Deadlock in flush_insert_instances
I don't think we can repeat the whole transaction (inserts in different tables in different places), but we commit every 10 files. I think dropping 10 files from the index is acceptable.
Deadlock in flush_insert_instances
So over the last month have one case where adding a new partition didn't crash the already running backup (luck?), but every time the backup crashed it was at the same time a new partition was…