[HADOOP-5652] Reduce does not respect in-memory segment memory limit when number of on disk segments == io.sort.factor

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.20.0
• Fix Version/s: 0.21.0
• Hadoop Flags: Reviewed
• Assignee: Christopher Douglas
• Reporter: Christopher Douglas
• Created: 10/Apr/09 03:51
• Updated: 24/Aug/10 20:37
• Resolved: 15/Apr/09 17:05

Description

If the number of on-disk segments is exactly io.sort.factor, then map output segments may be left in memory for the reduce contrary to the specification in mapred.job.reduce.input.buffer.percent.