I spent a while operating PostgreSQL running on an EC2 instance. We had a weird problem where sometimes queries would fail, with the error
could not open shared memory segment "/PostgreSQL.271741757".
For non-system (UID < 1000) users, logind will by default clear shared mem files upon logout. For things like databases, this is not at all desireable. One thing to note is that it’s unlikely for this to happen if you never login as the database user.
Setting RemoveIPC=no in /etc/systemd/logind.conf should resolve this. In theory, you can restart logind while the system is running. I did some tests, but I think it’s still probably safest to reboot the system.
Man page
RemoveIPC=
Controls whether System V and POSIX IPC objects belonging to
the user shall be removed when the user fully logs out. Takes
a boolean argument. If enabled, the user may not consume IPC
resources after the last of the user's sessions terminated.
This covers System V semaphores, shared memory and message
queues, as well as POSIX shared memory and message queues.
Note that IPC objects of the root user and other system users
are excluded from the effect of this setting. Defaults to
"yes".
Once changed, a full reboot will ensure the config is loaded. Alternatively, a restart of logind should also resolve this: systemctl restart systemd-logind.service
Useful links:
- https://github.com/systemd/systemd/issues/4532
- https://github.com/systemd/systemd/issues/2039
- https://stackoverflow.com/questions/49065733/redhat-or-centos7-systemd-may-remove-user-ipc-resources-unexpectedly
- https://stackoverflow.com/questions/73368759/why-systemd-remove-my-shm-file-but-not-postgresqls