Re: Need help identifying a periodic performance issue. - Mailing list pgsql-performance
On Nov 17, 2021, at 2:01 PM, Thomas Munro <thomas.munro@gmail.com> wrote:This message originated outside your organization.
On Thu, Nov 18, 2021 at 8:28 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:Justin Pryzby <pryzby@telsasoft.com> writes:It shows that the process is running FK triggers.
Indeed, and doing a seqscan therein. Normally I'd suppose that
this reflects a lack of an index, but RI_FKey_check should always
be doing something that matches the referenced table's unique
constraint, so why isn't it using that?
I wonder if the reference tables are empty sometimes, and there's an
unlucky sequence of events that results in cached RI plans with seq
scans being used later in the same session after the tables are
populated.
It seems to be important (so far) that we delete a ‘bucket’ in the re-creation of this problem. I’ve included a graphical copy of the schema courtesy of DataGrip. We’re trying to get the problem reproducible more quickly, but at the moment, it takes hours.
Attachment
pgsql-performance by date: