[REEF-1310] The Java Driver should ACK the Java Evaluator's DONE heartbeat

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.15
• Component/s: REEF, REEF Driver, REEF-Common
• Assignee: Andrew Chung
• Reporter: Andrew Chung
• Created: 04/Apr/16 16:55
• Updated: 14/Apr/16 21:28
• Resolved: 14/Apr/16 21:28

Description

The Driver should ACK the Evaluator's DONE heartbeat such that a race condition does not occur when the Evaluator ends. i.e. The Evaluator heartbeats DONE back to the Driver and the RM notices that the Evaluator process has exited. In this case, it is possible that the RM reports back to the Driver that the Evaluator is DONE before the Evaluator's DONE heartbeat goes back to the Driver, causing the Driver to invoke the FailedEvaluatorHandler due to an unexpected DONE message from the RM.

Issue Links

• relates to: REEF-976 Fix broken C# Tests caused by race condition of local RM
• relates to: REEF-1291 Driver gets FailedEvaluator message even if evaluator shuts down properly