संपादित करें

इसके माध्यम से साझा किया गया


Diagnostic functionality for Stateful Reliable Services

The Azure Service Fabric Stateful Reliable Services StatefulServiceBase class emits EventSource events that can be used to debug the service, provide insights into how the runtime is operating, and help with troubleshooting.

EventSource events

The EventSource name for the Stateful Reliable Services StatefulServiceBase class is "Microsoft-ServiceFabric-Services." Events from this event source appear in the Diagnostics Events window when the service is being debugged in Visual Studio.

Examples of tools and technologies that help in collecting and/or viewing EventSource events are PerfView, Azure Diagnostics, and the Microsoft TraceEvent Library.

Events

Event name Event ID Level Event description
StatefulRunAsyncInvocation 1 Informational Emitted when the service RunAsync task is started
StatefulRunAsyncCancellation 2 Informational Emitted when the service RunAsync task is canceled
StatefulRunAsyncCompletion 3 Informational Emitted when the service RunAsync task is finished
StatefulRunAsyncSlowCancellation 4 Warning Emitted when the service RunAsync task takes too long to complete cancellation
StatefulRunAsyncFailure 5 Error Emitted when the service RunAsync task throws an exception

Interpret events

StatefulRunAsyncInvocation, StatefulRunAsyncCompletion, and StatefulRunAsyncCancellation events are useful to the service writer to understand the lifecycle of a service, as well as the timing for when a service starts, cancels, or finishes. This information can be useful when debugging service issues or understanding the service lifecycle.

Service writers should pay close attention to StatefulRunAsyncSlowCancellation and StatefulRunAsyncFailure events because they indicate issues with the service.

StatefulRunAsyncFailure is emitted whenever the service RunAsync() task throws an exception. Typically, an exception thrown indicates an error or bug in the service. Additionally, the exception causes the service to fail, so it is moved to a different node. This operation can be expensive and can delay incoming requests while the service is moved. Service writers should determine the cause of the exception and, if possible, mitigate it.

StatefulRunAsyncSlowCancellation is emitted whenever a cancellation request for the RunAsync task takes longer than four seconds. When a service takes too long to complete cancellation, it affects the ability of the service to be quickly restarted on another node. This scenario might affect the overall availability of the service.

Performance counters

The Reliable Services runtime defines the following performance counter categories:

Category Description
Service Fabric Transactional Replicator Counters specific to the Azure Service Fabric Transactional Replicator
Service Fabric TStore Counters specific to the Azure Service Fabric TStore

The Service Fabric Transactional Replicator is used by the Reliable State Manager to replicate transactions within a given set of replicas.

The Service Fabric TStore is a component used in Reliable Collections for storing and retrieving key-value pairs.

The Windows Performance Monitor application that is available by default in the Windows operating system can be used to collect and view performance counter data. Azure Diagnostics is another option for collecting performance counter data and uploading it to Azure tables.

Performance counter instance names

A cluster that has a large number of reliable services or reliable service partitions will have a large number of transactional replicator performance counter instances. This is also the case for TStore performance counters, but is also multiplied by the number of Reliable Dictionaries and Reliable Queues used. The performance counter instance names can help in identifying the specific partition, service replica, and state provider in the case of TStore, that the performance counter instance is associated with.

Service Fabric Transactional Replicator category

For the category Service Fabric Transactional Replicator, the counter instance names are in the following format:

ServiceFabricPartitionId:ServiceFabricReplicaId

ServiceFabricPartitionId is the string representation of the Service Fabric partition ID that the performance counter instance is associated with. The partition ID is a GUID, and its string representation is generated through Guid.ToString with format specifier "D".

ServiceFabricReplicaId is the ID associated with a given replica of a reliable service. The replica ID is included in the performance counter instance name to ensure its uniqueness and avoid conflict with other performance counter instances generated by the same partition. Further details about replicas and their role in reliable services can be found here.

The following counter instance name is typical for a counter under the Service Fabric Transactional Replicator category:

00d0126d-3e36-4d68-98da-cc4f7195d85e:131652217797162571

In the preceding example, 00d0126d-3e36-4d68-98da-cc4f7195d85e is the string representation of the Service Fabric partition ID, and 131652217797162571 is the replica ID.

Service Fabric TStore category

For the category Service Fabric TStore, the counter instance names are in the following format:

ServiceFabricPartitionId:ServiceFabricReplicaId:StateProviderId_PerformanceCounterInstanceDifferentiator_StateProviderName

ServiceFabricPartitionId is the string representation of the Service Fabric partition ID that the performance counter instance is associated with. The partition ID is a GUID, and its string representation is generated through Guid.ToString with format specifier "D".

ServiceFabricReplicaId is the ID associated with a given replica of a reliable service. The replica ID is included in the performance counter instance name to ensure its uniqueness and avoid conflict with other performance counter instances generated by the same partition. Further details about replicas and their role in reliable services can be found here.

StateProviderId is the ID associated with a state provider within a reliable service. The state provider ID is included in the performance counter instance name to differentiate a TStore from another.

PerformanceCounterInstanceDifferentiator is a differentiating ID associated with a performance counter instance within a state provider. This differentiator is included in the performance counter instance name to ensure its uniqueness and avoid conflict with other performance counter instances generated by the same state provider.

StateProviderName is the name associated with a state provider within a reliable service. The state provider name is included in the performance counter instance name for users to easily identify what state it provides.

The following counter instance name is typical for a counter under the Service Fabric TStore category:

00d0126d-3e36-4d68-98da-cc4f7195d85e:131652217797162571:142652217797162571_1337_urn:MyReliableDictionary/dataStore

In the preceding example, 00d0126d-3e36-4d68-98da-cc4f7195d85e is the string representation of the Service Fabric partition ID, 131652217797162571 is the replica ID, 142652217797162571 is the state provider ID, and 1337 is the performance counter instance differentiator. urn:MyReliableDictionary/dataStore is the name of the state provider that stores data for the collection named urn:MyReliableDictionary.

Transactional Replicator performance counters

The Reliable Services runtime emits the following events under the Service Fabric Transactional Replicator category

Counter name Description
Begin Txn Operations/sec The number of new write transactions created per second.
Txn Operations/sec The number of add/update/delete operations performed on reliable collections per second.
Log Flush Bytes/sec The number of bytes being flushed to the disk by the Transactional Replicator per second
Throttled Operations/sec The number of operations rejected every second by the Transactional Replicator due to throttling.
Avg. Transaction ms/Commit Average commit latency per transaction in milliseconds
Avg. Flush Latency (ms) Average duration of disk flush operations initiated by the Transactional Replicator in milliseconds

TStore performance counters

The Reliable Services runtime emits the following events under the Service Fabric TStore category

Counter name Description
Item Count The number of items in the store.
Disk Size The total disk size, in bytes, of checkpoint files for the store.
Checkpoint File Write Bytes/sec The number of bytes written per second for the most recent checkpoint file.
Copy Disk Transfer Bytes/sec The number of disk bytes read (on the primary replica) or written (on a secondary replica) per second during a store copy.

Next steps

EventSource providers in PerfView