LangGraph run cancellation works locally but returns 404 in production

LangSmith Product Help Deployment
,
1

I’m experiencing an issue with the LangGraph run cancellation endpoint in my production environment. When I call

client.runs.cancel(threadId, runId)

in local development, it correctly processes the request with a 204 status code, but it does have an interesting warning message about cancelling “non-pending” runs:

info:    ▪ <-- POST /threads/96e219b8-2e54-4708-8ecd-f840170dd1de/runs/bca9c0fa-cbc5-449e-b75f-e399328602cb/cancel?wait=0&action=interrupt
warn:    ┏ Attempted to cancel non-pending run.
warn:    ┗ [1] { run_id: 'bca9c0fa-cbc5-449e-b75f-e399328602cb', status: 'error' }
info:    ┏ Cancelled runs
info:    ┃ [1] {
info:    ┃ [2]   run_ids: [ 'bca9c0fa-cbc5-449e-b75f-e399328602cb' ],
info:    ┃ [3]   thread_id: '96e219b8-2e54-4708-8ecd-f840170dd1de',
info:    ┃ [4]   action: 'interrupt'
info:    ┗ [5] }
info:    ▪ --> POST /threads/96e219b8-2e54-4708-8ecd-f840170dd1de/runs/bca9c0fa-cbc5-449e-b75f-e399328602cb/cancel?wait=0&action=interrupt 204

However, in production, the same code results in a 404 error:

[INFO] POST /threads/befa782a-1585-449a-b9b2-0afa8a31c538/runs/1f066639-ef80-693c-815c-3879748b8c87/cancel 404 16ms

With the error message: “HTTP 404: Some runs were not found or not cancelable”

I’ve verified that I can successfully retrieve the run information using the GET endpoint

/threads/{thread_id}/runs/{run_id}

so the run definitely exists in the system. This suggests there might be a difference in how the cancellation endpoint behaves between local and production environments.

Any insights on why this might be happening or how to resolve it would be greatly appreciated.

2

The 404 error “Some runs were not found or not cancelable” suggests timing or state differences between environments. In production, the run might have already completed/failed by the time you try to cancel it, while in local development it’s still in a cancelable state when you make the call.

Check the run status before canceling:

run = client.runs.get(threadId, runId)
if run.status in ['pending', 'running']:
    client.runs.cancel(threadId, runId)

The warning in local about “non-pending run” indicates you’re trying to cancel completed runs there too, but the local environment is more permissive. Production likely has stricter validation that only allows canceling active runs.

3

I’ve reproduced what you’re seeing, and it does appear to be a bug. I’ve written a test script that cancels a run. It works when running the in-memory server implementation used by langgraph dev. It does, however, show the warning about cancelling non-pending runs that you are seeing locally.

When I run the same script against the containerized implementation using langgraph up, I experience the 404 that you are seeing.

As a meta-point, langgraph up is a faster and easier way to test the full dockerized server implementation as part of a dev workflow than deploying to LangGraph Platform. In any case, we will get back to you soon about a fix for this!

1 Like
4

Okay, there was no bug on the production side. The in-memory server that runs on langgraph dev should have been failing all along, as you are attempting to cancel a run that is no longer in a cancellable state (it is already completed).

We’ve patched the in-memory server implementation. Version 0.6.1 reflects this new behavior.