[WHIRR-645] Race condition between collocated namenode and jobtracker start/init services

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.7.1
• Fix Version/s: 0.8.1, 0.9.0
• Component/s: core
• Assignee: Graham Gear
• Reporter: Graham Gear
• Created: 23/Aug/12 21:15
• Updated: 24/Sep/12 15:10
• Resolved: 23/Aug/12 23:25

Description

If the jobtracker role is defined before the namenode role on the same node (as per majority of the examples), the jobtracker service will poll for namenode and race the namenode HDFS dir config steps, often resulting in a failed write of 'mapred.system.dir' and a dead jobtracker process.

Issue Links

• breaks: WHIRR-656 HDFS directory creation can deadlock if the cluster being brought up is single node & preformatted