On Wed, Jul 27, 2022 at 11:26:44PM -0400, Tom Lane <tgl@xxxxxxxxxxxxx> wrote: > Scott Ribe <scott_ribe@xxxxxxxxxxxxxxxx> writes: > > On Jul 27, 2022, at 11:30 AM, Ian Lawrence Barwick <barwick@xxxxxxxxx> wrote: > >> If you are using a Systemd-enabled Linux, check that "RemoveIPC" is > >> set to "no". > > > That was the answer. I checked other servers, they had it set. But they are much older OSs (overdue for updates, RSN), different installers. > > I looked at this on my own machines and was interested to find that > RHEL8 defaults it to "no", although the upstream default is evidently > "yes" and Fedora follows that. Not sure what to make of that, > except that OS-vendor opinion isn't universal that this is a great > idea. > > regards, tom lane It is not a great idea. On Debian, I had a postgres user cronjob, and occasionally, systemd in its infinite wisdom would decide that the end of the cronjob meant that the postgres user had logged out and needed to have its IPC resources deleted, even though the IPC resources in question were created by a different long-running postgres user "session" that was still running (i.e., the database server). The consequence was error messages from the database claiming that the database was corrupt. Very scary. It wasn't corrupt. It just needed systemd to stop deleting its stuff. cheers, raf
