Raid

From SME Server
Warning.png Warning:
Please read this article before buying and deploying drives. https://raid.wiki.kernel.org/index.php/Timeout_Mismatch

The new type of SMR drives are NOT suitable for RAID arrays. Beware WD Red NAS drives, though recently they have made it clearer which models use SMR.

A drive failure can corrupt an entire array: RAID does not replace backup!



Important.png Note:
SME Servers RAID Options are largely automated, but even with the best laid plans things don't always go according to plan. See also: Raid:Manual Rebuild, Raid:Growing and Hard Disk Partitioning. There is a wiki on the Linux software raid, you will find many cool Tips here


Hard Drives

A software RAID array will be automatically configured as part of the installation process for servers which contain multiple hard drives. This is to ensure redundancy, so if one disk fails the system will still function.


Important.png Note:
As per the release notes, SME Server 10 RAID configuration is slightly different to previous versions. See Default RAID Rationale below for more details.


The specifics of the RAID setup depends on the number of drives available, to balance redundancy and capacity.

The root and swap volumes are configured using LVM on the RAID device /dev/md1 as follows:

  • 1 drive - no RAID
  • 2 drives - RAID 1
  • 3 drives - RAID 1 + hot spare
  • 4 drives - RAID 6
  • 5+ drives - RAID 6 + hot spare

The /boot volume and EFI partition if necessary is always a non-LVM RAID 1 array on the device /dev/md0.

If you use a hardware RAID controller to manage your drives, this should be configured to present a single volume, which SME will configure without software RAID.


Default RAID Rationale

The differences in RAID layout between SME Server 10 and previous versions is summarised below:

Number of Drives SME Server 10 Previous Versions
1 No software RAID Degraded RAID 1
2 Software RAID 1
3 RAID 1 + hot spare
4 RAID 6 RAID 5 + hot spare
5 RAID 6 + hot spare
6
7+ RAID 6 + hot spare

The main differences are no degraded RAID 1 for a single disk install, which better supports virtualised and hardware RAID use cases, and a preference for RAID 6 over RAID 5. This is to reduce the risk of a single disk failure bringing down the array. While consumer hard drives have got significantly larger over time, their unrecoverable read error rate (URE) has remained at 1 per 10^14 bits, or 12TB. As an example, imagine a server with 5 x 4TB drives. Under previous versions of SME Server this would have been configured as a 4 disk RAID 5 array with 1 hot spare. If one drive failed, the hot spare would become active and the array would begin to rebuild. This would require reading all 3 disks and, at some point during that 12TB operation, it’s very likely that an unrecoverable error would be encountered. At this point, the whole array would fail. In comparison, a RAID 6 array is tolerant to two disk failures. While this does not entirely solve the risk of a URE during rebuild, it significantly reduces the likelihood of it taking down the array. Note: RAID is a convenient method of protecting server availability from a drive failure. It does not remove the need for regular backups, which can be configured using the Server Manager.

Disk Layout

Mirroring drives in the same IDE channel (eg. hda and hdb) is not desirable. If that channel goes out, you may lose both drives. Also, performance will suffer slightly.

The preferred method is to use the primary location on each IDE channel (eg. hda and hdc). This will ensure that if you lose one channel, the other will still operate. It will also give you the best performance.

In a 2 drive setup put each drive on a different IDE channel:

IDE 1 Primary - Drive 1
IDE 1 Secondary - CDROM
IDE 2 Primary - Drive 2

Obviously this section is obsolete with SATA hard drives because each disk has its own channel.


Identifying Hard Drives

It may not always be obvious which physical hard drive maps to which logical device. The first step would be to be able to identify all block devices present on your server. This could be done by using two commands

 lsblk

or the following

 findmnt


Then, once you identified a block device , the simplest method to verify which physical drive it is, is using S.M.A.R.T. capability to map the serial number on the physical package with that displayed by smartctl. Assuming the device of interest is sda, (a SCSI drive), then you would issue the following command as root:

smartctl -i /dev/sda

Or if an IDE Drive

smartctl -i /dev/hda


Adding Additional Drives

For servers which were installed with 2+ drives and have a working RAID array, it is possible to add an additional drive which will become a hot spare, ready to be activated in case of drive failure.

Ensure that any new drives are the same size or larger than your existing drives.

  • Shut down the machine
  • Install one additional drive at a time
  • Boot up
  • At the login prompt log on as admin with the root password to get to the admin console
  • Go to #5 Manage disk redundancy
  • Accept the option to add an additional drive

If the Manage disk redundancy page displays the message "The free disk count must equal one" and "Manual intervention may be required", then you probably have additional hard drives that need to be disconnected while the RAID is set up. An external USB drive will have this effect, and should be unplugged.

Reusing Hard Drives

  • MBR formatted disks

If it was ever installed on a Windows machine, or any of the *BSDs, (or in some cases an old system with RAID and/or LVM) then you will need to clear the MBR first before installing it.

From the linux command prompt, type the following:

#dd if=/dev/zero of=/dev/hdx bs=512 count=1

or

#dd if=/dev/zero of=/dev/sdx bs=512 count=1

You MUST reboot so that the empty partition table gets read correctly.

For more information, check: http://bugs.contribs.org/show_bug.cgi?id=2154


  • For disks previously formatted as GPT this is insufficient. It's probably best to use gdisk or parted or partx to delete the partitions; there are other tools that will work. Parted has limited support for LVM.

Upgrading the Hard Drive Size

Note: these instructions are only applicable if you have a RAID system with more than one drive. They are not applicable to a single-drive RAID 1 system, and increasing the useable space on such a system by cloning the existing single drive to a larger drive is not supported. See http://bugs.contribs.org/show_bug.cgi?id=5311

  • CAUTION MAKE A FULL BACKUP!
  • Ensure you have e-smith-base-4.16.0-33 or newer installed. [or Update to at least 7.1.3]

HD Scenario - Current 250gb drives, new larger 500gb drives

  1. Shut down and install one larger drive in system for one old HD. Unplug any USB-connected drives.
  2. Boot up and login to the admin console and use option 5 to add the new (larger) drive to system.
  3. Wait for raid to fully sync.
  4. Repeat steps 1-3 until all drives in system are upgraded to larger capacity.
  5. Ensure all drives have been replace with larger drives and array is in sync and redundant!
  6. Issue the following commands:


Important.png Note:
SME9 uses /dev/md1 not /dev/md2.


mdadm --grow /dev/md2 --size=max
pvresize /dev/md2
lvresize -l +100%FREE main/root
ext2online -C0 /dev/main/root   

In the last command above, the -C0 is: dash C zero

If you receive an "command not found" error, try this:

resize2fs /dev/mapper/main-root &

TIP: I put an "&" at end to allow it to run in background even if I close ssh session.


Notes :

  • All of this can be done while the server is up and running with the exception of #1.
  • These instructions should work for any raid level you have as long as you have >= 2 drives
  • If you have disabled lvm , you don't need the pvresize or lvresize command, therefore the final line becomes
ext2online -C0 /dev/md2 #(or whatever / is mounted to)

or If you receive an "command not found" error, try this:

resize2fs /dev/md2 &

Replacing and Upgrading a Hard Drive after HD fail

Note: See Bugzilla: 6632 and Bugzilla:6630 a suggested sequence for Upgrading a Hard Drive size is detailed after issue when attempting to sync a new drive when added first as sda.

Note: These instructions are applicable if you have a faulty HD on a RAID system with more than one drive and intend to upgrade the sizes as well as replacing the failed HD. They are not applicable to a single-drive RAID 1 system, and increasing the useable space on such a system by cloning the existing single drive to a larger drive is not supported. See http://bugs.contribs.org/show_bug.cgi?id=5311

  • CAUTION MAKE A FULL BACKUP!
  • Ensure you have e-smith-base-4.16.0-33 or newer installed. [or Update to at least 7.1.3]

HD Scenario - Current 250gb drives, new larger 500gb drives

  1. Remove failed HDD from system, ensure remaining drive is on sda on its own and boot up.
  2. Shutdown, connect one new 500gb drive as sdb and boot up
  3. Login to the admin panel and manage raid to add new (larger) drive to system.
  4. Wait for raid to fully sync
  5. Do full reboot with those 2 drives in place (1 original, 1 new)
  6. Shutdown again, disconnect the original drive, and connect the new drive just sync'd as sda (in place of original)
  7. Boot up again with just the one new drive in place, and confirm it boots OK.
  8. Shutdown, and connected the other 500gb drive as sdb
  9. Boot up login to admin panel and add sdb to the array, and wait for raid to fully sync.
  10. Reboot and ensure all drives have been replaced with larger drives and array is in sync and redundant!
  11. Issue the following commands:


Important.png Note:
SME9 uses /dev/md1 not /dev/md2.


mdadm --grow /dev/md2 --size=max
pvresize /dev/md2
lvresize -l +100%FREE main/root
ext2online -C0 /dev/main/root   

In the last command above, the -C0 is: dash C zero

If you receive an "command not found" error, try this:

resize2fs /dev/mapper/main-root &

TIP: I put an "&" at end to allow it to run in background even if I close ssh session.

Notes :

  • These instructions should work for any raid level you have as long as you have >= 2 drives
  • If you have disabled lvm , you don't need the pvresize or lvresize command, therefore the final line becomes
ext2online -C0 /dev/md2 #(or whatever / is mounted to)

or If you receive an "command not found" error, try this:

resize2fs /dev/md2 &

RAID Notes

Many on-board hardware raid cards are in fact software RAID. Turn it off as cheap "fakeraid" cards aren't good for Linux. You will generally get better performance and reliability with Linux Software RAID (http://linux-ata.org/faq-sata-raid.html). Linux software RAID is fast and robust.

If you are insistent on getting a hardware RAID, buy a well supported RAID card which has a proper RAID BIOS. This hides the disks and presents a single disk to Linux (http://linuxmafia.com/faq/Hardware/sata.html). Please check that it is supported by the kernel and has some form of management. Also avoid anything which requires a driver. Try googling for the exact model of RAID controller before buying it. Please note that you won't get a real hardware raid controller cheap.

It rarely happens, but sometimes when a device has finished rebuilding, its state doesn't change from "dirty" to "clean" until a reboot occurs. This is cosmetic

Periodic scrub of RAID arrays

A Periodic scrub of RAID arrays (weekly raid check) is performed every week on Sunday at 04:22 local time, refer Bugzilla:3535 and Bugzilla:6160 for more information.

Theses operations are logged, however, no emails will be sent to admin as of the release of packages associated with Bug #6160 or the release of the 8.1 ISO.

Receive periodic check of RAID by email

There are routines in SME Server to check the RAID and sent mail to the admin user, when the RAID is degraded or when the RAID is resynchronizing. But the admin user may receive a lot of emails and sometimes messages can be forgotten. So the purpose is to have a routine which sends email to the user of your choice each week.

nano /etc/cron.weekly/raid-status.sh

You have to change the variable DEST="stephane@your-domaine-name.org" to the email you decide to use.

#!/bin/sh
# cron.weekly/mdadm-status -- weekly status of the RAID
# 2013 Pierre-Alain Bandinelli
# distributed under the terms of the Artistic Licence 2.0

# Get status from the RAID array and send the details by email.
# Email will go to the address specified in the commandline.
set -eu

MDADM=/sbin/mdadm
[ -x $MDADM ] || exit 0 # package may be removed but not purged

DEST="stephane@your-domaine-name.org"
exec $MDADM --detail  $(ls /dev/md*) | mail -s "RAID status SME Server" $DEST

save by ctrl+x

chmod +x /etc/cron.weekly/raid-status.sh

each sunday a 4h00 AM you will receive a mail which looks to this :

/dev/md1:
       Version : 0.90
 Creation Time : Sun Jan  6 20:50:41 2013
    Raid Level : raid1
    Array Size : 104320 (101.89 MiB 106.82 MB)
 Used Dev Size : 104320 (101.89 MiB 106.82 MB)
  Raid Devices : 2
 Total Devices : 2
Preferred Minor : 1
   Persistence : Superblock is persistent

   Update Time : Sun Dec 22 04:22:42 2013
         State : clean
Active Devices : 2
Working Devices : 2
Failed Devices : 0
 Spare Devices : 0

          UUID : 28745adb:d9cff1f4:fcb31dd8:ff24cb0c
        Events : 0.208

   Number   Major   Minor   RaidDevice State
      0       8        1        0      active sync   /dev/sda1
      1       8       17        1      active sync   /dev/sdb1
/dev/md2:
       Version : 0.90
 Creation Time : Sun Jan  6 20:50:42 2013
    Raid Level : raid1
    Array Size : 262036096 (249.90 GiB 268.32 GB)
 Used Dev Size : 262036096 (249.90 GiB 268.32 GB)
  Raid Devices : 2
 Total Devices : 2
Preferred Minor : 2
   Persistence : Superblock is persistent

   Update Time : Sun Dec 22 05:30:36 2013
         State : clean
Active Devices : 2
Working Devices : 2
Failed Devices : 0
 Spare Devices : 0

          UUID : c343c79e:91c01009:fcde78b4:bad0b497
        Events : 0.224

   Number   Major   Minor   RaidDevice State
      0       8        2        0      active sync   /dev/sda2
      1       8       18        1      active sync   /dev/sdb2

If you want to test the message sent without waiting the next sunday, you can do

/etc/cron.weekly/raid-status.sh

nospare

If you use the commandline parameter nospare during installation ("sme nospare"), the system will still count the missing spare towards the number of drives. A system with 6 physically present harddrives thus will be formated Raid6 _not_ Raid5. Resulting capacity of course will be "n-2". Note: with the release of version 7.6 and 8.0, the commandline parameter "sme nospare" has been changed to "sme spares=0" . In addition, you may also select the number of spare(s) implemented [0,1,or 2].

remove the degraded RAID message

When you install the smeserver with one drive with a degraded raid, you will see a 'U_' state but without warnings. If you want to leave just one 'U' in the /proc/mdstat and stop all future questions about your degraded raid state, then :

mdadm --grow /dev/md0 --force --raid-devices=1
mdadm --grow /dev/md1 --force --raid-devices=1

after that you will see this

# cat /proc/mdstat
Personalities : [raid1] 
md0 : active raid1 sda1[0]
     255936 blocks super 1.0 [1/1] [U]
     
md1 : active raid1 sda2[0]
     268047168 blocks super 1.1 [1/1] [U]
     bitmap: 2/2 pages [8KB], 65536KB chunk

unused devices: <none>

Resynchronising a Failed RAID

Information.png Tip:
You can refer to 'man mdadm' or Mdadm Man page or Raid:Manual_Rebuild


Sometimes a partition will be taken offline automatically. Admin will receive an email DegradedArray event on /dev/md2.


Important.png Note:
This will happen if, for example, a read or write error is detected in a disk in the RAID set, or a disk does not respond fast enough, causing a timeout. As a precaution, verify the health of your disks as documented in: Monitor_Disk_Health and specifically with the command:


smartctl -a /dev/hda

Where hda is the device to be checked; check all of them.

You may check the health of your array using the Admin Console. Login as root, type console. Select Item 5. "Manage disk reduncancy"

--------Disk Reduncancy status as of Thursday Dec 22 -------
             	Current RAID status:
             
             	Personalities : [raid1]
             		md2 : active raid1 hda2[0] <-- NOTICE hdb2[#] is missing. Means hdb2[#] failed.	
                                     38973568 blocks [2/1] [U_]
             
             		md1 : active raid1 hda1[0] hdb1[1]
                   			104320 blocks [2/2] [UU]
             
             			unused devices: <none>
             	Only Some of the RAID devices are unclean.  <-- NOTICE This message and 
             	Manual intervention may be required. <-- this message.

Notice the last 2 sentences of the window above. You have some problems.
If your system is healthy however the message you will see at the bottom of Raid Console window is:

All RAID devices are in clean state

If you have no software RAID devices you will see the message at the bottom of the Console window:

Your system only has a single disk drive installed or is using hardware 
mirroring. If you would like to enable software mirroring, please shut
down, install a second disk drive (of the same capacity) and then return
to this screen.

Additionally, the details of the raid can be seen by inspecting the mdstat file from the shell prompt.

[root@sme]# cat /proc/mdstat
Personalities : [raid1]
md1 : active raid1 hda3[0] hdb3[1]
     38837056 blocks [2/2] [UU]

md2 : active raid1 hdb2[1]        <--    Shows current active partition - note there is one missing
     1048704 blocks [2/1] [_U]    <--    '_' = partition missing from array

md0 : active raid1 hda1[0] hdb1[1]
     255936 blocks [2/2] [UU]

Make a note of the raid partition that has failed, shown by [_U]
In this case it is md2, the device being /dev/md2.

The failed drive partition is indicated by the '_' underline character. In the above example _U indicates that the first drive partition on md2, (Multi-Device 2) has failed. The second drive partition on md2, symbolized by the character 'U' is still part of the md2. If the second drive partition had failed, that is hdb2 then the details would be reversed. E.g. [U_] . Placing the _ Underline second in the details.

Determine the missing physical partition, Look carefully at the sample above and fill in the gap for which drive is missing.
In this example, it's hda2, the device being /dev/hda2

md1 : active raid1 hda3[0] hdb3[1]
md2 : active raid1 hda2[0] hdb2[1]   <--- In the above sample hda2[0] is missing
md0 : active raid1 hda1[0] hdb1[1]

If the raid has a failed disk that has not yet been kicked out of the array then mdstat will show something like the following:

md2 : active raid1 hda2[0](F) hdb2[1]   <--    Shows current active partition - with one FAILED (F)
     1048704 blocks [2/1] [_U]          <--    '_' = partition missing from array

In this case before you add the disk back in you will need to remove the disk as per:

[root@sme]# mdadm --remove /dev/md2 /dev/hda2

However if the drive has already been removed by the operating system then removing the drive is unnecessary. To determine this use the command:

mdadm --query --detail /dev/md2

Of course use the proper md# based on your configuration. This command will give you several lines of data, including the size of the array. Near the end of the output you will see the following if the drive has been removed already. There is no need to remove the drive since it has already been removed.

   Number   Major   Minor   RaidDevice State
      0       3        2        0      active sync   /dev/hda2
      1       0        0        -      removed      <-- NOTE THIS 


To add the physical partition back and rebuild the raid partition.

[root@sme]# mdadm --add /dev/md2 /dev/hda2

Once you type the command the following message will appear, appropriate for your device.

 [root@sme]  mdadm: hot added /dev/hda2

It important to know that your devices are likely to be different, E.G your device could be /dev/sda2 or you may have more than two disks, including a hot standby. These details can always be determined from the mdstat file. Once the raid resync has been started, the progress will be noted in mdstat. You can see this real time by:

[root@sme]# watch -n .1 cat /proc/mdstat

or you can see this in a snapshot by:

[root@sme]# cat /proc/mdstat
Personalities : [raid1]
md1 : active raid1 hda3[0] hdb3[1]
      38837056 blocks [2/2] [UU]

md2 : active raid1 hda2[2] hdb2[1]
      1048704 blocks [2/1] [_U]
      [=>...................]  recovery =  6.4% (67712/1048704) finish=1.2min speed=13542K/sec
md0 : active raid1 hda1[0] hdb1[1]
      255936 blocks [2/2] [UU]

When recovery is complete, the partitions will all be up:

[root@sme]# cat /proc/mdstat
Personalities : [raid1]
md1 : active raid1 hda3[0] hdb3[1]
      38837056 blocks [2/2] [UU]

md2 : active raid1 hda2[0] hdb2[1]
     1048704 blocks [2/2] [UU]

md0 : active raid1 hda1[0] hdb1[1]
      255936 blocks [2/2] [UU]

If this action is required regularly, you should test your disks for SMART errors and physical errors, check your disk cables, and make sure no two hard drives share the same IDE port. See also: http://wiki.contribs.org/Monitor_Disk_Health

Also check your driver cards, since a faulty card can destroy the data on a full RAID set as easily as it can a single disk.


Information.png Tip:
we could use a shortcut for the raid rebuild :
mdadm -f /dev/md2 /dev/hda2 -r /dev/hda2 -a /dev/hda2


Convert Software RAID1 to RAID5

Important.png Note:
these instructions are only applicable if you have SME8 or greater and a RAID1 system with 2 hd in sync; new drive(s) must be of the same size or larger as the current drive(s)


Warning.png Warning:
Please make a full backup before proceeding


Warning.png Warning:
Newer versions of mdadm use the v1.x superblocks stored at the beginning of the block device, which could overwrite the filesystem metadata. You’ll need to be starting with a v0.9 metadata device for the above instructions to work (which was the default for years).First, check the existing superblock version with:

mdadm –detail /dev/md0

Then, when re-creating the RAID 5 array, make sure you add the –metadata=0.9 tag so the superblock is recreated in the right place. Unfortunately, v1.0 give a new size for the md device (smaller than the original array), v1.1 and v1.2 corrupts the filesystem outright, so best to avoid these cases entirely. Creating a new array with v1.x superblocks when the original was v0.9 is likewise outright destructive.


  1. Login as root
  2. Move to /boot (we must create a new initrd image to load raid5 driver).
    cd /boot
    
  3. Make a backup copy
    mv initrd-`uname -r`.img initrd-`uname -r`.img.old
    
  4. Create the new image
    mkinitrd --preload raid5 initrd-`uname -r`.img `uname -r`
    
  5. Shut down and install new drive(s) in system.
  6. Boot up with SME cd and enter the rescue mode.
    sme rescue
    
  7. Skip network setup.
  8. Skip mounting the current SME installation.
  9. Now, create on the new drive(s) the correct partition table.
    sfdisk -d /dev/sda > tmp.out
    

    sfdisk /dev/sdc < tmp.out

  10. Repeat the last step for each new hd (sdd, sde etc.).
  11. Create the new array
    mdadm --create /dev/md2 -c 256 --level=5 --raid-devices=2 /dev/sda2 /dev/sdb2
    

    mdadm: /dev/sda2 appears to be part of a raid array: level=raid1 devices=2 ctime=Fri Dec 18 13:17:49 2009 mdadm: /dev/sdb2 appears to be part of a raid array: level=raid1 devices=2 ctime=Fri Dec 18 13:17:49 2009 Continue creating array? y mdadm: array /dev/md2 started.

  12. Wait for resync; monitor the status with
    cat /proc/mdstat
    

    root# cat /proc/mdstat Personalities : [raid0] [raid1] [raid5] md2 : active raid5 sdb1[2] sda1[0] 1048512 blocks level 5, 256k chunk, algorithm 2 [2/1] [U_] [==>..................] recovery = 12.5% (132096/1048512) finish=0.8min speed=18870K/sec

  13. Reboot
    exit
    
  14. Login as root
  15. Add the new drives to the array
    mdadm --add /dev/md2 /dev/sdc2
    
  16. Repeat the last step for each new hd (sdd2, sde2 etc.)
  17. Grow the array
    mdadm --grow /dev/md2 --raid-devices=N
    
  18. N is the total number of drives: minimum is 3
  19. Wait for array reshaping. This part can take a substantial amount of time; monitor it with
    cat /proc/mdstat
    

    root# cat /proc/mdstat Personalities : [raid0] [raid1] [raid5] md2 : active raid5 sdc1[2] sdb1[1] sda1[0] 1048512 blocks super 0.91 level 5, 256k chunk, algorithm 2 [3/3] [UUU] [==>..................] reshape = 12.5% (131520/1048512) finish=2.5min speed=5978K/sec

  20. Issue the following commands:
    pvresize /dev/md2
    

    lvresize -l +100%FREE main/root resize2fs /dev/main/root

Notes :

  • If you have disabled lvm
  1. you don't need the pvresize or lvresize command
  2. the final line becomes resize2fs /dev/md2 (or whatever / is mounted to)
  3. More info: http://www.arkf.net/blog/?p=47