STONITH

STONITH, an acronym used in high-availability cluster management, stands for a mechanism that forcibly isolates a malfunctioning or unreachable node to protect data integrity. The term is often explained with a tongue-in-cheek expansion, but its practical purpose is to prevent a split-brain scenario where two or more nodes believe they are the active primary and access the same shared resources.

The primary goal of STONITH is to ensure that a failed or partitioned node cannot continue accessing any shared storage or resources. When a cluster detects a node is unresponsive or misbehaving, it uses a fence device or fence agent to "shoot" that node from the cluster, typically by power-cycling or severing network access. A successful fence action should decisively render the node harmless to the shared resources. Modern cluster stacks, such as Pacemaker with Corosync, manage STONITH devices via fence agents and support a range of hardware and software fencing methods.

Common fencing methods include hardware-based power fencing (IPMI, iDRAC, ILO, and smart PDUs) and software or network-based fencing techniques. Fence agents, such as fence_ipmilan, fence_ipmi, or vendor-specific controllers, implement the communication with the fence device. STONITH configurations should be highly reliable and auditable, with authentication and logging, and are considered essential in many clusters to guarantee data integrity.

In some environments, STONITH can be disabled only under tightly controlled conditions, but doing so increases the risk of data corruption in case of failover. Proper fencing is a foundational component of robust high-availability architectures.