Manually switching (migrating) the service group to a secondary node is the standard, controlled method for planned maintenance. It ensures service remains online for users on the secondary system while the primary node is taken offline for upgrades. The other options cause unnecessary downtime (offlining globally), prevent failover during maintenance which is risky (freeze), or leave the service completely down (NeverFailover).

When a service group faults, VCS engine logs and the detailed resource state (via ‘hares -display’) contain the specific reasons for failure, such as a probe script returning a failure, a resource dependency not being met, or an agent error. This is the targeted diagnostic step. Checking system messages is too broad, moving the group doesn’t diagnose the root cause, and rebooting is disruptive and doesn’t address the underlying issue.

The behavior upon primary node recovery is governed by the AutoFailback attribute of the service group. If set to 0 (default), the group stays on its current node (NodeB). If set to 1, VCS will automatically migrate the group back to its highest-priority, available node (NodeA). Immediate failback is not default behavior, and a properly configured cluster should not experience split-brain in this routine scenario.

Best practice for the private heartbeat network is redundancy and performance. Configuring at least two dedicated network interfaces (e.g., cross- over cables or via a switch) in an active-active configuration allows LLT to load balance heartbeat traffic and provides immediate failover if one link fails. A single link is a single point of failure. Different subnets would complicate LLT. Using the public network is strongly discouraged due to latency and potential traffic interference.

The command ‘gabconfig -px’ (where x is a port number) is used to disable a specific GAB port. This simulates the failure of a network link used for cluster communication, effectively isolating the node on that port and allowing you to test split-brain detection and fencing mechanisms. ‘-c’ configures a port, ‘-n’ changes node membership, and ‘-D’ provides debug output, none of which directly simulate link failure.

Each Veritas agent has its own log file (e.g., Nic.log, Mount.log, Volume.log) in the /opt/VRTSvcs/log directory. When a specific resource like a NIC shows an UNKNOWN or faulted state, its dedicated agent log will contain the detailed execution output, error messages from its monitor (probe) or online/offline methods, and is the primary source for debugging that resource’s behavior. The engine log is for higher-level cluster coordination.

In VCS, storage resources must be modeled with clear dependencies. The DiskGroup (CVM disk group) is a foundational resource. Volume resources (CVM volumes) depend on the DiskGroup being online. Therefore, you create the DiskGroup resource first, then create Volume resources specifying their dependency on the DiskGroup, and finally add these resources to the service group in the correct hierarchical order. VCS does not auto-discover or auto-create resources from CVM operations.

This scenario calls for a custom monitoring script (probe) for the Process resource. The probe would check memory usage. The FailureInterval and RetryLimit (or similar) attributes on the resource control how many probe failures (e.g., memory threshold breaches) must occur within a specified time window before VCS declares the resource faulted, which would then trigger the service group’s fault response, such as a failover or restart.

A core requirement for shared storage in a cluster is that all nodes must have concurrent physical access to the same disk LUNs. This is achieved through proper SAN zoning and LUN masking, presenting the LUNs to all cluster nodes. The storage can be from multiple arrays, device numbers can differ (VxVM uses Disk Access Records or DIDs to handle this), and while VxVM is required, CVM is a specific clustered configuration of VxVM.

Manually switching (migrating) the service group to a secondary node is the standard, controlled method for planned maintenance. It ensures service remains online for users on the secondary system while the primary node is taken offline for upgrades. The other options cause unnecessary downtime (offlining globally), prevent failover during maintenance which is risky (freeze), or leave the service completely down (NeverFailover).0