For completeness, I didn’t ever figure exactly what went wrong, but I ended up fixing it by recreating the volume:
- Figured out which volume was the one in use:
fly volumes list - Created a new one to be used
fly volumes create pg_data --region syd --size 10 - Scaled back up (and luckily the new VM used the new volume):
fly scale count 2 - Deleted the dud volume:
fly volumes destroy $vol_id
I was then still left with the no keeper info available logs, but according to this post, this is fixed in a newer postgres image. So I updated using fly image update and hopefully they will stop after ~48 hours.