Dear Fujii-san,
> Thanks for the review! At first I pushed 0001 patch.
I found your commit. Thanks!
> BTW, 0002 patch adds the regression test that checks
> pg_stat_activity.application_name. But three months before, we added the similar
> test when introducing postgres_fdw.application_name GUC and
> reverted/removed it because it's not stable [1]. So we should review carefully
> whether the test 0002 patch adds may have the same issue or not. As far as I read
> the patch, ISTM that the patch has no same issue. But could you double check
> that?
I agreed we will not face the problem.
When postgres_fdw_disconnect_all() is performed, we just send a character 'X' to
remote backends(in sendTerminateConn() and lower functions) and return without any blockings.
After receiving 'X' message in remote backends, proc_exit() is performed and processes
will be died. The test failure is caused because SELECT statement is performed
before dying backends perfectly.
Currently we search pg_stat_activity and this is not affected by residual rows
because the condition is too strict to exist others.
Hence I think this test is stable. How do you think?
Best Regards,
Hayato Kuroda
FUJITSU LIMITED