Az, Currently, /dev/md/? devices are stopped before LVM. This causes a complaint when you shut the machine down (can't stop /dev/md/0, device busy -- 'cause LVM is still using it). So LVM should probably be stopped before RAID devices.
Kai Lindenberg <kai@ldbg.de> writes: > Hi Gentooguys, > > I'm very happy with my Gentoo 1.4rc1, but can you tell me, why we do > first load lvm and then start md, repectively first shut down md > and then lvm in /etc/init.d/{checkfs,halt.sh}? > One of my VGs is on md, so I have to exchange the lvm start/stop > with those of md and, that's the business, I have have to control > the update of those files. > I wonder if there is any reason for this way. > > Thanks Kai
Ok, so I guess nobody will ever have raid devices on LVM ?
hmmmmm good question
Maybe use raid autostart ?
raid autostart: yes this works, but the shutdown is a problem.
how do other distros handle this? I'm guessing they do: check for any LVM volumes are part of raidsets (via some nasty awk command) shutdown raidsets that are found shutdown LVM shutdown any other md's. Would that work? -eric (who is currently having problems with this same bug, but who only has LVM in raid, and no raid in LVM...)
MDK for one only stop LVM .. they leave RAID for the kernel I guess ...
Recap: LVM and RAID are started AND stopped in the wrong order. There is also a BUG in the AWK command that searches for mounted filesystems. If you have a VG named 'rootvg' the current awk command in 'halt.sh' -- awk '!/(^#|proc|devfs|tmpfs|^none|^\/dev\/root|[[:space:]]\/[[:space:]])/ {print $2}' /proc/mounts |sort -r -- causes all LV's in rootvg to stay mounted at shutdown. Take out the '^\/dev/root|' and it works properly. Is there any logical reason to omit any FS that starts with 'root'?
BTW, why not just switch the order around? I did it in checkfs and halt.sh and it works fine... I don't see why this problem needs to hang around for another four months even IF my suggestion turns out to be a bandaid :)
Because the halt.sh and checkfs in baselayout 1.8.6.3 looks totally different ?
No chance for a 1.8.5.9 that will fix the problem until 1.8.6.3 is ready to go? How would it cause problems with later versions if you fixed two bugs that are in a version that is completely different? Yes, I have no clue how the dev process works. I'm just trying to understand the reasoning behind your answer.
This bug's been inactive for at least 180 days. baselayout-1.8.6.10-r1 is currently stable. Is this bug still relevant? Please reopen if the problem persists.
