Re: mdadm 2.5.5 external bitmap assemble problem
From: dean gaudet <hidden>
Date: 2006-11-06 08:26:16
On Mon, 6 Nov 2006, Neil Brown wrote:
quoted
hey i have another related question... external bitmaps seem to pose a bit of a chicken-and-egg problem. all of my filesystems are md devices. with an external bitmap i need at least one of the arrays to start, then have filesystems mounted, then have more arrays start... it just happens to work OK if i let debian unstale initramfs try to start all my arrays, it'll fail for the ones needing bitmap. then later /etc/init.d/mdadm-raid should start the array. (well it would if the bitmap= in mdadm.conf worked :) is it possible to put bitmaps on devices instead of files? mdadm seems to want a --force for that (because the device node exists already) and i haven't tried forcing it. although i suppose a 200KB partition would be kind of tiny.... but i could place the bitmap right beside the external transaction log for the filesystem on the raid5.Create the root filesystem with --bitmap=internal, and store all the other bitmaps on that filesystem maybe?
yeah i only have the one external bitmap (it's for a large raid5)... so things will work fine once i apply your patch. thanks.
I don't know if it would work to have a bitmap on a device, but you can always mkfs the device, mount it, and put a bitmap on a file there??
yeah this was the first thing i tried after i found mdadm -b /dev/foo wasn't accepted... without modifying startup scripts there's no way to use any filesystem other than root... it's just due to ordering of init scripts: # ls /etc/rcS.d | grep -i 'mount\|raid' S02mountkernfs.sh S04mountdevsubfs.sh S25mdadm-raid S35mountall.sh S36mountall-bootclean.sh S45mountnfs.sh S46mountnfs-bootclean.sh i'd need to run another mdadm-raid after the S35mountall, and then another mountall. anyhow, i don't think you need to change anything (except maybe a note in the docs somewhere), i'm just bringing it up as part of the experience of trying external bitmap. i suspect that in the wild and crazy direction debian and ubuntu are heading (ditching sysvinit for event-driven systems) it'll be "easy" to express the boot dependencies. -dean