[REEF-541] Notifying evaluator restart failure entails calling AllocatedEvaluatorHandler

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.12
• Component/s: REEF Driver, REEF.NET Driver
• Assignee: Andrew Chung
• Reporter: Andrew Chung
• Created: 30/Jul/15 23:58
• Updated: 05/Aug/15 18:48
• Resolved: 05/Aug/15 18:48

Description

On restarts, if an evaluator fails, we need to create a brand new EvaluatorManager for it, which calls AllocatedEvaluatorHandler. This does not really make sense as the evaluator was never really "allocated," and we should call EvaluatorFailedHandler directly. In addition, in the C# interop portion for AllocatedEvaluatorHandler, we inspect fields in the Evaluator object, which causes a NullReferenceException because on restarts, we do not initialize certain fields for the "allocated" evaluator. Perhaps there should be a new EvaluatorFailed handler specifically for restarts as well.

Issue Links

• blocks: REEF-560 Add a configurable timeout for driver to recover evaluators on restart
• links to: Link to PR