Here is some more detailed feedback based on my latest testing.
At the beginning, the system is still behaving normally. There is a short spike right after startup (all clients were closed before), but it quickly settles down again.
I tested the typical use cases first (web UI, Outlook on Windows, iOS clients, etc.), which all worked fine.
Then I powered on my Mac and opened Apple Mail. From that point on, memory usage started increasing steadily at roughly 10% per minute.
This continued until one nginx worker process reached around 50% memory usage. At that point I only had a single mail client connected.
Shortly after, a second worker process started to grow as well, reaching around 40–45% memory usage, as also visible in the screenshot.
Another noticeable behavior is that gromox-http shows a constant CPU load of around 10%, which does not look normal, along with the increased network traffic.
After some more time (10 minutes), the system becomes increasingly unresponsive (admin web UI, etc.), and eventually the OOM killer is triggered. In my case, nginx processes are being killed due to excessive memory usage (as confirmed via dmesg).
After the nginx worker processes are killed and I close the Apple Mail client, the system recovers and memory usage returns to normal. Only one woker with 40% of mem usage is left.
I hope I tested everything as requested. For reference, memory usage observed directly on the console (ps/top) matches what is shown in Proxmox as well as in the admin dashboard during the increase.
