[HAWQ-979] Resource Broker Should Reconnect Hadoop Yarn When Failed to Get Cluster Report

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.1.0.0-incubating
• Component/s: Resource Manager
• Assignee: Wen Lin
• Reporter: Wen Lin
• Created: 04/Aug/16 08:38
• Updated: 18/Feb/17 05:27
• Resolved: 05/Aug/16 02:30

Description

While HAWQ with yarn mode is running, sometimes the heartbeat thread of libyarn maybe fail(e.g. YARN RM restarts) and quit,

2016-08-03 18:45:27.913838 PDT,,,p34645,th-1290610400,,,,0,con4,,seg-10000,,,,,"WARNING","01000","YARN mode resource broker failed to get YARN queue report of queue default. LibYarnClient::getQueueInfo, Catch the Exception:LibYarnClient::libyarn AM heartbeat thread has stopped.",,,,,,,0,,"resourcebroker_LIBYARN_proc.c",1840,

resource broker process should re-register HAWQ to YARN in this case, but actually not.

The reason is:
In function handleRM2RB_GetClusterReport(), when RB2YARN_getQueueReport() failed, function sendRBGetClusterReportErrorData() is called, but sendRBGetClusterReportErrorData() returns OK(should return RESBROK_ERROR_GRM)

Issue Links

• links to: GitHub Pull Request #834
• links to: GitHub Pull Request #836