This post appeared originally in our sysadvent series and has been moved here following the discontinuation of the sysadvent microsite
Migrating disk volumes to new storage solutions can be a bit of a chore. Cloning file systems are usually done after taking your system offline in order to preserve consistency. But every so often you simply can’t afford to have your data or system unavailable for that period of time. So what to do?
Our problem
We had invested in new SAN storage solution and had to move the disk volumes for our virtual machines from the old SAN to the new one. The new solution did not have any handy migration tools ready, and scheduling downtime for each VM/server, for internal changes in infrastructure, was something we wanted to avoid presenting to our customers. Not to mention the overhead involved with stopping, migrating and starting thousands of VMs, most of which had inter-dependencies.
Software RAID to the rescue!
mdadm
mdadm is the tool used to manage Software RAID in Linux. It has the ability to
create a RAID that is not persistent, the super-block is kept in memory rather
then on the device.
The command will then take the form:
/sbin/mdadm --build /dev/DEVICE --level 1 --raid-disks=2 /dev/mapper/ORGINALDISK missing
/sbin/mdadm --add /dev/DEVICE /dev/mapper/NEWDISK
The “normal” way of creating a Software RAID is by using --create, but here
we used --build. See the man page of mdadm for details and warning.
Then when the Software RAID has finished the synchronization, the RAID device is stopped.
/sbin/mdadm --stop /dev/DEVICE
And the /dev/mapper/NEWDISK can be used instead of /dev/mapper/ORGINALDISK
We always have the opportunity to stop the Software RAID device and revert to
using the /dev/mapper/ORGINALDISK
SAN migration
Then it was just down to a few simple steps:
- Stop the Virtual Machine
- Create a Software RAID volume out of the original volume on the old SAN and an equal sized volume on the new SAN
- Start the Virtual Machine pointing to the Software RAID volume instead of the old SAN volume
- Wait for the synchronization to finalize
- Stop the Virtual Machine
- Stop the RAID device
- Start the Virtual Machine pointing to the volume on the new SAN
This can be scripted, so in wall clock time, it mounts to 2 reboots of the Virtual Machine.
Real example
As I do not have pastes from the actual migration, I ran through the steps on
normal LVM disk volumes.
Got some extra space on the rotating rust.
$ pvs
PV VG Fmt Attr PSize PFree
/dev/sda2 system lvm2 a-- 74.38g 5.00g
/dev/sdb slowhd lvm2 a-- 2.73t 1.38t
Make two volumes for the task.
$ lvcreate -n orgvolume -L 100G slowhd
Logical volume "orgvolume" created.
$ lvcreate -n destvolume -L 100G slowhd
Logical volume "destvolume" created.
$ mkfs.xfs /dev/slowhd/orgvolume
meta-data=/dev/slowhd/orgvolume isize=256 agcount=4, agsize=6553600 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=0 finobt=0
data = bsize=4096 blocks=26214400, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=0
log =internal log bsize=4096 blocks=12800, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
Then mount it and create some data on the disk.
$ mount /dev/slowhd/orgvolume /mnt/
$ dd if=/dev/urandom of=/mnt/somefile.data count=10 bs=1M
10+0 records in
10+0 records out
10485760 bytes (10 MB) copied, 0.514991 s, 20.4 MB/s
$ md5sum /mnt/somefile.data
598fab89fe76e6019be2593779eea1e8 /mnt/somefile.data
Then it is time for some mdadm stuff
$ umount /mnt
$ /sbin/mdadm --build /dev/md126 --level 1 --raid-disks=2 /dev/slowhd/orgvolume missing
mdadm: array /dev/md126 built and started.
$ /sbin/mdadm --add /dev/md126 /dev/slowhd/destvolume
mdadm: added /dev/slowhd/destvolume
$ cat /proc/mdstat
Personalities : [raid1]
md126 : active raid1 dm-17[2] dm-16[0]
104857600 blocks super non-persistent [2/1] [U_]
[>....................] recovery = 0.3% (355264/104857600) finish=39.2min speed=44408K/sec
unused devices: <none>
$ /sbin/mdadm -D /dev/md126
/dev/md126:
Version :
Creation Time : Thu Nov 26 14:00:34 2015
Raid Level : raid1
Array Size : 104857600 (100.00 GiB 107.37 GB)
Used Dev Size : 104857600 (100.00 GiB 107.37 GB)
Raid Devices : 2
Total Devices : 2
State : clean, degraded, recovering
Active Devices : 1
Working Devices : 2
Failed Devices : 0
Spare Devices : 1
Rebuild Status : 1% complete
Number Major Minor RaidDevice State
0 254 16 0 active sync /dev/dm-16
2 254 17 1 spare rebuilding /dev/dm-17
And the main feature here: It’s possible to use the RAID while synchronization goes on.
mount /dev/md126 /mnt/
And we add some data to the disk while it is being synchronized.
$ dd if=/dev/urandom of=/mnt/someotherfile.data count=100 bs=1M
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 5.00504 s, 21.0 MB/s
$ md5sum /mnt/someotherfile.data
c5dfaa813c47f998e8367a576e881a05 /mnt/someotherfile.data
md spends some time synchronizing the block devices. Eventually, it will complete:
$ cat /proc/mdstat
Personalities : [raid1]
md126 : active raid1 dm-17[2] dm-16[0]
104857600 blocks super non-persistent [2/2] [UU]
unused devices: <none>
$ /sbin/mdadm -D /dev/md126
/dev/md126:
Version :
Creation Time : Thu Nov 26 14:00:34 2015
Raid Level : raid1
Array Size : 104857600 (100.00 GiB 107.37 GB)
Used Dev Size : 104857600 (100.00 GiB 107.37 GB)
Raid Devices : 2
Total Devices : 2
State : clean
Active Devices : 2
Working Devices : 2
Failed Devices : 0
Spare Devices : 0
Number Major Minor RaidDevice State
0 254 16 0 active sync /dev/dm-16
2 254 17 1 active sync /dev/dm-17
Then we can change back to the new device
$ umount /mnt
$ /sbin/mdadm --stop /dev/md126
mdadm: stopped /dev/md126
$ mount /dev/slowhd/destvolume /mnt/
$ md5sum /mnt/some*
598fab89fe76e6019be2593779eea1e8 /mnt/somefile.data
c5dfaa813c47f998e8367a576e881a05 /mnt/someotherfile.data
We used this process to migrate a number of virtual servers from an old SAN to a new one. The downtime, for each virtual server, has been reduced to two reboots.
Thoughts on the CrowdStrike Outage
Unless you’ve been living under a rock, you probably know that last Friday a global crash of computer systems caused by ‘CrowdStrike’ led to widespread chaos and mayhem: flights were cancelled, shops closed their doors, even some hospitals and pharmacies were affected. When things like this happen, I first have a smug feeling “this would never happen at our place”, then I start thinking. Could it?
Broken Software Updates
Our department do take responsibility for keeping quite a lot ... [continue reading]