fix: Change retry logic to extract next run from traceback

[adinhodovic]

task_meta.result is an Exception for me with Celery v5.2.5, I can't grasp where result.when would come from. The retry logic simply does not work as the result is an Exception and does not have any fields indicating when the next run is. Accessing when just crashes the progress bar and returns a 500 error.

Still kept it in and checked if it's iterable, maybe it works for others.

Here's a PR that works for me, if there's other solutions let me know. But atm retries are causing 500 errors.

[czue]

Thanks for the contribution!

I've never actually used the retry functionality. @EJH2 or @OmarWKH - do you have any thoughts on this?

Or @adinhodovic do you have a quick way to reproduce the scenario on the celery side so I can play with it?

[adinhodovic]

Thanks for the contribution!

I've never actually used the retry functionality. @EJH2 or @OmarWKH - do you have any thoughts on this?

Or @adinhodovic do you have a quick way to reproduce the scenario on the celery side so I can play with it?

from celery import Task
from celery_progress.backend import ProgressRecorder

class BaseTask(Task):
    autoretry_for = (Exception,)
    max_retries = 5
    retry_backoff = 300

@app.task(
    base=BaseTask, bind=True
)
def my_test(self):
    progress_recorder = ProgressRecorder(self)
    progress_recorder.set_progress(0, 100, description="Starting")
    raise Exception("Test")

We run retries on all tasks so it becomes core. Internal server errors on any task that is retrying as the API crashes.

task_meta.result is an exception on retries, accessing when on the result just fails.

[czue]

Thanks again for reporting and for the repro steps. I played with it and made some tweaks to your original PR in #112

Updates:

• Always assume result is an exception for RETRY states
• Remove seconds to date conversions (I ran into some timezone issues with this)

If you can confirm it works for you I'll close this and merge that one.