This is an automated comment for commit 37f2fb2 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

This PR has multiple flaws I'd like to address

1. Tests require system.errors having a specific state at start. Re-running tests may produce incorrect results. I haven't found another reasonable way to query errors state (maybe using part_log?).
2. Deleting a mutation when a table doesn't exist is a cure for a specific case. The general problem is that mutations' info doesn't include neither retry count, nor error_code (only message, which isn't a reliable filter). However, building a proper fix is quite a big task.
3. This fix (currently) doesn't work for replicated tables. The only solution I see is to delete mutations on each replica individually
4. It is possible that a mutation for table A referencing table B starts when table B is not loaded yet. In this case I believe my solution is incorrect

A good solution could manage an exponential backoff for failed mutations, but we have only start time (not finish time) in system.mutations, and time is not used while planning the mutation job to MergeTreeBackgroundExecutor

As Mikhail is working on a better approach for mutations, closing this PR