Re: Migrating to gmirrored RAID1
On 20.8.2008, at 10.48, DA Forsyth wrote: Good for you. I see I'm a bit late coming in with my advice, which is to follow the instructions at http://www.onlamp.com/pub/a/bsd/2005/11/10/FreeBSD_Basics.html?page=1 with a few mods because you (and I) were not building a new machine. This is the one I basically followed - I read both that one and the freebsddiary one. I decided to reinstall the server though, because I was any way going to reinstall world and lots of packages were outdated as well. Dumping and restoring would've worked as well, I think. One thing with gmirror is that it won't like to suffer repeated power failures. If the time between failures is less than the time it takes to rebuild the mirror the mirror will break and require manual fsck and then a rebuild. This happened to my test server a few weeks back but today I get to install a UPS for it so this should not happen again. OK, good to know. Luckily I have a small UPS which should provide five to ten minutes of full power. In Finland the power cuts are quite rare and usually measured in mere seconds than minutes, so it's sufficient most of the time and if not, apcupsd shuts the server down after few minutes if something bigger happens. -- Henry Karpatskij http://ripe.net/fcgi-bin/whois?searchtext=HK1203-RIPE ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Migrating to gmirrored RAID1
Thanks a lot folks, I managed to get the server working, only one file could not be read from the bad disk, but it wasn't required. It took hours to do it and it certainly will remind me next time I choose to be lazy and install a server to non-mirrored disk. :-) -- Henry Karpatskij http://ripe.net/fcgi-bin/whois?searchtext=HK1203-RIPE ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Migrating to gmirrored RAID1
Hi, I have a failing IDE disk which is running my 7.0-p1 server. I've been investigating the possible solutions and I've decided to go with two new IDE disks and gmirror. However, I'm not too familiar with disk internals, I know how to install the system and somehow understand the concept of slices and partitions, but that's about it. I found some examples on how to install the gmirror on a running system, but they all have in common that they just add new spare disk to the system and turn on the mirroring to it, but I need to replace the current disk which is not the same size as the new ones. Any suggestions how one would do such an operation? Should I just re- install the server to a new disk, turn on the mirroring and then restore the configuration and files from the failing disk? Or is it easier to add the disks to the running system, turn on mirroring and then somehow dump the current disk to the mirror and then re-configure it to boot from the gmirror and remove the failing disk? Current df output: Filesystem 1K-blocksUsedAvail Capacity Mounted on /dev/ad0s1a507630 159262 30775834%/ devfs 1 10 100%/dev /dev/ad0s1e507630 56 466964 0%/tmp /dev/ad0s1f 33573476 6044408 2484319020%/usr /dev/ad0s1d 1762414 381632 123979024%/var devfs 1 10 100%/var/named/dev Thanks in advance, -- Henry Karpatskij http://ripe.net/fcgi-bin/whois?searchtext=HK1203-RIPE ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Migrating to gmirrored RAID1
Henry Karpatskij wrote: Hi, I have a failing IDE disk which is running my 7.0-p1 server. I've been investigating the possible solutions and I've decided to go with two new IDE disks and gmirror. However, I'm not too familiar with disk internals, I know how to install the system and somehow understand the concept of slices and partitions, but that's about it. I found some examples on how to install the gmirror on a running system, but they all have in common that they just add new spare disk to the system and turn on the mirroring to it, but I need to replace the current disk which is not the same size as the new ones. Any suggestions how one would do such an operation? Should I just re-install the server to a new disk, turn on the mirroring and then restore the configuration and files from the failing disk? Or is it easier to add the disks to the running system, turn on mirroring and then somehow dump the current disk to the mirror and then re-configure it to boot from the gmirror and remove the failing disk? If the new drives have the same size as the old one, you could do it gradually: - Mirror the old drive to new drive #1 - Replace the old drive with new drive #2, mirror it from new drive #1 If the sizes are different, this won't work and the best would be to install a fresh system on a new drive and transfer the configuration from the old one. Booting will always happen from a single drive. Since up to the point file systems are mounted all relevant drive access is read-only, the BIOS will happily boot one of the mirrored pair, then the kernel will pick up and activate mirroring before file systems are mounted. This is why all instructions start with one drive and later add a new drive to hold the mirror; it's easy to set up a single drive and then simply mirror it. signature.asc Description: OpenPGP digital signature
Re: Migrating to gmirrored RAID1
On Monday 18 August 2008 05:39:10 am Henry Karpatskij wrote: Hi, I have a failing IDE disk which is running my 7.0-p1 server. I've been investigating the possible solutions and I've decided to go with two new IDE disks and gmirror. However, I'm not too familiar with disk internals, I know how to install the system and somehow understand the concept of slices and partitions, but that's about it. I found some examples on how to install the gmirror on a running system, but they all have in common that they just add new spare disk to the system and turn on the mirroring to it, but I need to replace the current disk which is not the same size as the new ones. Any suggestions how one would do such an operation? Should I just re- install the server to a new disk, turn on the mirroring and then restore the configuration and files from the failing disk? Or is it easier to add the disks to the running system, turn on mirroring and then somehow dump the current disk to the mirror and then re-configure it to boot from the gmirror and remove the failing disk? I think the latter approach is easier and makes the most sense for your situation. Install the disks, set up the mirror(s) that you want, transfer data and then do a boot test. Something along these lines should work. Substitue device and volume names to match your hardware and tastes. #set up a single mirror to use the whole disk (versus mirroring individual slices/partitions) gmirror label myraid1 /dev/ad4 /dev/ad6 #install a partition table and the boot0 code fdisk -BI /dev/mirror/myraid1 #install a default label and the main boot code bsdlabel -wB /dev/mirror/myraid1s1 #create BSD partitions by hand. remember to set EDITOR if you don't like vi bsdlabel -e /dev/mirror/myraid1s1 #This is the tricky part. Create the partitions you want on the mirror. Use the output of bsdlabel /dev/ad0s1 as a guide. Remember that a should be root, b is traditionally swap, c is the raw partition and should not be changed, and d - h are other partitions. I find a spreadsheet to be handy for figuring out the correct values, though a calculator is adequate (I've used dc more than once..). The units you are dealing with are 512-byte sectors. Best practice (which sysinstall doesn't follow but bsdlabel -w does) is to leave 16 sectors at the start of the slice for the boot code (but both swap and UFS will avoid clobbering it even if you don't do this). If you follow the best practice and do the partitions in order, then the offset for a is 16, and the offset for any other partition is the offset of the previous one plus the size of the previous one. Assuming your last filesystem wants to use the remainder of the slice, figure its offset as above then subtract it from the total (the size of c) for the size. For filesystem partitions the fstype should be 4.2BSD, and use 2048 16384 0 for the last three columns unless you have reason to do otherwise. (The bps is recalculated when you create a filesystem so it won't be 0 later. That's expected.) The fstype for swap space is swap and the last three columns are omitted. Save and exit the editor when finished.) #Create filesystems newfs /dev/mirror/myraid1s1a #(repeat for other filesystems, changing the partition letter as appropriate) #Make temp mountpoints mkdir /newroot #(again repeat as needed) #Mount new filesystems mount /dev/mirror/myraid1s1a /newroot #(repeat as needed) #Dump/restore filesystems cd /newroot dump -0 -L -C32 -f - / | restore -r -f - rm restoresymtable #(repeat as needed, changing the filesystem argument to dump and the cwd for your new filesystems. one or two messages from restore about getting a different inode than expected is normal.) #edit /newroot/etc/fstab. Change the device for / to /dev/mirror/myraid1s1a. Make a similar change for other filesystems. #edit /newroot/boot/loader.conf. Make sure it includes this line: geom_mirror_load=YES #shut down, remove the original disk, and try booting Good luck! JN Current df output: Filesystem 1K-blocksUsedAvail Capacity Mounted on /dev/ad0s1a507630 159262 30775834%/ devfs 1 10 100%/dev /dev/ad0s1e507630 56 466964 0%/tmp /dev/ad0s1f 33573476 6044408 2484319020%/usr /dev/ad0s1d 1762414 381632 123979024%/var devfs 1 10 100%/var/named/dev Thanks in advance, ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
