On Sun 2019-09-01 20:58:30 -0400, Tom Lane wrote:
> >> A separate question is how come the particular query you're complaining
> >> about has (seemingly) a fairly wide window where it never does any
> >> CHECK_FOR_INTERRUPTS call before terminating. Perhaps there's someplace
> >> we need to sprinkle a few more of those.
>
> > Yeah, it is strange. There are many queries in this system with statement
> > timeouts and this is the only one where I see this and it happens several
> > times a day.
>
> It'd be interesting to see EXPLAIN ANALYZE output from a typical execution
> of that query. Even better would be a test case (w/ tables and dummy
> data) so somebody else could reproduce it; but maybe looking at EXPLAIN
> ANALYZE would be enough to determine where we're missing checks.
I put output from a query with values that caused the error here:
https://gist.githubusercontent.com/horgh/653e60b8c2f071d859424517d653eb4e/raw/5d16e11d0ac884ed9dc120c4521e2de8f2e7c3d6/test.txt
I sanitised some of the names and values again. Sorry about that.
As far as a test case: I'm not sure how to reproduce it myself. It just
happens periodically (15 times in the past 24 hours). The tables are
somewhat large (60+ GiB per partition). I could try to generate dummy
versions if you like, but they would have to be significantly dummied, so
I'm not sure it would end up being representative.
