gentoo emerge in tmpfs

One of my gentoo pain points has been the extremely (long and) resource intensive emerge for large packages. However, it is expected and many have come up with their own tricks to make this more efficient.

For me, this has been a gradual step-by-step approach, as I encountered these issues and discovered more about Gentoo.

I stated out with my portage configured as default -- compiling everything on the same disk/filesystem mount on my rotating (not ssd) disk. Any emerge affected everything else, which was too invasive to ignore.

I want emerge to run gently in the background, not affecting any foreground tasks. So I used nice, and later also ionice. I did so on the commandline for each emerge. Anytime I forgot, the system would crawl for almost the entirety of the emerge.
# nice -n19 ionice -c3 emerge ...

And then I found out about making them the default options in man make.conf and /usr/share/portage/config/make.conf.example[1].

$ cat /etc/portage/make.conf | grep NICE

For ionice, your kernel needs the CFQ ioscheduler. Check your kernel config. I found my system suffering (sometimes extreme) lags due to CFQ, and I switched back to noop. Apparently, CFQ is more efficient on larger systems and heavier loads. Most of the time, my minimal-system resource usage is almost nil. You might have a different experience.

I still felt emerge invasive enough, although much less than before. Longer emerge and larger packages have a noticeable impact on foreground tasks. A kindly soul on the gentoo forums suggested[3] compiling on tmpfs. I did a search and found this approach on the Gentoo Wiki[2].

Since I keep a tight control on what is on my system, I find that I usually have more memory to play with than most others. So I migrated to Portage TMPDIR on tmpfs[2].

Portage TMPDIR is /var/tmp/portage. To move it to tmpfs, we just mount it there. This can be set at boot in /etc/fstab.

$ cat /etc/fstab | grep portage
tmpfs /var/tmp/portage tmpfs defaults,noatime,mode=775,uid=portage,gid=portage,size=3G

All was well, till emerge started crashing.. for larger packages like firefox and libreoffice. Again the gentoo forum helped me understand why. I had to temporarily reset the Portage TMPDIR back to disk, by unmounting /var/tmp/portage, for larger packages.

The next step was identifying packages which needed more emerge diskspace than my tmpfs, and just setting those to emerge on disk, rather than tmpfs[2].
cat /etc/portage/package.env
app-office/libreoffice notmpfs.conf
games-action/supertuxkart notmpfs.conf
sys-devel/gcc notmpfs.conf
www-client/firefox notmpfs.conf
cat /etc/portage/env/notmpfs.conf

Now I did't need to unmount /var/tmp/portage when these larger packages hit emerge. But the pain points are back, as emerge on slow-disk, particularly for large packages, becomes a bottleneck again. The larger packages demanding more resources and consequent longer emerges was that particular pain point, and just for this I was entirely back on disk again, rather than tmpfs for even part of it. Still, this is much better than when I started with default portage config.

Looking back in hindsight, all this seems logical. But till I hit the wall, I didn't realise what needed to be done or how. And it happened, one step at a time. I seem to be learning from *my* mistakes, rather than learning from others'. Hmm..


No comments:

Post a Comment

most viewed