[REEF-316] Improve reporting when containers are killed by the RM

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Component/s: REEF-Runtime-YARN
• Assignee: Unassigned
• Reporter: Markus Weimer
• Created: 10/May/15 22:50
• Updated: 10/May/15 22:50

Description

YARN sometimes kills containers. Most commonly, when those containers outgrow their resource limit. REEF today doesn't provide any indication about the reason for the lost Evaluator in the FailedEvaluator message. We should investigate whether we can do a better job here, e.g. if YARN provides us with a reason for the kill we can relay.