Determine which types of keys to repair and which to leave as failed
Closed this issue · 2 comments
There are some cases where resetting a failed key will cause an infinite loop (e.g. when the model name does not exist). Ideally, the janitor will be able to decide when it should stay in failed (404), and when it just needs to be retried due to a server issue or 429 error.
It gets the errors from the redis entry itself, so perhaps the consumer will need to have a flag for "resettable". (but then why not just retry the request if it knows it just needs to be retried).
This is currently not a useful issue.
All functionality has been removed, except monitoring processing queues and making sure they are still valid.
We can make a separate request to add back the failure inspection/repair as well.
Another issue can be created if the need arises.
For now, I am closing this issue as we have decided (for now) to have the janitor not restart any failures.