pg_ctl is a control utility on top of the postgres binary allowing to do many operations on a server. A couple of its subcommands offer a way to wait for the operation to finish before exiting pg_ctl, which is useful to ensure the state of the server before moving on with other things in the upper application layer. Sub-commands start, restart and register (Windows-only to register Postgres as a service in the in-core SCM), are not waited to finish by default, but this can be done by using the -w switch. Reversely, the sub-command stop implies to wait, and this can be disabled by using the switch -W. In Postgres 10, the sub-command promote has gained this option, per the following commit:
commit: e7010ce4794a4c12a6a8bfb0ca1de49b61046847 author: Peter Eisentraut <firstname.lastname@example.org> date: Wed, 21 Sep 2016 12:00:00 -0400 pg_ctl: Add wait option to promote action When waiting is selected for the promote action, look into pg_control until the state changes, then use the PQping-based waiting until the server is reachable.
Before this commit, what pg_ctl did was to just write in PGDATA a file called “promote” to let the startup process know that it needs to exit recovery, to take a couple of end-of-recovery action, and to jump to a new timeline before switching the server to read-write mode. Once pg_ctl was done, it was necessary to have some additional logic for example querying pg_is_in_recovery() on the newly-promoted server to see if the server was ready for read-write queries or not. While not complicated, that is always an additional task to do for the server maintainer when doing a failover.
With pg_ctl promote -w, such additional logic becomes unnecessary and the whole process is actually more responsive. pg_ctl checks periodically for the control file of the server and see if it has been switched to an in-production state, then it considers that the promotion is completed, a check happening every second until a timeout ends (can be defined by the user). However, the control file is updated closer to the point where the backends are told that they are authorized to generate WAL, as improved by this commit.
Even if pg_ctl is not used, and some custom utility is in charge of telling a Postgres instance to perform an upgrade, looking directly at the on-disk contents of the control file would improve the responsiveness and the reliability of the promotion process. Something to keep in mind when working on your cluster management tools.