What is quorum in clustered databases?
In the world of clustered databases like Oracle RAC, Quorum is the "legal majority" required for the cluster to keep running.
Think of a cluster as a small committee. If the committee members (servers) lose the ability to talk to each other, they need a set of rules to decide who is still in charge and who should be "fired." Quorum is that set of rules.
The primary reason for Quorum is to prevent Split-Brain Syndrome.
Imagine a 2-node cluster where the network cable between Node A and Node B is cut.
Node A thinks: "I can't see Node B. It must have crashed. I'll take over everything!"
Node B thinks: "I can't see Node A. It must have crashed. I'll take over everything!"
If both nodes start writing to the same data files independently, they will corrupt the database beyond repair in seconds. Quorum prevents this.
Every node in a cluster is given a vote. To remain "alive," a sub-group of nodes must hold a majority of the total possible votes (usually $50\% + 1$).
In an even-numbered cluster (like 2 nodes), you have a problem. If they lose contact, they each have 50% of the votes—a tie.
To break the tie, Oracle uses a Voting Disk (stored on shared storage).
The Voting Disk acts like a "Golden Vote."
The node that successfully "locks" the Voting Disk first gets the extra vote and wins.
The losing node is instantly "evicted" (rebooted) to protect the data.
Quorum isn't just for servers; it’s also for storage. In Oracle ASM, Quorum determines if a Disk Group can stay "mounted" based on how many disks are healthy.
Normal Redundancy (2-way): Requires at least 2 "Failure Groups."
High Redundancy (3-way): Requires at least 3 "Failure Groups."
The Quorum Device: In a "Flex" or "Extended" cluster (spread across two buildings), you might put a third "Quorum Disk" in a third building. This disk doesn't store your data; it only stores a "vote" to decide which building stays online if the fiber optic cable between them is cut.
Most architects prefer an odd number of nodes (3, 5, etc.) or an odd number of voting devices.
In a 3-node cluster, if 1 node dies, 2 are left ($66\%$). The majority is clear.
In a 2-node cluster, you are 100% dependent on that shared Voting Disk to act as the "third man."
When a node loses Quorum, Oracle Clusterware performs a Fast Reboot (fencing).
I/O Fencing: The node is blocked from writing to the disks.
Panic: The node's OS is forced to reboot immediately.
Recovery: The surviving node(s) perform "Instance Recovery" to clean up any half-finished work the dead node left behind.
| Component | Role |
| Nodes | The servers that hold active votes. |
| Voting Disks | The "referees" stored on shared storage. |
| Interconnect | The network used to count the votes. |
| Majority | The $n/2 + 1$ rule that keeps the cluster up. |