Do you have an idea why autovac was failing to clear the issue on that one problem table, though?
Before I intervened, the maintenance_work_mem was only 16MB (they had encoded and carried over PG 8 defaults in the chef recipe). I bumped it to 512MB before kicking off my freezeThat's the big factor I can think of. This is a huge table obviously as well. By the time it caught my attention there were 850M dead tuples to be cleaned up. My VACUUM FREEZE has been running just just about 5 days and is probably half way done. We shouldn't be at risk of hitting wraparound though (only 52% there).