[HBASE-16842] Chaos policies can terminate all masters for extended periods of time

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: integration tests
• Assignee: Unassigned
• Reporter: Andrew Kyle Purtell
• Created: 14/Oct/16 16:44
• Updated: 14/Oct/16 21:53

Description

Running ITBLL with the slowDeterministic monkey I observe our primary and backup masters in the test cluster can be both shut down by signals followed by no attempt to start replacements for an extended period of time. Meanwhile other actions continue to run that churn the regionserver fleet. Other monkeys may enter a similar state, but I haven't observed it. The outcome of the convergence of these behaviors is the eventual time out and termination of the running integration test, which is obvious and expected and unhelpful, so I believe unintentional.