Question: best practice for multi-process/multi-host logging

[stefanfoulis]

For the case where a web request triggers a task on a queue to later execute: is it ok if the eliot action originally in the request has "finished" already, but the async task running in another process still adds messages to the sub-action?

Pseudocode (Django-ish, Celery-ish):

# in the web process handling requests
def a_view(request):
    with eliot.start_action('my-web-request') as action:
        my_background_task.delay(
            eliot_action_uuid=action.serialize_task_id().decode('ascii')
        )
    # at this point the action as already finished, but the code in
    # my_background_task has not run yet.

# in a worker process handling background tasks
def my_background_task(eliot_action_uuid):
    with eliot.Action.continue_task(task_id=eliot_action_uuid):
        # here we're logging stuff somewhere in the subtree of an 
        # action that is actually already "finished".
        do_stuff()

Is this ok? Or should this be avoided?

[itamarst]

This is OK, yes - it's one of the reasons Eliot does causal ordering rather than time-based ordering. It makes certain use cases (e.g. profiling) trickier, but it's an accurate representation of many situations.

This should probably be some sort of FAQ or documentation entry, so leaving issue open for that.

[stefanfoulis]

Thanks @itamarst that information helped me a lot in figuring out how to use eliot :-)