a run is active - exception (c#, open AI assistant)

[JP-droidic]

Hi team,

not sure if anyone else has noticed this, but we are getting more and more often the following exceptions:

System.ClientModel.ClientResultException: HTTP 400 (invalid_request_error: ) 
Can't add messages to thread_8vvUmiYP5yNpmkwL2DF01Usj while a run run_xq3NOfM700rAyoJJ6tINKzFM is active. 
at OpenAI.ClientPipelineExtensions.ProcessMessageAsync(ClientPipeline pipeline, PipelineMessage message, RequestOptions options) 
at OpenAI.Assistants.InternalAssistantMessageClient.CreateMessageAsync(String threadId, BinaryContent content, RequestOptions options) 
at OpenAI.Assistants.AssistantClient.CreateMessageAsync(String threadId, MessageRole role, IEnumerable`1 content, MessageCreationOptions options, CancellationToken cancellationToken) 
at Microsoft.SemanticKernel.Agents.OpenAI.Internal.AssistantThreadActions.CreateMessageAsync(AssistantClient client, String threadId, ChatMessageContent message, CancellationToken cancellationToken) ]
at ...

The scenario is,

a user asks a question (new chat), we create an assistant and start a thread.
using streaming result is shown to the user
user asks a second question
we get the error message

This is completely random, I would say one in 7 chats gets the error message

[crickman]

HI @JP-droidic - On the surface this seems to indicate the perhaps the user message is being added while the run is still active. I might double check any potential concurrency issues and also verify that the all of the (IAsyncEnumerable) streamed responses have been processed.

Also, can you confirm whether you are targeting the Open AI endpoint or an Azure AI endpoint?

In the meantime, I'll setup a local stress test to explore this dynamic further. If I am able to reproduce this locally, it will aid in understanding what may be at play here.

[JP-droidic]

It is very strange indeed. LLM stopped returning results(stream) maybe even a minute after the second message was sent. We use Open AI endpoint.

Thank!

[crickman]

Review the API docs and saw this confirmation on what is going on:

https://platform.openai.com/docs/assistants/deep-dive#thread-locks

The error message is pretty clear. I suppose the question is why is the run active? I reviewed the code and termination states. My only thought on how a run might be active is if an exception was thrown; although, I'd expect such a failure to bubble up into your call-stack. I'll keep looking, but can you also please examine logs to acertain if perhap an exception has occurred? If so, this will help guide the fix.