Description of problem:
Today, our fencing logic tells one host to fence another over a fencing network and without any validation logic to determine if there are real host issues, or simply network connectivity issues.
How reproducible:
Seen multiple use cases at several customers
Steps to Reproduce:
1. Engine loses direct network connectivity and cannot see any host
2. Uplink switch loses network connectivity, breaking route between engine and host
3. Host loses direct network connectivity causing engine to see it as down
Actual results:
We need to add additional logic to make an attempt to validate a real issue and minimize the number of false positives introduced by today's fencing logic.
Expected results:
Improvement to fencing logic, knowing there is no way to have 100% accuracy
This BZ will be used to centralize the core fencing improvements and will have dependencies on Infra, Storage and Network components for the immediate solution.
Targeting tracker for RHEV 3.6. Individual fencing logic checks will be targeted for 3.5 or 3.6 depending on dev capacity, and features will be introduced over 2 releases.