On Sat, May 28, 2022 at 01:40:14AM +0800, 徐志宇徐 wrote: > vm.swappiness=0 I think this is related to the problem. swappiness=0 means to *never* use swap, even if that means that processes are killed. If you really wanted that, you should remove the swap space. Swap is extremely slow and worth avoiding, but this doesn't let you use it at all. You can't even look at your swap usage as a diagnostic measure to tell if things had been paged out at some point. I *suspect* the problem will go away if you set swappiness=1 in /proc (and in sysctl.conf). -- Justin
