From comment seems like each backend should have its own copy of PgBackendStatus, it means st_changecount should be different for each process. If st_changecount was corrupted for 1/2 process, how can it impact newly created process. So could you please help to understand then how come if we run new query via new console (means new process) that also is getting stuck.
neeraj kumar <neeru.cse@gmail.com> writes: > Took some time to get stack trace as we didn't had root permission. > Attaching stack trace of two process (out of many) stuck for same query > below[1][2]
Hmm, the line numbers in your stack traces don't agree with either v10 or HEAD branches for me. But assuming that you've correctly identified where it's stuck:
the only really obvious theory is that some process left its st_changecount odd, which would more or less have to imply that something threw an error between pgstat_increment_changecount_before and pgstat_increment_changecount_after. There's only one place where that seems very plausible, namely pgstat_bestart, which is doing a rather scary amount of stuff in between. Are you using either SSL or GSS?