I can try reproducing it, I didn't quite get what is actually triggering that. Let me share some more details:
Postgres instance is getting loaded of 4 big tables (not at the same time, sequentially):
Table1: 2.5B rows
Table2: 2.2B rows
Table3: 7.7B rows
Table4: 1.3B rows
Autovacuum to prevent wrap around kicks in on those table, I was actually wondering if this is somehow related to autovacuum_multixact_freeze_max_age parameter and how. That behavior should be referred to row locking on multiple transaction but this database is still static and not accessible yet, meaning there shouldn't be row level concurrency. For completeness, the load for each table is executed by 40 parallel sessions running the copy command.
pgxxxxx=> show autovacuum_multixact_freeze_max_age;
autovacuum_multixact_freeze_max_age
-------------------------------------
400000000
Thanks
Alessandro
Here you go:
pgxxxxx=> show vacuum_freeze_table_age;
vacuum_freeze_table_age
-------------------------
150000000
I can't think of any other innocent explanations, so maybe it is a bug. But if it were easy to hit, I'd expect it to have been noticed before now.
Can you reproduce it readily?
Cheers,
Jeff
