[HDFS-8547] The space reservation for RBW block is leaking

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Duplicate
• Target Version/s: 2.8.0
• Assignee: Unassigned
• Reporter: Kihwal Lee
• Created: 05/Jun/15 21:27
• Updated: 15/Jun/15 15:41
• Resolved: 15/Jun/15 15:40

Description

HDFS-6898 added the feature to reserve space on datanode when creating RBW replicas. We have noticed that nonDfsUsed increasing on some of the datanodes. Heap dump of datanodes has revealed that reservedForRbw was huge for each volume. There weren't many rbw blocks at that time, so datanode must leaking it.

Issue Links

• is duplicated by: HDFS-8072 Reserved RBW space is not released if client terminates while writing block
• relates to: HDFS-6898 DN must reserve space for a full block when an RBW block is created