Re: Out-of-swap killer and SIGTERM signal
- Reply: Ambert : "Re: Out-of-swap killer and SIGTERM signal"
- In reply to: ralph41096 : "Out-of-swap killer and SIGTERM signal"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Wed, 12 Jan 2022 03:21:25 UTC
On 2022-Jan-10 12:55:56 +0000, ralph41096 <ralph41096@protonmail.com> wrote: >Do you think it would be a good idea to introduce the following soft >mode for the Out-of-swap killer? There have been lots of discussions about this in the past, starting in about 1998, (though I agree that it's been about 4 years since the last discussion). I suggest you search for "freebsd+sigdanger" for previous discussions. >New behaviour: > 1) If there is less than 2 GiB of free swap space, send SIGTERM > signals to the condemned processes, then wait for a 5 seconds > grace period, then send SIGKILL signals to the condemned processes > who are still running. 2 GiB seems quite a high limit, especially with a 5s grace period. That implies the system is using swap at 410 MiB/s - which seems unrealistic. It also means that you need to overprovision swap by 2 GiB since the last 2 GiB is effectively unusable. >These two system-wide parameters (2 GiB of free swap space, 5 seconds >grace period) could be defined in a configuration file, to allow the >administrator to change them. The FreeBSD model would be to make them sysctls. -- Peter Jeremy