On 09/12/2018 19.41, Andrei Borzenkov wrote:
09.12.2018 14:13, Carlos E. R. пишет:
The "kswapd0" kernel process was stuck at 25% CPU, so something was trashing like hell, in a loop. Thus starting any process took minutes. I was lucky to even be able to ssh in and finally kill firefox.
25% is likely one logical CPU (or core), i.e. kswapd consumes 100% of allocated CPU resources. kswapd is started (woken up) when memory becomes short and attempts to free some memory. So in your case it looks like it fails to free enough memory (at least for for long time) and is constantly started again. Every time it needs to traverse page list to find candidates to release so it is time consuming.
Aha. Ah, wait, that machine has 2 cores, so it was half of one core used.
When you stop firefox you free large amount of memory which returns system to normal state.
So if the next time I kill Chrome perhaps it also returns to normal.
It is impossible to say why it happened from top output alone. sysrq-m output may provide some clue.
I know, there is not enough information. How can I do sysrq-m over ssh? Because locally the machine did not respond at all, black screen. <http://www.linuxhowtos.org/Tips%20and%20Tricks/sysrq.htm> Alt+SysRQ+m prints memory information to the console. Ok, so, is there a command to dump that information on the ssh session? -- Cheers / Saludos, Carlos E. R. (from 42.3 x86_64 "Malachite" at Telcontar)