tuwunel locks up / RocksDB Write Stalling

[gymnae]

I'm running tuwunel inside an unprivileged LXC on a Proxmox host. Tuwunel runs as binary on a fedora guest. Since I added some users, I often appear to be suffering write stalling. The write ahead log grows, sometimes it jumps from 1 to 20 or 40 or even 60. If this happens, the app does not crash, but I notice no messages go in or out. I then have to restart tuwunel.

tuwunel has exclusive access to two cores of an i5-12400, 8GB RAM and runs on an NVME. I have only three active users. So that hardware should be fine?

Are the database or other settings, which can ensure reliable functionality in shared environments, where no process of tuwunel has exclusive access to I/O or CPU cycles?

[x86pup]

@gymnae Are you still having these issues on the latest Tuwunel version 1.9.0?

[gymnae]

Hi there, yeah. I still see logs popping up and my automated restart script jumps in:

Tue Sep  1 03:25:02 PM CEST 2026: Stall detected (16 logs). Triggering safe restart.
Tue Sep  1 04:45:01 PM CEST 2026: Stall detected (17 logs). Triggering safe restart.
Tue Sep  1 06:30:02 PM CEST 2026: Stall detected (25 logs). Triggering safe restart.
Tue Sep  1 07:45:02 PM CEST 2026: Stall detected (19 logs). Triggering safe restart.
Tue Sep  1 09:25:02 PM CEST 2026: Stall detected (30 logs). Triggering safe restart.
Tue Sep  1 10:10:02 PM CEST 2026: Stall detected (24 logs). Triggering safe restart.
Wed Sep  2 12:25:02 AM CEST 2026: Stall detected (26 logs). Triggering safe restart.
Wed Sep  2 04:50:01 AM CEST 2026: Stall detected (18 logs). Triggering safe restart.
Wed Sep  2 08:25:01 AM CEST 2026: Stall detected (24 logs). Triggering safe restart.
Wed Sep  2 10:50:02 AM CEST 2026: Stall detected (43 logs). Triggering safe restart.

[x86pup]

Can you elaborate more on what this "automated restart script" does? Do you have any tuwunel server logs to share when this happens too?

Are you running tuwunel on a CoW filesystem like btrfs?

[x86pup]

I filed a proper bug report / issue on your behalf so I'd move further responses there so it can be properly tracked and investigated: #572