Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Prefect Community

Hi team - I am trying to run a local agent from a virtualenv alongside the new server, but when I try and trigger a flow from the UI, my agent reports `[2020-04-07 22:52:35,213] ERROR - agent | Error while deploying flow: FileNotFoundError(2, "No such file or directory: 'prefect'")`

I tried dialling up the logging but found nothing more

Hey <@ULQLBDKSP>!  I’ve seen this error before whenever one runs the agent by directly sourcing from a virtual env like `/.../venv/prefect_env_name/prefect agent start`

You need to make sure you actually _activate_ the prefect virtual environment, as the agent will pass down the environment that it is running with

that’s done it - thank you <@UKNSNMUE6> for your incredibly timely response

One more <@UKNSNMUE6> - I’m trying to use a local agent with some local flows using the server

```[2020-04-08 04:24:50,068] ERROR - prefect.CloudTaskRunner | Unexpected error: KeyError('state')
Traceback (most recent call last):
  File "/venv/lib/python3.7/site-packages/prefect/engine/runner.py", line 48, in inner
    new_state = method(self, state, *args, **kwargs)
  File "/venv/lib/python3.7/site-packages/prefect/engine/cloud/task_runner.py", line 226, in check_task_is_cached
    for key, res in (candidate_state.cached_inputs or {}).items()
  File "/venv/lib/python3.7/site-packages/prefect/engine/cloud/task_runner.py", line 226, in &lt;dictcomp&gt;
    for key, res in (candidate_state.cached_inputs or {}).items()
KeyError: 'state'```

is it expected that my local agent would be using the CloudTaskRunner?

and I would actually like to use my own subclass of TaskRunner that includes some caching, is this config something I can set with environment variables on the local agent? (via `PREFECT__ENGINE__TASK_RUNNER__DEFAULT_CLASS` ?)

I’m honestly not sure where that `KeyError` is coming from, I’d need to look deeper to better understand it; for changing the default task runner class, yes you can do that!  You’ll need to start your agent like so:
```prefect agent start -e PREFECT__ENGINE__TASK_RUNNER__DEFAULT_CLASS="full.importable.path.to.your.runner.class"```
but note that you’ll want to subclass the `CloudTaskRunner` as it knows how to update states through the Prefect API

Cheers - I think something funky was going on in my subclass. All resolved now - thanks <@UKNSNMUE6>