Arch Linux

ulidtko

Member

Registered: 2013-05-11

Posts: 10

Cinnamon 5.2.7 goes lagging every few seconds after a day of uptime

Hi forum. I'm trying to figure out an extremely annoying behavior: roughly every few seconds, desktop freezes for about a second. Reboot helps for a day or two, then it starts the same again.

So far, I noticed a perfect correlation of the lag with huge spikes in minor pagefaults, here demonstrated with dstat -a --vm:

As can be seen, the background levels of minpf are in tens to thousands; during lag, it spikes to ~50k per second.

Further, I'd captured flamegraphs using perf record -e minor-faults -a -g -- sleep 30. Freshly booted system (no lag):

And system with 1 day of uptime (lagging):

I posted the flamegraphs and source data to this gist — go there to zoom in.

The rightmost tower is from Slack (the IM) — but it's apparently not the problem, because killing it does not resolve the lag. Same for Firefox.

What does appear to be the problem — is `cinnamon` process and others related to it, `cinnamon-subprocess-helper`, `JS Helper`. On the next-to-bottom row of flamegraphs, you can see how small proportion of cinnamon's pagefaults is on the "baseline" graph VS how those 3 processes become the dominating pagefault cause on the "repro" graph. Perhaps this is easier to see on perf report screenshots:

I did check ~/.xsession-errors — there're some junk logs from cinnamon, but nothing new appears during the lag.

Any suggestions how to identify the issue?
I'd appreciate links to cinnamon source repos — or better yet, any ways to enable debug output.
Thanks in advance

ulidtko

Member

Registered: 2013-05-11

Posts: 10

Re: Cinnamon 5.2.7 goes lagging every few seconds after a day of uptime

Ah... Definitely confirmed this is Cinnamon bug. Just remembered that Win+L opens "Melange - Cinnamon Debugger" (instead of locking the screen as I used to expect) and it has a "Restart Cinnamon" button.

After restart (without reboot), pagefaults did not spike for maybe 2 minutes. Then started ramping up again, 30k... 40k... No lag visible yet, but I'm convinced there's a big performance issue here.