Hi > The buildfarm is unhappy after this one: > https://buildfarm.postgresql.org/cgi-bin/show_history.pl?nm=culicidae&br=HEAD > > If I use -DEXEC_BACKEND when compiling the tests complain about a > timeout in 003_recovery_targets. Here is what the buildfarm reports, I > can see the failure by myself as well: > # Postmaster PID for node "standby_6" is 26668 > # poll_query_until timed out executing this query: > # SELECT '0/3022690'::pg_lsn <= pg_last_wal_replay_lsn() > # expecting this output: > # t > # last actual query output: > # f > # with stderr: > Timed out while waiting for standby to catch up at > t/003_recovery_targets.pl line 34.
I can reproduce this and notice wrong assign settings order. For example standby_6 has > recovery_target_name = '$recovery_name' > recovery_target_xid = '$recovery_txid' > recovery_target_time = '$recovery_time' But recoveryTarget was set to RECOVERY_TARGET_XID Without -DEXEC_BACKEND all fine. As far as I understand the guc.c code with EXEC_BACKEND all processes uses different config processing logic. We serialize all GUC and restore at process start. And we sort GUC by name in build_guc_variables - so we restore settings in wrong order. I was afraid that the GUC system does not guarantee the order of settings... What is preferable solution? Seems we can not simple change this logic. I think i can reorganize all new recovery_target_* GUC into single one with format, for example, recovery_target = "xid:607" (was mentioned in patch discussion). regards, Sergei
