Bug#633024: initramfs-tools: Race condition when root filesystem is on a LV and mdadm array
On Mon, 11 Jul 2011, Rain Maker wrote: > - The rootdelay=X option does as advertised: "Also it adds time to > scsi device to settle before calling mdadm or lvm thus excluding > potential races.". > In this situation, I seem to need a delay between assembling mdadm and > activating the VG's. sure this *is* the recommended workaround for such races. such timing bugs occure in many different linux subsystems. > - scsi_mod.scan=sync does not introduce any visual delay, but doesn't > help me either. of course as no scsi disks are involved. > - I've also tried a couple of break= commands, but there doesn't seem > to be a breakpoint between mdadm and LVM. the breaks only help to debug stuff inbetween aka for manual interaction. so false critic. > Thank you for the suggestion. thank you. -- maks -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#633024: initramfs-tools: Race condition when root filesystem is on a LV and mdadm array
2011/7/8, maximilian attems : > On Fri, 08 Jul 2011, Roel Brook wrote: > >> When booting the system, it seems there is a race condition between >> assembling >>.. >> "lvm vgdisplay" after running a "lvm vgchange -ay" > > tried to use rootdelay=X with a small value? > see http://wiki.debian.org/InitramfsDebug > > -- > maks > I tried the suggestions on the page you metioned. Unfortunatly, they do not help in this situation. - The rootdelay=X option does as advertised: "Also it adds time to scsi device to settle before calling mdadm or lvm thus excluding potential races.". In this situation, I seem to need a delay between assembling mdadm and activating the VG's. - scsi_mod.scan=sync does not introduce any visual delay, but doesn't help me either. - I've also tried a couple of break= commands, but there doesn't seem to be a breakpoint between mdadm and LVM. Thank you for the suggestion. Roel -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#633024: initramfs-tools: Race condition when root filesystem is on a LV and mdadm array
On Fri, 08 Jul 2011, Roel Brook wrote: > When booting the system, it seems there is a race condition between assembling > a mdadm array, and scanning and / or activating VGs which have it's PV on this > array. > > I have a mdadm array (ICH9R fakeraid BIOS array), where one of the partitions > is a PV. On this PV, there is a LV for the root filesystem. > > Since ~ 2.6.37, the initramfs does not find the VG all the time. This means, I > have to boot the system a couple of times, before the VG is found, and the > root > filesystem is mounted. > > There is no real logic in how often initramfs succesfully finds the VG. It's > about a 1 in 3 chance of it booting succesfull. However, sometimes I have to > reboot the system about 10 times before the VG is finally found. Sometimes it > finds it immediatly. > > When in the shell for the initramfs, the RAID array seems to be assembled fine > (this is also reported in the console output). Also, when executing a "lvm > vgdisplay", the VG is there. Once the VG is activated, the LV's can be mounted > fine. > > Attached are the dmesg output, the contents of /proc/mdstat and the output of > "lvm vgdisplay" after running a "lvm vgchange -ay" tried to use rootdelay=X with a small value? see http://wiki.debian.org/InitramfsDebug -- maks -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
