Re: Non-overlapping updates blocking each other

> > On Mon, Oct 16, 2017, at 10:55 AM, Tom Lane wrote:
> > I'd bet on the last one, especially since you found that the problem
> > was a page-level lock. Did you look to see which relation the page
> > lock was in?

On Mon, Oct 16, 2017, at 12:34 PM, Seamus Abshere wrote:
> The specific relation varies, but it appears to always be compound GIN
> index on (jsonb, text)

This is definitely GIN fastupdate. I turned off fastupdate and the
blocks go away. I have a feeling, however, that my UPDATEs will actually
get slower (testing now).

I'm most interested in the fastest UPDATEs possible, even if reads
suffer or similar UPDATEs take wildly different amounts of time.

Should I crank maintenance_work_mem and gin_pending_list_limit way up,
and autovacuum thresholds way down?