行几天，没有任何问题。然而昨天我们注意到其中一项服务进入了“错误”状态并且此后一直没有恢复。

该服务是一种可靠的有状态服务，具有 1 个分区和 3 个副本(即一个主副本，两个辅助副本)。该服务实现长时间运行的 RunAsync()，并在 RunAsync() 期间同时读取和写入许多 ReliableDictionary 条目。

我们注意到主节点上的事件日志中存在以下错误:

Description: The application requested process termination through System.Environment.FailFast(string message).
Message: ProgressVectorEntry.Lsn == failureLsn
Stack:
   at System.Environment.FailFast(System.String)
   at Microsoft.ServiceFabric.Replicator.ProgressVector.FindSharedVector(Microsoft.ServiceFabric.Replicator.ProgressVector, Microsoft.ServiceFabric.Replicator.ProgressVector)
   at Microsoft.ServiceFabric.Replicator.ProgressVector.FindCopyModePrivate(Microsoft.ServiceFabric.Replicator.CopyContextParameters, Microsoft.ServiceFabric.Replicator.CopyContextParameters, Int64)
   at Microsoft.ServiceFabric.Replicator.ProgressVector.FindCopyMode(Microsoft.ServiceFabric.Replicator.CopyContextParameters, Microsoft.ServiceFabric.Replicator.CopyContextParameters, Int64)
   at Microsoft.ServiceFabric.Replicator.LoggingReplicator.GetLogRecordsToCopy(Microsoft.ServiceFabric.Replicator.ProgressVector, System.Fabric.Epoch, Microsoft.ServiceFabric.Replicator.LogicalSequenceNumber, Microsoft.ServiceFabric.Replicator.LogicalSequenceNumber, Int64, Int64, Microsoft.ServiceFabric.Replicator.LogicalSequenceNumber ByRef, Microsoft.ServiceFabric.Replicator.LogicalSequenceNumber ByRef, Microsoft.ServiceFabric.Data.IAsyncEnumerator`1<Microsoft.ServiceFabric.Replicator.LogRecord> ByRef, Microsoft.ServiceFabric.Replicator.BeginCheckpointLogRecord ByRef)
   at Microsoft.ServiceFabric.Replicator.LoggingReplicatorCopyStream+<GetNextAsyncSafe>d__3.MoveNext()
   at System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1[[System.__Canon, mscorlib, Version=4.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089]].Start[[Microsoft.ServiceFabric.Replicator.LoggingReplicatorCopyStream+<GetNextAsyncSafe>d__3, Microsoft.ServiceFabric.Data.Impl, Version=5.0.0.0, Culture=neutral, PublicKeyToken=31bf3856ad364e35]](<GetNextAsyncSafe>d__3 ByRef)
   at Microsoft.ServiceFabric.Replicator.LoggingReplicatorCopyStream.GetNextAsyncSafe(System.Threading.CancellationToken)
   at Microsoft.ServiceFabric.Replicator.LoggingReplicatorCopyStream.GetNextAsync(System.Threading.CancellationToken)
   at System.Fabric.StateProviderBroker+AsyncEnumerateOperationDataBroker.<BeginGetNext>b__8(System.Threading.CancellationToken)
   at System.Fabric.Interop.Utility.WrapNativeAsyncMethodImplementation(System.Func`2<System.Threading.CancellationToken,System.Threading.Tasks.Task>, IFabricAsyncOperationCallback, System.String, System.Fabric.Interop.InteropApi)
   at System.Fabric.StateProviderBroker+AsyncEnumerateOperationDataBroker.BeginGetNext(IFabricAsyncOperationCallback)

我们从未在本地开发环境中观察到这种情况，到目前为止，我们只在 Azure 集群中观察到这种情况。

1. 这里发生了什么？它看起来像损坏的复制信息。
2. 我们的代码在做什么会导致这个问题？

Answer 1

这是 Service Fabric 中的一个问题，已在 5.3.311 中修复.