I'm trying to test out my error handling code when calling into Microsoft Azure Cognitive Services. I'm trying to make sure I handle any potential return error codes. Is there a way to simulate or force Cognitive services to return a specific error? Or am I going to have to mock cognitive services and send my error codes to my tests?
Actually as I known, there are only two kind of HTTP response status codes in MS Azure Cognitive Services: 20x (such as 200, 202) represent to success calling, and not 20x like 400, 415, 500 represent to failed calling with error.
So I think you just need to handle two potential return status codes: 20x and not 20x. Error is error, not for all error codes to make the related handle code, just record it with logging.
Of couse, there be the API references of all cognitive service https://westus2.dev.cognitive.microsoft.com/docs/services/, you can collect the error codes of response status code for all APIs of all services by yourself to build a mock service to test your code.
Related
Message Pact is non Http approach, see for more details:
https://docs.pact.io/getting_started/how_pact_works#non-http-testing-message-pact
ActiveSupport::Notification - is part of [Rails ActiveSupport][1], see for more details:
https://apidock.com/rails/ActiveSupport/Notifications
As I understand, ActiveSupport::Notification is using memory for a queue underneath, but not the external requests that are expected by Pact, so probably it can be done only via some other Message Queue, like Kafka, for example:
ActiveSupport::Notifications.subscribe("my_message") do |payload|
Kafka.produce(queue: 'test-queue', message: payload)
end
where Kafka.produce can be handled by Pact.
However, in this way, there is makes sense to remove ActiveSupport::Notifications and keep using only Kafka but this is the next step.
Note: Help with the immediate problem would be great, but mostly I'm looking for advice on troubleshooting gRPC timing issues in general (this isn't my first such issue).
I am adding a new server streaming service to a C++ module which has an existing server streaming service, and the two appear to be conflicting. Specifically, the completion queue Next() call on the server is crashing intermittently after the C# client calls Cancel() on the cancellation token for one of the services. This doesn't happen if I run each service independently.
On the client, I get this at the response stream MoveNext() call:
System.InvalidOperationException
HResult=0x80131509
Message=Shutdown has already been called
Source=Grpc.Core
StackTrace:
at Grpc.Core.Internal.CompletionQueueSafeHandle.BeginOp()
at Grpc.Core.Internal.CallSafeHandle.StartReceiveMessage(IReceivedMessageCallback callback)
at Grpc.Core.Internal.AsyncCallBase`2.ReadMessageInternalAsync()
at Grpc.Core.Internal.ClientResponseStream`2.<MoveNext>d__5.MoveNext()
at System.Runtime.ExceptionServices.ExceptionDispatchInfo.Throw()
at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
at System.Runtime.CompilerServices.TaskAwaiter`1.GetResult()
at MyModule.Connection.<DoSubscriptionReceives>d__7.MoveNext() in C:\snip\Connection.cs:line 67
On the server, I get this at the completion queue next() call:
Exception thrown: read access violation.
core_cq_tag->**** was 0xDDDDDDDD.
The stack trace:
MyModule.exe!grpc_impl::CompletionQueue::AsyncNextInternal(void * * tag, bool * ok, gpr_timespec deadline) Line 59 C++
> MyModule.exe!grpc_impl::CompletionQueue::Next(void * * tag, bool * ok) Line 176 C++
...snip...
It appears something is being added to the queue after shutdown. The difficulty is I have little visibility into what is being added into the queue and in what order.
I'm trying to write a server-side interceptor to log all requests & responses, but there seems to be no documentation. So far, poking through the API hasn't gotten me very far. Is there any documentation available on wiring up an interceptor in C++? Or, are there other approaches for troubleshooting timing conflicts between services?
Windows 11, Grpc.Core 1.27
What I've tried:
I first played with the GRPC_TRACE & GRPC_VERBOSITY environment variables. I was able to get some unhelpful output from the client, but nothing from the server. Of course, there's been lots of debugging, stripping the client & server down to barebones, disabling keep alives, ensuring we aren't using deadlines, having the services share a cancellation token, etc.
Update: I have found that the crash only happens when the client is run from an NUnit test. In that environment, the completion queue is getting more hits on Next(), but I'm still trying to figure out where they are coming from.
Is 1.27 the version you are using? That seems pretty old.. There might have been fixes since then.
For using the C++ server interception API, I think you would find this very useful - https://github.com/grpc/grpc/blob/0f2a0f5fc9b9e9b9c98d227d16575d106f1e8d43/test/cpp/end2end/server_interceptors_end2end_test.cc#L48
One suggestion I have is to run the code another sanitizers https://github.com/google/sanitizers to make sure that we don't have a heap-use-after-free type bug.
I would also check for API misuse issues. (If you had posted the code, I could've given a look to see if anything seems weird..)
I am working on Spring Kafka implementation and my use case is consume messages from Kafka topic as batch (using batch listener). when I consumer the list of messages, will iterate and call the REST endpoint for message enrichment. In case REST API fails for any runtime exception, I have implemented retry logic using spring retry. I want to stop the container, after the number of retries fails. So planning to use KafkaContainerStoppingErrorHandler to achieve this. Does the KafkaContainerStoppingErrorHandler commits the previous success messages - say if we receive 10 messages, and for message 1,2,3,4, enrichment call is success and for message 5 enrichment API call fails. so when we restart the container, will I get all 10 again or will I receive messages 5- 10?
or is there a way we can achieve above use case? I looked into all types of error handles of Spring kafka and need input on how to achieve above requirement.
You will get them all again.
You can use the DefaultErrorHandler (with a custom recoverer) and throw a BatchListenerFailedException to indicate which record in the batch failed.
The error handler will commit the offsets up to that record and call the recoverer with the failed record; in your custom recoverer you can stop the container (use the same logic as the container stopping error handler).
In versions before 2.8, this same functionality is provided by the RecoveringBatchErrorHandler.
From time to time, I see this error in Application Insights in Failures => failed dependencies :
Been searching through the documentation, but cannot see this mentioned anywhere. Does this status mean that the operation was cancelled through the token, or is it similar to the cross-partition response that used to be an 400 error? (https://github.com/Azure/azure-cosmos-dotnet-v2/issues/606#issuecomment-427909582)
Also, will this action be retried or is there loss of data for this?
We use the JS amaxon-dax-client (1.2.3) and we receive a bunch of
ERROR Failed to pull from <cluster>.cache.amazonaws.com (-.-.-.-): TimeoutError: Connection timeout after 10000ms
at SocketTubePool.alloc (/opt/nodejs/node_modules/amazon-dax-client/src/Tube.js:227:64)
at /opt/nodejs/node_modules/amazon-dax-client/generated-src/Operations.js:215:30
My understanding is this is a warning, not an error? We don't see any adverse effects (at least not noticeable) and the functions can access DAX just fine, but it pollutes our logs.
If the DAX endpoint isn't properly configured, the DB operations fail which isn't what we're seeing here.