Maintenance mode switching works with a delay

[0ranki]

Question: This is a long shot, but I have recently been getting this issue that when I enable or disable maintenance mode, there's a 1-2 minute delay of maintenance mode actually switching state.

I'm running the Apache image (supervisord+cron), with Postgres, Redis and ClamAV containers. The setup is slightly unorthodox in that I'm running the containers in a rootless pod with Podman in an LXC container. Current NC version is 23.0.3, the rest are running latest versions.

The behaviour appeared probably with the 23.0.2 update, though I switched from MariaDB to Postgres when running that version. This is mostly an inconvenience, as my backup scripts already pause between enabling maintenance and doing the actual backup, but since the behaviour is new I thought to ask here. After enabling maintenance, status.php shows "maintenance": true, but the instance is still accessible. There should be no caching happening downstream of the containers. When the maintenance mode kicks in, it doesn't seem to be related to cron, and NC logs (container or nextcloud.log) show no activity.

Any idea where to start looking on what's causing this?