Saturday, May 7, 2011

RHCS(Red Hat Cluster Suite) quorum disk

The last post "RHCS I/O fencing" is about dealing with split-brain situation, in which cluster members lost heartbeat communication and each believe it is legitimate to write data to the shared storage.
Methods to deal with split-brain situation:
1. Redundant heartbeat path
network port communication plus serial port communication
2. I/O fencing
Remaining nodes separate failed node from its storage either by shutdown/reboot power port or storage port
3. Quorum disk
Quorum disk is a kind of I/O fencing, but the reboot action is executed by failed node's own quorum daemon. It also has additional feature: contributing vote to cluster. if you want the last standing node to keep the multiple-nodes cluster  running, quorum disk appears to be the only solution.
RHCS (Red Hat Cluster Suite) Quorum disk facts
- A shared block device (SCSI/iSCSI/FC..), Device size requirement is approximately 10MiB
- Supports maximum 16 nodes, nodes id must be sequentially ordered
- Quorum disk can contribute  votes. In multiple nodes cluster, together with quorum vote, the last standing node can still keep the cluster running
- single node votes+1 <=Quorum's disk vote < nodes total votes
- The failure of the shared quorum disk won’t result in cluster failure, as long as Quorum's disk vote < nodes total votes
- each node write its own health information in its own region, the health is determined by external checking program such as "ping"
Setup Quorum disk
#initialise quorum disk once in any node 
mkqdisk -c /dev/sdx -l myqdisk 
Add quorum disk to cluster 
Use luci or system-config-cluster to add quorum disk, following is the result xml file
<clusternodes>
<clusternode name="station1.example.com" nodeid="1" votes="2">
<fence/>
</clusternode>
<clusternode name="station2.example.com" nodeid="2" votes="2">
<fence/>
</clusternode>
<clusternode name="station3.example.com" nodeid="3" votes="2">
<fence/>
</clusternode>
</clusternodes>
#expected votes =9=(nodes total votes + quorum disk votes) = (2+2+2+3)       
<cman expected_votes="9"/> 
#Health check result is writen to quorum disk every 2 secs
#if health check fails over 5 tko, 10 (2*5) secs, the node is rebooted by quorum daemon
#Each heuristic check is run very 2 secs and earn 1 score,if shell script return is 0
<quorumd interval="2" label="myqdisk" min_score="2" tko="5" votes="3">
<heuristic interval="2" program="ping -c1 -t1 192.168.1.60" score="1"/>
<heuristic interval="2" program="ping -c1 -t1 192.168.1.254" score="1"/>
</quorumd>
Start quorum disk daemon
The daemon is also one of daemons automatically started by cman
service qdiskd start
Check quorum disk information
$ mkqdisk -L -d
mkqdisk v0.6.0
/dev/disk/by-id/scsi-1IET_00010002:
/dev/disk/by-uuid/55fbf858-df75-493b-a764-5640be5a9b46:
/dev/sdc:
Magic:                eb7a62c2
Label:                myqdisk
Created:              Sat May  7 05:56:35 2011
Host:                 station2.example.com
Kernel Sector Size:   512
Recorded Sector Size: 512
Status block for node 1
Last updated by node 1
Last updated on Sat May  7 15:09:37 2011
State: Master
Flags: 0000
Score: 0/0
Average Cycle speed: 0.001500 seconds
Last Cycle speed: 0.000000 seconds
Incarnation: 4dc4d1764dc4d176
Status block for node 2
Last updated by node 2
Last updated on Sun May  8 01:09:38 2011
State: Running
Flags: 0000
Score: 0/0
Average Cycle speed: 0.001000 seconds
Last Cycle speed: 0.000000 seconds
Incarnation: 4dc55e164dc55e16
Status block for node 3
Last updated by node 3
Last updated on Sat May  7 15:09:38 2011
State: Running
Flags: 0000
Score: 0/0
Average Cycle speed: 0.001500 seconds
Last Cycle speed: 0.000000 seconds
Incarnation: 4dc4d2f04dc4d2f0
The cluster is still running with last node standing
Please note Total votes=quorum votes=5=2+3, if quorum disk vote is less than  (node votes+1), the cluster  wouldn’t have  survived
$cman_tool status
..
Nodes: 1
Expected votes: 9
Quorum device votes: 3
Total votes: 5
Quorum: 5  
..

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.