@Marvin

Encountered exception during execution: RuntimeError('Cannot send a request, as the client has been closed.')
Traceback (most recent call last):
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/flow_engine.py", line 1316, in run_context
    yield self
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/flow_engine.py", line 1378, in run_flow_async
    await engine.call_flow_fn()
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/flow_engine.py", line 1331, in call_flow_fn
    await self.handle_success(result)
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/flow_engine.py", line 923, in handle_success
    await self.set_state(terminal_state)
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/flow_engine.py", line 873, in set_state
...
  File "/home/worker/.local/lib/python3.10/site-packages/httpx/_client.py", line 1585, in request
    return await self.send(request, auth=auth, follow_redirects=follow_redirects)
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/client/base.py", line 324, in send
    response = await self._send_with_retry(
  File "/home/worker/.local/lib/python3.10/site-packages/prefect/client/base.py", line 248, in _send_with_retry
    response = await send(request, *send_args, **send_kwargs)
  File "/home/worker/.local/lib/python3.10/site-packages/httpx/_client.py", line 1661, in send
    raise RuntimeError("Cannot send a request, as the client has been closed.")
RuntimeError: Cannot send a request, as the client has been closed.

The error message "Cannot send a request, as the client has been closed" usually occurs when an HTTP client is used after it has been closed, often due to improper management of the client's lifecycle in asynchronous contexts. Here are some potential solutions and insights based on the gathered information: 1. Ensure Proper Client Management: When using asynchronous clients, such as the one provided by

httpx

, make sure to manage the client's lifecycle properly. Use the client within a context manager (

async with

) to ensure it is opened and closed correctly. For example:

python
   async with get_client() as client:
       # perform operations with client

This pattern ensures that the client is properly closed after its operations are complete. 2. Check for Premature Closure: Verify that the client isn't being closed prematurely in your code. Ensure that all operations requiring the client are completed before exiting the context manager. 3. Review Your Flow Code: The issue might be related to the flow's structure or how asynchronous operations are handled. Carefully review the flow code to ensure that asynchronous operations are awaited properly and the client isn't used outside its intended scope. 4. Update Prefect: Make sure you are using the latest version of Prefect. Sometimes, these issues are resolved in newer releases. You can check the latest updates and fixes in the release notes. 5. Review Related Issues: There might be similar issues reported by other users. For example, this GitHub issue discusses a related problem with client management. If these steps don't resolve the issue, consider reaching out to Prefect support or checking the Prefect documentation on troubleshooting for further guidance. If you need more specific guidance, feel free to share more details about your flow setup or any additional error logs.

hi @Ji-Oh Yoo! I've seen this one before, are you passing

client

into something? like

Block.load

or something like that

Hi, thanks for quick response. This is happening on the job worker side; after the flow is finished it seems it can't update the status because of that error.

what type of worker are you using?

not sure if this is just network issue or if the client is closed then if it is supposed to recreate a client

hrmm yeah its usually something like

def thing_needing_client(*, client):
  # closes client somewhere here

client = get_client()
thing_needing_client(..., client=client)
client.something() # raises

maybe its in your worker impl if you're using a custom worker?

yeah it could be in some impl of the custom worker. I don't think we're manually retrieving the client object and closing the client though