In certain situations, images consistently fail to be indexed by Clair. The cause might be a malformed image, a key missing or any other issue in tbetween. On these failures, Clair returns a 500 to Quay saying that indexing failed, at which point Quay requeues the image to be scanned again and sends the same manifest to Clair. This causes several issues:

1) it can significantly slow down indexing of other images stored in Quay, because the same image is queued over and over.
2) it can create significant transfer costs to the client, due to the same issue.

Instead, we propose the following: If a manifest fails to be indexed more than n times, it should be marked as "failed to be indexed" and removed from the indexing pool. The number of consecutive scans should not be bigger than 5, and there should be a small delay between enqueueing the same image. The Quay UI should also show "Failed to be indexed" error and there should be a button in the UI which would reset the number of scanning attempts so that the image can be rescanned again at user's discretion.

To make this work, Clair should return the failed manifest SHA in the error message or in the response header so errors could be tracked.

This requirement is not needed because the worker already knows which manifest is failing.