Date: Mon, 1 Mar 2021 08:23:17 -0800 From: Mark Millard <marklmi@yahoo.com> To: tech-lists <tech-lists@zyxst.net> Cc: freebsd-hackers@freebsd.org Subject: Re: The out-of-swap killer makes poor choices Message-ID: <132A65F3-E06A-40C1-8488-C421AD02F8F5@yahoo.com> In-Reply-To: <YD0O28EX/Xa2jc87@cloud.zyxst.net> References: <CAOtMX2jYmrK7ftx62_NEfNCWS7O=giHKL1p9kXCqq1t5E1arxA@mail.gmail.com> <YD0O28EX/Xa2jc87@cloud.zyxst.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2021-Mar-1, at 07:57, tech-lists <tech-lists at zyxst.net> wrote: >=20 > On Tue, Feb 23, 2021 at 01:49:49PM -0700, Alan Somers wrote: >=20 > [...lots...] >=20 > I've found that if I take a system already configured with swap, then > add say 2* 16GB swapfiles as per the handbook, enable the swapfiles, > that some things will use the swapfile and others not, even with = plenty > available in the swapfiles. >=20 > example: 12-stable amd64 system running poudriere. >=20 > it will use the swap configured by default (think it's 2GB) as a > partition. In this example, all of swap is on ssd - the partition and > both swapfiles. Compiling some things like rust or firefox will use = all > of the partition swap and some of the *swapfile* but not all of it[1]. = It's > almost like some processes have more of an affinity for swap as a > partition than others, and others won't use swap as a swapfile at all. >=20 > In the end, because I don't know how to debug this kind of issue > further, it was easier to blat the ssd and install 32GB of swap on > reinstallation of the OS. Swapfiles arent used now, on this particular > install. The 32GB swap partition is fine and works as expected. >=20 > Thought the above might be of use as I dunno, a data point. I wonder, > though, why some programs have seemingly more affinity for = swap-as-partition than swap-as-swapfile. >=20 > [1] by "not all of it" I mean fail with an OOM message before all of = the > swapfile is used See https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D206048 for why swapfiles are problematical and likely should be avoided. In particular comments #7 and #8 that quote Konstantin Belousov on the subject (or reference his material). Part of the text says: As result, swapfile swapping is more prone to the trivial and = unavoidable deadlocks where the pagedaemon thread, which produces free memory, needs more free memory to make a progress. Swap write on the raw partition = over simple partitioning scheme directly over HBA are usually safe, while = e.g. zfs over geli over umass is the worst construction. =3D=3D=3D Mark Millard marklmi at yahoo.com ( dsl-only.net went away in early 2018-Mar)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?132A65F3-E06A-40C1-8488-C421AD02F8F5>