fix DeleteFolder action being sporadically issued by Podbox

[ichub]

fixes https://linear.app/0xparc-pcd/issue/0XP-1061/pretixpipelines-deletefolder-action-is-sporadic

determine whether to delete a pipeline's folder based on whether or not it has loaded at least once since server start, rather than based on whether the current instance of the class corresponding to the pipeline has loaded

[artwyman]

I have some thoughts, though I might be thinking in the wrong direction given my limited context on how the PODBox backend is structured. I'm glad to review this, though you might want to solicit a second review from another dev with more context too.

[artwyman]

This has a cost (to fetch from the DB) so I have some thoughts on reducing duplication of that work:

1. It seems like there's duplicate effort from loading the tickets above. Is it difficult to get this information from the same fetch. Maybe it's tricky since there could be no tickets for the user's email. At least if tickets is non-empty we can assume there are atoms without fetching again, right?
2. It should at least be possible to do this only once per pipeline instance if the first time you do this fetch you set this.everLoaded.

There are some assumptions here, like the fact that the atom DB is never empty in an active/loaded system. What would happen if an admin simply deleted all the tickets? Would that mean no atoms, and no delete, so that the client-side cleanup would never happen? Or are atoms a more persistent concept than that?

[robknight]

I wonder if any of the problems occurred due to async requests occuring inbetween the this.db.clear() and this.db.save(), which are not atomic.

There's also a small possibility that a pipeline might serve stale atoms from the atom DB after it has been restarted, and maybe some mismatch between the new post-restart config and pre-restart atoms might cause a problem (I can't think of one but it did make me wonder).

[robknight]

In #1830 I've solved the problem of concurrent reads occuring inbetween clear() and save() by serializing access to the atom DB using PQueue, with clearing and saving occuring atomically.

[ichub]

subsumed by #1839