On Mon, Oct 30, 2023 at 6:57 AM Alexander Kukushkin <cyberdemn@gmail.com> wrote:
...
> As Michael already said, the following workflow works just fine (I did it dozens of times):
> 1. enable checksums on the standby node
> 2. start the standby and let it catch up with the primary
> 3. switchover to a standby node
> 4. enable checksums on the former primary (now replica).
There is also a good trick described in
https://www.crunchydata.com/blog/fun-with-pg_checksums to avoid
accidental start of Postgres:
after pg_ctl stop and before pg_checksums --enable, do:
mv data/pg_twophase data/pg_twophase.DO_NOT_START_THIS_DATABASE
and once pg_checksums --enable is done, move it back.
Additionally, I compiled some thoughts about running pg_checksums
without downtime (Patroni-friendly, of course) here:
https://twitter.com/samokhvalov/status/1719961485160689993.