[TEZ-3366] Tez timeline client reporting different domains for same entity

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate

Environment

centos 6.6
apache hadoop 2.6.4
tez 0.6.2

• Assignee: Unassigned
• Reporter: Nikhil Mulley
• Created: 20/Jul/16 08:10
• Updated: 20/Jul/16 15:05
• Resolved: 20/Jul/16 15:02

Description

Hi,

Timeline server service logs on 2.6.4 cluster (no security, no acls) show often these error and then an exception follows when tez job runs. Closely inspecting the code shows there is a possibility of tez itself reporting different domain for the same entity (one that is already also in the timeline store) and then getting skipped to handle the event and store the event timeline information.

>>>>
ERROR org.apache.hadoop.yarn.server.timeline.TimelineDataManager: Skip the timeline entity:

{ id: tez_container_1468970783049_0021_01_000002, type: TEZ_CONTAINER_ID }

>>>

>>>
org.apache.hadoop.yarn.exceptions.YarnException: The domain of the timeline entity

{ id: tez_container_1468970783049_0021_01_000002, type: TEZ_CONTAINER_ID }

is not allowed to be changed.
>>>

Issue Links

• is duplicated by: TEZ-3177 Non-DAG events should use the session domain or no domain if the data does not need protection