I tried taking into account how states override each other

Avoid pushing resync jobs for deleted users

Regarding syncrotons needs:

• I think we need to get away from the timestamp comparisons and instead store whatever we need to as part of the synchronization state (tied to the synckey)
• Ideally we find something better to store than all entries in a folder that are part of the tag, or at least a more efficient representation than huge urls.

just as an fyi, purging users became less of a priority, so I'm parking this diff for now.

Rely on onlyTrashed

Do not disable a pending mandate

Really only test the IMAP_READY status

Rewrote resync to use separate queries for created/deleted

Do not retry the payment, we will on the next charge cycle, but make sure email notifications are not interrupted by topup failing.

Check updated_at/created_at

Fixed some typos, now the query exeutes in reasonable time

Questions:

• Is this good enough to detect inactivity?
• Is the degraded state good enough or do we need to check the wallet balance?

This uses the same mechanisms as the horizon dashboard, and allows us to largely ignore the dashboard and implement monitoring (to e.g. notice if a queue gets stuck or so).

Makes sense in principle, but yields about 20k fewer entries it seems (even when filtering only for the "pushed" jobs).

Only count actually pushed jobs

In D5014#62904, @machniak wrote:

Also, whereNot('status', '&', User::STATUS_DELETED) would probably do the same. This status is set after user deletion from LDAP/IMAP was successful.