RAC - Voting Disk

Voting Disk –

Voting disk is a shared file used by clusterware for various functions. Primary purpose of voting disk is node membership. It contains information about all the nodes in the cluster. Clusterware uses voting disk to decide whether a node is part of the cluster. Every node writes 2 types of heartbeats in voting disk. 1. Disk heartbeat 2. Network heartbeat. If the node is able to communicate with node master it will put disk heartbeat. If the node is able to communicate with other nodes of the cluster, it will put network heartbeat. If CSSD is able to see both the heartbeat of a node in voting disk, it means the node is healthy. If any of the heartbeat is missing, there is some error. If disk heartbeat is missing, cssd updates kill block of node. Kill block of node is checked every second. Till 11gr1, voting disk was placed on raw device. From 11gr2, its mandatory to put voting disk on ASM storage.  It creates one more issue. If a node is coming up, it needs to read voting disk, but asm cant be up until node is up. So how will it read the voting disk. There is a solution to this issue. Once asm is up, cluster puts voting disk information at a fixed location and the location is put in asm disk header. So even if asm is not up, cluster can read voting disk. Till 11gr1, voting disk had to be backed up separately. From 11gr2, voting disk as it resides in asm, is backed up automatically with disk backup. We should always keep odd number of voting disk. There is a reason for it. At any point of time any node in cluster should be able to access more than half number of total voting disks. [Number of voting disk - failure tolerated, 1 –> 0, 2   –> 0, 3 –> 1, 4 –> 1, 5 –> 2, 6 –> 2, 2n-1 –> n, 2n -> n] As for 2n-1 and 2n number of voting disks, failure is n, then it’s better to keep 2n-1 voting disk instead of keeping one redundant voting disk. So, we always keep odd number of voting disks. We can keep maximum of 15 voting disk, minimum of 3. Oracle recommends not more than 5. The size of voting disk is usually 280mb. The number of voting disk will be decided on the asm disk redundancy we are using.  We can query voting disk by (crsctl query css votedisk)

[oracle@testing ~]$ crsctl query css votedisk

##  STATE    File Universal Id                File Name Disk group

--  -----    -----------------                --------- ---------

 1. ONLINE   53e5e46930f14f0ebfdc49abfed65701 (/dev/raw/raw11) [OCRVD]

 2. ONLINE   2dd1ebe0dc1f4feabf3545c3f589d407 (/dev/raw/raw12) [OCRVD]

 3. ONLINE   74814e9a48d14fc1bfc48ec5afc75b90 (/dev/raw/raw13) [OCRVD]

Located 3 voting disk(s).