Tom Lane wrote:
"Craig A. James" <cjames@xxxxxxxxxxxxxxxx> writes:
Bottom line is that I was expecting "instant death" with SIGTERM, but
instead got an agonizing, drawn out -- but safe -- death of the query.
What was the query exactly?
Our expectation is that all or at least most queries should respond to
SIGINT or SIGTERM interrupts pretty rapidly, say on a less-than-a-second
timescale. However there are various loops in the backend that fail to
execute CHECK_FOR_INTERRUPTS sufficiently often :-(. We've been
gradually finding and fixing these, and will be glad to fix your case
if you provide enough details to pin it down. You might be interested
in this current thread about a similar problem:
http://archives.postgresql.org/pgsql-patches/2006-07/msg00039.php
Thanks, this is good information. The qsort is a distinct possibility. The query is a big
insert into some_hitlist (select id from another_hitlist join data_table on (...))
where the hitlists are unindexed. So it may be using a merge-join with qsort. When I have a few minutes, I'll turn on logging in the app and find the exact SQL, and run an EXPLAIN ANALYZE and see what's really happening.
It's also possible that the INSERT itself is the problem, or adds to the problem. The SIGINT may come after a few million rows have been inserted, so it would have to clean that up, right?
Thanks,
Craig
