@feld@lanodan Is there any way to switch to a different message broker than the one that's backed by Postgres? I'm wondering if the I/O issue that @sun and I ran into might be exacerbated by a DB-backed queue. #Pleroma
@tk@lanodan@sun I personally think it's bad queue design and I'm working on a branch to remediate that. Too many queues, some most expensive operations (DELETE) allowed too many concurrently.
@feld@lanodan@sun I wonder if the auto-vacuum could be involved here, too. A lot of deletes would require the auto-vacuum to do a lot of I/O work. :blobfoxthink:
@icedquinn@feld@lanodan@sun All I know is that when I got Postgres back up after migrating to the new volume, the vacuum worker was doing a ton of I/O.
Add comment