On 17.07.2019 18:14, Andres Freund wrote:
To me that means that we need prioritization across databases, and
between tables, and probably by multiple criteria. I suspect there
need
to be multiple criteria how urgent vacuuming is, and autovacuum ought
to
try to make progress on all of them.
I'm not a contributor, and don't know the code base, so don't normally
comment. But it occurs to me in the present case that there may be a
short-term workaround for the current problem: could autovacuum visit
tables in a random order? Then at least all tables would get vacuumed
eventually (in the current case). If, later, more complex priorities are
implemented, they could be probabilistic weights.
Best,
-- Shaun Cutts
