Questions re swap-on-zfs

Beeblebrox zaphod at berentweb.com
Sun Dec 1 17:29:46 UTC 2013


Results of testing have producced two problems:

1. The "gconcat label" device does not persist through reboots. I suspect
the reason is due to swap #1 being on a zpool (tank0/swap) and this gets
mounted before gconcat can do its magic. There's no "canmount=noauto"
setting, unless of course I were to disable "org.freebsd:swap=on" (which is
probably what needs to be done to make it work).

2. The more important problem however, is one of system freezes during
poudriere runs. 4 cores = 4 build jails; during build of larger ports (like
firefox, www/webkit-gtk*), if the number of such concurrent builds (large
ports) gets to 3, screens freeze, HDD activity continues but eventually
produces a kernel panic (I have tried to capture one on dumpdev but no
success).
If however, the 1G partition is added normally;
Device:       1024-blocks     Used:
/dev/zvol/tank0/swap   4194304    126932
/dev/ada0p1     1048576    126608
I do not get any system freezes and poudriere goes about its business
normally. For example:
	[01]: www/webkit-gtk3                  build           (00:16:45)
	[02]: x11/kdelibs4                     configure       (00:02:45)
	[03]: www/webkit-gtk2                  stage           (00:16:23)
	[04]: graphics/ImageMagick             install         (00:16:45)

Conclusion: gconcat or any other virtual device setup seems doomed to fail
and a "real device" seems what's needed for stability.




-----
FreeBSD-11-current_amd64_root-on-zfs_RadeonKMS
--
View this message in context: http://freebsd.1045724.n5.nabble.com/Questions-re-swap-on-zfs-tp5848720p5865165.html
Sent from the freebsd-fs mailing list archive at Nabble.com.


More information about the freebsd-fs mailing list