Hello!
Addendum. I'll add obervations, which may be related.
Update. Due to updates, I had to shutdown all and restart. This make verything shine golden and remind me, that immidiately after starting the x2go client (and server!), perf is really well (I described my hw/sw already). Its usually over time, that perf degrades that much. Over time is someting really over days, because I usually do no shutdowns, but hibernate my workstation. What I saw before the restart, was additionally a laggy scrolling behavior in Firefox and it reminds me, that I saw this bevore - looks probably related to the memory leak of the x2goagent. After the restart, scrolling works fine.
Related. I have two lxc-containers running on my gateway, one for dns, one for a squid proxy. Both works well, but sometimes, perf is in question. I usually run a terminal to the gateway host and monitor this containers using lxc-top. The related thing is: relative high cpu (20%) and growing mem-use, the reason, they must finally rebooted each two weeks. Now, I've disabled x2goserver in both containers and I am running the third day now: 0%(!!) cpu, NO memory leak. They all run debian buster (fully upgraded) on kernel 5.10, x2goserver 4.1.0.3-4.
May be this helps. So the memory-leak seems to be obvious. In the current, new cycle for my main-use, mem climbed already up from 0.6% to 5.0%.
Regards, Manfred
----- Original Message ----- From: webman@manfbraun.de [mailto:webman@manfbraun.de] To: <x2go-user@lists.x2go.org> Sent: Tue, 11 Apr 2023 02:27:46 +0200 Subject: [X2Go-User] Performance problems with x2go - hickups, pause, probably memory-leak
Hello!
I probably detected strong problems with x2go, not very easy to explain. I am running a debian bullseye desktop, on it a kvm guest with debian bullseye, both have kernel6. The kvm guest x2goserver: 4.1.0.3-5, the host (my workstation) x2goclient: 4.1.2.2-2. Usually everything updated.
The problem is the performance of firefox, but not generally. It stucks sometimes, up to several minutes and is completely unresponsive, especially(!), if it comes to youtube videos, but most time, it is not video. Astoundigly, in the many cases, it recovers. I am run dstat permanently (in the guest) and usually see a memory bottleneck (mem < 1 GB), not too strong and additionally high cpu usage, load: 6/4/4, but not too long. Often, this box crashes, if mem goes below 500 MB. Known, with firefox.
For unknown reasons, this situation is not stabil (I am using up to 500 Tabs), one will never know, when this may happen and pre-actively saveing the session and continue with, say, 50 Tab, doesnt change the problem!
The VM got 8 cores(!) on a AMD EPYC-Rome Processor with max 3.4 Ghz and 24 GB(!) mem, it got two network interfaces, one, to communicate with my host over a local bridge (10-36 Gb/s!), this is the "x2go-channel" and a lan-IF (1 Gb/s, hardware-passthrough) to the gateway (has 500 Mb/s) and nothing overloaded ever. Additionally some diskwrite, but with 600MB/s (in the VM!) this should not indicate a bottleneck.
But since days, while another hickup, I saw, that the fox needs about only 20% cpu, BUT 'x2goagent' neds 100% !! It looks like the hickup appears exactly at those times. I setup a little processmonitor for the x2goagent and this seems to show a memory leak. I append this at the end. I installed 'glances' and that show me the x2goclient as the second mem consumer, growing. What language is the x2goagent written? Looks like a failing heap-management, garbage collection, which needs a lot of cpu?
Some notes would be really welcome!
Thanks, Manfred
--- x2go-"monitor" ----------------------------------- (to keep this short, it has been cutted in the middle)