On Wed, Sep 30, 2020 at 3:41 PM Adrian Klaver <adrian.kla...@aklaver.com> wrote:
> On 9/30/20 2:30 PM, Adam Sjøgren wrote: > > Adrian writes: > > > >> I don't have an answer. Not even sure if this is relevant to the > >> problem, but how are the jobs getting into the queue? > > > > Plain INSERTs - often a lot at the same time. I inserted 400K jobs to > > clean up after a bug earlier today, for instance. > > > > What were you suspecting? > > Honestly a fishing expedition. All the discussion, as far as I could > remember, had to do with the retrieval part of the process. Just thought > for completeness it might be useful to flesh out the insertion portion > of the process. Though it does get one to wondering how the appearance > of a large number of jobs at once affects the those programs looking for > new jobs? > Is autovacuum/analyze keeping up with the changes to allow the planner to make prudent choices? With so many updates/deletes, I would hope so but if still using the 20% or 10% default values for scale_factor, and at times the table has many millions of rows, perhaps not.
