[TEZ-2250] Consolidate the Internal Error transition for DAG/Verex/Task/TaskAttempt

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Unassigned
• Reporter: Jeff Zhang
• Created: 30/Mar/15 07:19
• Updated: 20/May/15 05:33

Description

Currently, invalid state machine transition will trigger the internal error. Here's current logics

• Invalid transition in Task/TaskAttempt will send DAGEvent with type of INTERNAL_ERORR to dag, and dag will trigger its InternalErrorTransition
• Invalid transition in Vertex will finish this vertex without sending INTERNAL_ERORR event to dag, although this will also cause the dag to move to ERROR
• Invalid transition in DAG will cause it to kill all the vertices.

2 issues:

• Invalid transition in DAG/Vertex/Task/TaskAttempt aren't consistent, I think any invalid transition should send INTERNAL_ERROR event to and then trigger it to kill vertices/task/task attempts
• Currently DAG/Vertex move to ERROR without wait ing for vertices/task to complete. It would cause the aborting conflict with the running task attempt.

Issue Links

• relates to: TEZ-2456 Refactor recovery event logging to ensure it meet the recovery event spec