On Tue, May 04, 2021 at 12:43:53PM +0530, Bharath Rupireddy wrote:
> And having a retry test case with clobber cache enabled doesn't make
> sense because all the cache entries are removed/invalidated for each
> query, but the test case covers the code on non-clobber cache
> platforms, so I would like to keep it.
Yeah, I'd rather keep this test around as it is specific to connection
caches, and it is not time-consuming on fast machines in its new shape
either. Another trick we could use here could be an aggregate
checking for the number of rows returned, say:
SELECT count(pg_terminate_backend(pid, 180000)) >= 0
FROM pg_stat_activity
WHERE application_name = 'fdw_retry_check';
But using CALL as you are suggesting is much cleaner.
(Worth noting that I am out this week for Golden Week, so if this can
wait until Monday, that would be nice. I am not willing to take my
chances with the buildfarm now :p)
--
Michael