On Tue, Aug 25, 2020 at 12:32 AM Ed Greshko <ed.gres...@greshko.com> wrote: > > On 2020-08-25 13:58, Chris Murphy wrote: > > On Mon, Aug 24, 2020 at 11:30 PM Ed Greshko <ed.gres...@greshko.com> wrote: > >> With F32 and previous versions I've been able to install Workstation and > >> other spins in a QEMU VM > >> with about 1200MB~1248MB of memory. With F33 the install process crashes > >> during the "installing software" > >> phase. > >> > >> Is this expected? > >> > >> FWIW, the "Everything-netinst" install of a KDE system did install just > >> fine with 1248MB assigned. > >> > >> Currently using... > >> > >> Fedora-Everything-netinst-x86_64-33-20200824.n.0.iso > >> Fedora-Workstation-Live-x86_64-33-20200824.n.0.iso > >> Fedora-KDE-Live-x86_64-33-20200824.n.0.iso > > Are there any hints why it crashed in the journal? That is pretty > > tight memory wise. Minimum RAM 2G is what's listed on getfedora.org, > > and 4G i s recommended. The biggest difference memory wise I can think > > of between F32 and F33 is no disk based swap by default. It uses zram > > based swap. And in such a tight case, only 50% may not be enough. > > > > Even before the zram change though, I'd see ~250M of swap used with a > > 2G VM. So that means it's already under provisioned, and walking on > > thin ice. > > > > But we should make sure we have a complete explanation before making > > adjustments. You could see if zram set to 75% and 100% help, just as > > an extra data point. Create /etc/systemd/zram-generator.conf and edit > > to include > > > > [zram0] > > zram-fraction=0.75 > > > > and then 'systemctl restart swap-create@zram0' > > > > confirm with swapon and/or zramctl that its size is 75% that of totalmemory > > > > You can do this prior to beginning the installation. > > > > OK, doing the above still resulted in a crash. But, it happened later in the > process. The > journal did show > > Aug 25 02:27:15 localhost-live anaconda[2042]: program: stderr: > Aug 25 02:27:14 localhost-live anaconda[2042]: program: Running... umount > /run/install/source > Aug 25 02:27:13 localhost-live earlyoom[1015]: sending SIGTERM to process > 2042 uid 0 "anaconda": ba > dness 77, VmRSS 55 MiB > Aug 25 02:27:12 localhost-live earlyoom[1015]: low memory! at or below > SIGTERM limits: mem 4.00%, > swap 10.00% > Aug 25 02:27:12 localhost-live earlyoom[1015]: mem avail: 44 of 1191 MiB > ( 3.77%), swap free: > 0 of 892 MiB ( 0.00%) > > So, it would seem, zram based swap is less forgiving than disk based swap.
It's getting clobbered by earlyoom. It really doesn't have enough memory if it's swapping out 892M. If that were on disk, it'd be kindof a dreadful experience performance wise, but yeah it would eventually succeed if big enough. Another test would be to use Custom partitioning to create disk-based swap equal to the amount of memory. To do that: keep the btrfs scheme and have it create the layout for you, then reduce the Btrfs partition (click on / or /home, and on the right side UI find 'Volume' and click the Modify button, use fixed size and set a value - now you can create swap same size as RAM). I'm curious how much swap it ends up using but it'll probably be more than 800M in which case, it's just seriously under resourced. Still another test, you can keep increasing the zram device size up to 2x memory, i.e. ' zram-fraction=2.0' That has its own consequences but will still be way faster than disk based swap. -- Chris Murphy _______________________________________________ test mailing list -- test@lists.fedoraproject.org To unsubscribe send an email to test-le...@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/test@lists.fedoraproject.org
