Drive Replacement Infrastructure SOP

From FedoraProject

(Difference between revisions)
Jump to: navigation, search
m (SOP Formatting)
(redirect page to new infra-docs)
 
(4 intermediate revisions by 2 users not shown)
Line 2: Line 2:
 
{{shortcut|ISOP:DRIVE}}
 
{{shortcut|ISOP:DRIVE}}
  
{{admon/note|At present this SOP only works for the X series IBM servers.}}
 
  
We have multiple machines with lots of different drives in them. For the most part now though, we are trying to standardise on IBM X series servers. At present I've not figured out how to disable onboard raid, as a result of this many of our servers have two raid 0 arrays then we do software raid on this.
+
This SOP has moved to the fedora Infrastructure SOP git repo. Please see the current document at: http://infrastructure.fedoraproject.org/infra/docs/ibm-drive-replacement.txt
  
= Database - DriveReplacement =
+
For changes, questions or comments, please contact anyone in the Fedora Infrastructure team.
  
== Contact Information ==
 
Owner: Fedora Infrastructure Team
 
 
Contact: #fedora-admin, sysadmin-main
 
 
Location: All
 
 
Servers: All
 
 
Purpose: Steps for drive replacement.
 
 
== Verify the drive is dead ==
 
 
<pre>
 
$ cat /proc/mdadm
 
Personalities : [raid1]
 
md0 : active raid1 sdb1[1] sda1[0]
 
      513984 blocks [2/2] [UU]
 
 
md1 : active raid1 sdb2[2](F) sda2[0]
 
      487717248 blocks [2/1] [U_]
 
</pre>
 
 
This indicates that md1 is in a degraded state and that /dev/sdb2 is the failed drive.  Notice that /dev/sdb1 (same physical drive as /dev/sdb2) is not failed.  /dev/md0 (not yet degraded) is showing a good state.  This is because /dev/md0 is /boot.  If you run:
 
 
<pre>
 
touch /boot/t
 
sync
 
rm /boot/t
 
</pre>
 
 
That should make /dev/md0 notice that its drive is also failed.  If it does not fail, its possible the drive is fine and that some blip happened that caused it to get flagged as dead.  It is also worthwhile to log in to xenX-mgmt to determine if the RSAII adapter has noticed the drive is dead.
 
 
If you think the drive just had a blip and is fine, see "Re-adding" below
 
 
=== Re-adding a drive (poor man's fix) ===
 
 
Basically what we're doing here is making sure the drive is, infact, dead.  Obviously you don't want to do this more then once on a drive, if it continues to fail.  Replace it.
 
 
<pre>
 
# cat /proc/mdadm
 
Personalities : [raid1]
 
md0 : active raid1 sdb1[1] sda1[0]
 
      513984 blocks [2/2] [UU]
 
 
md1 : active raid1 sdb2[2](F) sda2[0]
 
      487717248 blocks [2/1] [U_]
 
# mdadm /dev/md1 --remove /dev/sdb2
 
# mdadm /dev/md1 --add /dev/sdb2
 
# cat /proc/mdstat
 
md0 : active raid1 sdb1[1] sda1[0]
 
      513984 blocks [2/1] [U_]
 
        resync=DELAYED
 
 
md1 : active raid1 sdb2[2] sda2[0]
 
      487717248 blocks [2/1] [U_]
 
      [=>...................]  recovery =  9.2% (45229120/487717248) finish=145.2min speed=50771K/sec
 
</pre>
 
 
So we removed the bad drive, added it again and you can now see the recovery status.  Watch it carefully.  If it fails again, time for a drive replacement.
 
 
== Actually replacing the drive ==
 
 
Actually replacing the drive is a bit of a todo.  If the box is in a RH owned location, we'll have to file a ticket and get someone access to the colo.  If it is at another location, we may be able to just ship the drive there and have someone do it on site.  Please follow the below steps for drive replacement.
 
 
=== Data ===
 
 
There's a not insignifcant amount of data you'll need to place the call.  Please have the following information handy:
 
 
1) The hosts machine type (this is not model number).
 
<pre>
 
# lshal | grep system.product
 
  system.product = 'IBM System x3550 -[7978AC1]-'  (string)
 
</pre>
 
 
In the above case, the machine type is encoded into [7978AC1].  And is just the first 4 numbers.  So this machinetype is 7978.  M/T (machine type) is always 4 digits for IBM boxes.
 
 
2) Machine's serial number
 
<pre>
 
# lshal | grep system.hardware.serial
 
  system.hardware.serial = 'FAAKKEE'  (string)
 
</pre>
 
 
The above's serial number is 'FAAKKEE'
 
 
3) Drive Stats
 
 
There are two ways to get the drive stats.  You can get some of this information via hal, but for the full complete information you need to either have someone physically go look at the drive (some of which is in inventory) or use RaidMan.  See "Installing RaidMan" below for more information on how to install RaidMan.
 
 
Specifically you need:
 
 
Drive Size (in G)
 
Drive Type (SAS or SATA?)
 
Drive Model
 
Drive Vendor
 
 
To get this information run:
 
 
<pre>
 
# cd /usr/RaidMan/
 
# ./arcconf GETCONFIG 1
 
</pre>
 
 
4) The phone number and address of the building where the drive is currently located.  This will go to the RH cage.
 
 
This information is located in the contacts.txt of private git repo on puppet1 (only available to sysadmin-main people)
 
 
=== Call IBM ===
 
 
Call 1-770-858-5079 and follow the directions they give you.  You'll need to use the M/T above to get to the correct rep.  They will ask you for the information above (you wrote it down, right?)
 
 
When they agree to replace the drive, make sure to tell them you need the shipping number of the drive as well as the name of the tech who will do the drive replacement.  Sometimes the tech will just bring the drive.  If not though, you need to open a ticket with the colo to let them know a drive is coming.
 
 
=== Get the package, give access to the tech ===
 
 
As SOON as you get this information, open a ticket with RH. at is-ops-tickets redhat.com.  Request a ticket ID from RH.  If the tech has any issues getting into the colo, you can give the AT&T ticket request to the tech to get them in.
 
 
NOTE: this can often take hours. We have 4 hour on site response time from IBM.  This time goes very quickly, sometimes you may need to page out someone in IS to ensure it gets created quickly.  To get this pager information see contacts.txt in puppet1's private repo (if puppet1 is down for some reason see the dr copy on backup2.fedoraproject.org:/srv/
 
 
=== Prepwork before the tech arrives ===
 
 
Really the big thing here is to remove the broken drive from the array.  In our earlier example we found /dev/sdb failed.  We'll want to remove it from both arrays:
 
 
<pre>
 
# mdadm /dev/md0 --remove /dev/sdb1
 
# mdadm /dev/md1 --remove /dev/sdb2
 
</pre>
 
 
Next get the current state of the drives and save it somewhere.  See "Installing RaidMan" for more information if RaidMan is not installed.
 
 
<pre>
 
# cd /usr/RaidMan
 
# ./arcconf GETCONFIG 1 > /tmp/raid1.txt
 
</pre>
 
 
Copy /tmp/raid1.txt off to some other device and save it until the tech is on site.  It should contain information about the failed drive.
 
 
=== Tech on site ===
 
 
When the tech is on site you may have to give him the rack location.  All of our Mesa servers are in one location, "the same room that the desk is in".  You may have to give him the serial number of the server, or possibly make it blink.  It's either the first rack on the left labeled: "01 2 55" or "01 2 58".
 
 
Once he's replaced the drive, he'll have you verify.  Use the RaidMan tools to do the following:
 
 
<pre>
 
# cd /usr/RaidMan
 
# ./arcconf RESCAN 1
 
# ./arcconf GETCONFIG 1 > /tmp/raid2.txt
 
# # arcconf CREATE <Controller#> LOGICALDRIVE [Options] <Size> <RAID#> <Channel# ID#>
 
# ./arcconf create 1 LOGICALDRIVE 476790 Simple_volume 0 1
 
</pre>
 
 
First we're going to re-scan the array for the new drive.  Then we'll re-get the configs.  Compare /tmp/raid2.txt to /tmp/raid1.txt and verify the bad drive is fixed and that it has a different serial number.  Also make sure its the correct size.  Thank the tech and send him on his way.  The last line there creates a new logical drive from the physical drive.  "Simple_volume" tells it to create a raid0 array of one drive.  The size was pulled out of our initial /tmp/raid1.txt (should match the other drive).  The last two numbers are the Channel and ID of the new drive.
 
 
=== Rebuild the array ===
 
 
Now that the disk has been replaced we need to put a partition table on the new drive and add it to the array:
 
 
* /dev/sdGOOD is the *GOOD* drive
 
* /dev/sdBAD is the *BAD* drive
 
 
<pre>
 
# dd if=/dev/sdGOOD of=/tmp/sda-mbr.bin bs=512 count=1
 
# dd if=/tmp/sda-mbr.bin of=/dev/sdBAD
 
# partprobe
 
</pre>
 
 
Next re-add the drives to the array:
 
 
* /dev/sdBAD1 and /dev/sdBAD2 are the partitons on the new drive which is no longer bad.
 
<pre>
 
# mdadm /dev/md0 --add /dev/sdBAD1
 
# mdadm /dev/md1 --add /dev/sdBAD2
 
# cat /proc/mdadm
 
</pre>
 
 
This starts rebuilding the arrays, the last line checks the status.
 
 
== Installing RaidMan ==
 
 
Unfortunately there is no feasible alternative to managing IBM Raid Arrays without causing downtime.  You can get and do this via the pre-POST interface.  This requires downtime, and if the first drive is the failed drive, may result in a non-booting system.  So for now RaidMan it is until we can figure out how to get rid of the raid controllers in these boxes completely.
 
 
<pre>
 
yum -y install compat-libstdc++-33
 
rpm -ihv http://infrastructure.fedoraproject.org/rhel/RaidMan/RaidMan-9.00.i386.rpm
 
</pre>
 
 
To verify installation has completed successfully:
 
 
<pre>
 
# cd /usr/RaidMan/
 
# ./arcconf GETCONFIG 1
 
</pre>
 
 
This should print the current configuration of the raid controller and its logical drives.
 
  
 
[[Category:Infrastructure SOPs]]
 
[[Category:Infrastructure SOPs]]

Latest revision as of 19:28, 19 December 2011

Infrastructure InfrastructureTeamN1.png
Shortcut:
ISOP:DRIVE


This SOP has moved to the fedora Infrastructure SOP git repo. Please see the current document at: http://infrastructure.fedoraproject.org/infra/docs/ibm-drive-replacement.txt

For changes, questions or comments, please contact anyone in the Fedora Infrastructure team.