Applications fail between hosts, but ibstat shows Active. Which test is next?
Why D is the answer
ibping or a perftest
Test the path, since local state is fine.
In detail
- Local state is fine, so test the path.
- Local state is fine, so test the path.
Use case
An illustrative scenario, not a real incident.
Ports are Active, but an application fails, so ibping tests the path.
See it in different ways
flowchart TD
A{"What do you want to know?"} -->|"Local port state"| B["ibstat"]
A -->|"Which nodes exist"| C["ibnodes"]
A -->|"Link details"| D["iblinkinfo"]
A -->|"Fabric errors"| E["ibdiagnet"]
A -->|"RDMA speed"| F["ib_write_bw and ib_write_lat"]stateDiagram-v2 [*] --> LocalPort: ibstat LocalPort --> Fabric: ibnodes and iblinkinfo Fabric --> Errors: ibdiagnet Errors --> Path: ibping Path --> Speed: ib_write_bw and ib_write_lat Speed --> [*]
flowchart TD
Q{"Does a path test work?"}
Q -->|"Yes"| A0["Check the application"]
Q -->|"No"| A1["Check the fabric path"]Key point: Test the path next.
Why the other options are wrong
- A. A rack power audit
- Power is not the symptom.
- B. A GPU driver reinstall
- A GPU driver is not on the path.
- C. A BCM head node rebuild
- BCM is not on the path.