[HADOOP-726] HDFS locking mechanisms should be simplified or removed

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.11.0
• Assignee: Raghu Angadi
• Reporter: Sameer Paranjpye
• Created: 15/Nov/06 23:40
• Updated: 08/Jul/09 16:42
• Resolved: 25/Jan/07 21:02

Description

HDFS includes a locking mechanism that allows clients to lock files and directories in the filesystem. This requires that the Namenode, in addition to all other management of filesystem state also has to act like a lock manager on behalf of clients. Rather than burden the Namenode with this Hadoop should include a distinct general purpose distributed lock manager.

In the interim, the locking functionality on the filesystem should be simplified to permit only file locks or removed altogether. At this point, removing or simplifying the functionality is not likely to break client code, removing it later will be much harder. In any case, locks in HDFS don't work very well, see HADOOP-656. Also, directory locks are overkill most UNIX fses don't support it.

Issue Links

• relates to: HADOOP-656 dfs locking doesn't notify the application when a lock is lost