How do you protect BCM against a head node failure?
Why D is the answer
Configure head node high availability
A second head node takes over.
In detail
- A second head node takes over.
- Test failover before relying on it.
Use case
An illustrative scenario, not a real incident.
A head node fails, and a second one takes over.
See it in different ways
flowchart LR A["Head node"] --> B["Software image"] B --> C["Category"] C --> D["Compute nodes"] A --> E["Standby head node"]
mindmap
root((BCM))
Head node
High availability
cmsh
Category
Shared settings
Software image
Node OS
Workloads
Slurm
Enroot and PyxisstateDiagram-v2 state "Primary head node" as s0 state "Failed" as s1 state "Secondary takes over" as s2 [*] --> s0 s0 --> s1: failure s1 --> s2 s2 --> [*]
Key point: Configure head node HA.
Why the other options are wrong
- A. Back up the configuration once a month
- Backups help recovery, not failover.
- B. Install BCM on each compute node
- Compute nodes do not run BCM that way.
- C. Run a nightly HPL burn-in on each head node and its peer
- That tests compute nodes.