We have two RAIDs on the HPC
If you're looking at the front of the HPC you'll see four rows of drives. From the bottom:
The Linux kernel has the md (mirrored devices) driver for software RAID devices. There are currently two 80 GB IDE hard drives connected to the server, /dev/hda and /dev/hdd. These were set up as five RAID devices during the install of Rocks/CentOS.
Here is information on their configuration:
# mount | grep md /dev/md0 on / type ext3 (rw) /dev/md3 on /boot type ext3 (rw) /dev/md2 on /scratch type ext3 (rw) /dev/md1 on /export type ext3 (rw) # df -h | grep md /dev/md0 29G 11G 17G 39% / /dev/md3 190M 60M 121M 34% /boot /dev/md2 35G 177M 33G 1% /scratch /dev/md1 25G 5.5G 18G 24% /export
It should be noted that /dev/md4 is being used as swap:
# swapon -s Filename Type Size Used Priority /dev/md4 partition 2168632 0 -1
A snapshot of the software RAID's health:
# cat /proc/mdstat
Personalities : [raid1] [raid0]
md3 : active raid1 hdd1[1] hda1[0]
200704 blocks [2/2] [UU]
md1 : active raid1 hdd3[1] hda3[0]
26627648 blocks [2/2] [UU]
md2 : active raid0 hdd5[1] hda5[0]
36868608 blocks 256k chunks
md4 : active raid1 hdd6[1] hda6[0]
2168640 blocks [2/2] [UU]
md0 : active raid1 hdd2[1] hda2[0]
30716160 blocks [2/2] [UU]
unused devices: <none>
When a disk is failing you need to replace the drive. First, look at the RAID configuration to see which partitions are in use by which arrays. For example:
# cat /proc/mdstat
Personalities : [raid1] [raid0]
md3 : active raid1 hdd1[1] hda1[0]
200704 blocks [2/2] [UU]
md1 : active raid1 hdd3[1] hda3[0]
26627648 blocks [2/2] [UU]
md2 : active raid0 hdd5[1] hda5[0]
36868608 blocks 256k chunks
md4 : active raid1 hdd6[1] hda6[0]
2168640 blocks [2/2] [UU]
md0 : active raid1 hdd2[1] hda2[0]
30716160 blocks [2/2] [UU]
unused devices: <none>
If /dev/hda is having problems, set all its RAID1 partitions as failed and remove them:
# mdadm /dev/md0 --fail /dev/hda2 --remove /dev/hda2 # mdadm /dev/md1 --fail /dev/hda3 --remove /dev/hda3 # mdadm /dev/md3 --fail /dev/hda1 --remove /dev/hda1 # mdadm /dev/md4 --fail /dev/hda6 --remove /dev/hda6
/dev/md2 is a RAID0 stripe mounted as /scratch, so we have to umount it and then stop it (you can't remove volumes from a stripe):
# umount /dev/md2 # mdadm --stop /dev/md2
Shut the server down and replace the faulty drive with a new one. After booting your drive letters may have shifted around, so just be sure to verify which is which before proceeding. Clone the partition table from the good drive to the bad one:
# sfdisk -d /dev/hda | sfdisk --force /dev/hdd
Verify the new partitions can be seen:
# partprobe -s /dev/hda: msdos partitions 1 2 3 4 <5 6> /dev/hdd: msdos partitions 1 2 3 4 <5 6> /dev/sda: msdos partitions 1 /dev/sdb: msdos partitions 1 /dev/sdc: msdos partitions 1
Re-create the scratch partition (RAID0):
# mdadm --create --verbose /dev/md2 --level=0 --raid-devices=2 /dev/hda5 /dev/hdd5 # mkfs.ext3 /dev/md2 # mount /dev/md2 /scratch
You can now add the new partitions back to the RAID1 arrays:
# mdadm /dev/md0 --add /dev/hdd2 # mdadm /dev/md1 --add /dev/hdd3 # mdadm /dev/md3 --add /dev/hdd1 # mdadm /dev/md4 --add /dev/hdd6
After adding you can monitor the progress of the RAID rebuilds by looking in /proc/mdstat:
Personalities : [raid1] [raid0]
md3 : active raid1 hdd1[1] hda1[0]
200704 blocks [2/2] [UU]
md1 : active raid1 hdd3[2] hda3[0]
26627648 blocks [2/1] [U_]
[===================>.] recovery = 95.4% (25407552/26627648) finish=0.7min speed=28648K/sec
md2 : inactive hda5[0]
18434304 blocks
md4 : active raid1 hdd6[2] hda6[0]
2168640 blocks [2/1] [U_]
resync=DELAYED
md0 : active raid1 hdd2[1] hda2[0]
30716160 blocks [2/2] [UU]
unused devices: <none>
A 3ware 9500S-12 SATA RAID card using the 3w-9xxx kernel module. It has 12 channels. The HPC is configured to use RAID5 for all of its RAID arrays on the hardware RAID.
We have one RAID controller, 'c1'. Disks are plugged into ports, 'p0' - 'p11'. The disks are then grouped into units (basically the rows), 'u0' - 'u2'.
| Port 8 | Port 9 | Port 10 | Port 11 |
| Port 4 | Port 5 | Port 6 | Port 7 |
| Port 0 | Port 1 | Port 2 | Port 3 |
There is a utility, tw_cli, which can be used to control/monitor the hardware raid controller.
Study the output of show to know which controller to manage. Then you can use /c1 show to show the status of that particular controller. Things to look for:
tw_cli tool before you can swap drives.
Remove the faulty port:
maint remove c1 p5
Insert a new drive and rescan:
maint rescan
Rebuild the degraded array:
maint rebuild c1 u2 p5
Check the status of the rebuild by monitoring /c1 show, but I have a feeling this might disturb the rebuild process. In any case, you can check the status by following the output of dmesg:
3w-9xxx: scsi1: AEN: INFO (0x04:0x000B): Rebuild started:unit=2. 3w-9xxx: scsi1: AEN: INFO (0x04:0x0005): Background rebuild done:unit=2.
This sucks:
3w-9xxx: scsi1: AEN: INFO (0x04:0x0029): Background verify started:unit=0. 3w-9xxx: scsi1: AEN: INFO (0x04:0x002B): Background verify done:unit=0. 3w-9xxx: scsi1: AEN: ERROR (0x04:0x0002): Degraded unit detected:unit=0, port=3
$ sudo tw_cli Password: //hpc-ilri> /c1 show Unit UnitType Status %RCmpl %V/I/M Stripe Size(GB) Cache AVrfy ------------------------------------------------------------------------------ u0 RAID-5 DEGRADED - - 64K 698.461 ON OFF u1 RAID-5 OK - - 64K 698.461 ON OFF u2 RAID-5 OK - - 64K 698.461 ON OFF Port Status Unit Size Blocks Serial --------------------------------------------------------------- p0 OK u0 232.88 GB 488397168 WD-WMAEP2714804 p1 OK u0 232.88 GB 488397168 WD-WMAEP1570106 p2 OK u0 232.88 GB 488397168 WD-WMAEP2712887 p3 DEGRADED u0 232.88 GB 488397168 WD-WMAEP2714418 p4 OK u2 232.88 GB 488397168 WD-WCAT1C715001 p5 OK u2 232.88 GB 488397168 WD-WMAEP2713449 p6 OK u2 232.88 GB 488397168 WD-WMAEP2715070 p7 OK u2 232.88 GB 488397168 WD-WMAEP2712590 p8 OK u1 232.88 GB 488397168 WD-WMAEP2712574 p9 OK u1 232.88 GB 488397168 WD-WMAEP2734142 p10 OK u1 232.88 GB 488397168 WD-WMAEP2702155 p11 OK u1 232.88 GB 488397168 WD-WMAEP2712472
Looks like another drive failed.