Hi, On Wed, Jan 29, 2025 at 07:01:00PM +0000, Ben Harris wrote: > On Wed, 29 Jan 2025, Ben Harris wrote
I tried to reproduce the reinit failure, but it always worked for me. Not sure what is different on your side :-/ > > If I add "WorkingDirectory=~" to patroni@.service, even on a current > > Patroni package, reinit starts working again. This gives me a > > convenient workaround, but I'll try to work out the cause as well. > > And after a bit of fighting with strace, I've spotted where the bug (or > rather the anti-bug) is. Here, in pg_clonecluster_patroni: > > rm -f etc/postgresql/$VERSION/$CLUSTER/postgresql.base.conf > > Note the lack of a '/' before "etc". Ouch, thanks for spotting that. > This means that if the script is run with its current directory being / (as > systemd runs services by default), then in will successfully delete > postgresql.base.conf, pg_dropcluster will fail, and so reinit will fail. But > if the current directory is set by systemd then the "rm" will fail, > pg_dropcluster will succeed, and so reinit will succeed. > > I would suggest that the proper fix here is to insert the missing '/' and > move the "rm" command back to being after the "pg_dropcluster" command. Or > just drop the "rm" command entirely. I did the former now in the salsa git repo, however, that will only be part of the 4.0.x releases. Michael
