repmgrd configuration questions about repmgr

[martinmarques]

If your worries are related with using a superuser to create such functions, I have to say that not loading the functions will not change anything, as you need superuser privileges for cloning as well.

About repmgrd, you need to start it on all standby nodes which you want to monitor. If you want to monitor lag from all standbys, then you have to have it running on all standby nodes.

You should have it ready on the primary, so after switchover it can be easily started.

[gp4git]

Concerning question 1, my worries are that I want to install the minimum required for the features I plan to use. And correct me If I'm wrong, these functions are only in the scope of repmgrd not plain repmgr features.

As for question 2, I ask where to start repmgrd daemon because I have seen it started on both master and standby nodes. Is it safe to do so ? May be it can help to hasten switchover process ?
And a last question : as the standbys are in read only mode, how repmgrd is able to write into the monitoring table ?
Thanks again for your feedback
Regards

[martinmarques]

On the master, repmgrd is only needed in case of auto-failover. It's good to have it anyway, so in case of switchover it will start storing monitoring information in the new standby

About how repmgrd works on a standby, it gathers information from the standby node but writes it to the primary node, that's why you need normal libpq connections as well as replication connections from the standby in pg_hba.conf.

[gp4git]

Sorry to be picky, but my question revolves whether it is safe/better to run repmgrd with ONLY --monitoring-history option on the master or if it will burn CPU cycles for nothing and should be started only during switchover.?

Also, as homegrown replication monitoring scripts use pg_stat_replication view on the master, may I say that repmgrd monitoring is more accurate as it gets its info from standby nodes ?

Regards
And thank you for your patience ;-)

[martinmarques]

I will not burn CPU cycles. It just checks if it needs to reload the configuration every monitor_interval_secs seconds.

Also repmgr stats are more accurate as they compare LSN from the standby and it's primary. It also gives time lag.

[gp4git]

OK thanks for your valuable feedback
Regards