On Fri, Jul 22, 2022 at 1:09 PM Michael Paquier <mich...@paquier.xyz> wrote: > On Thu, Jun 23, 2022 at 02:30:13PM +0900, Michael Paquier wrote: > > One idea I got to limit the useless output generated is to check the > > status of the cluster after running the regression test suite as > > restart_on_crash is disabled by default in Cluster.pm, and avoid any > > follow-up logic in these tests if the cluster is not running anymore, > > as of the attached. > > So, this is still an open item.. > > Thomas, any objections about this one? Checking for the status of the > node after completing pg_regress still sounds like a good idea to me, > because as restart_after_crash is disabled we would generate a ton of > logs coming from regression.diffs for nothing. On top of that the > parallel connections make harder finding which query failed, and the > logs of the backend provide enough context already on a hard crash.
What if the clue we need to understand why it crashed was in the regression diffs that we didn't dump? I wonder if we should move the noise suppression check closer to pg_regress, so that it works also for the "main" pg_regress run, not only the one in this new TAP test. As discussed in this thread, inconclusively: https://www.postgresql.org/message-id/flat/CA%2BhUKGL7hxqbadkto7e1FCOLQhuHg%3DwVn_PDZd6fDMbQrrZisA%40mail.gmail.com
