[HDFS-3937] The BlockManager should not use the FSN lock

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: namenode
• Assignee: Unassigned
• Reporter: Eli Collins
• Created: 14/Sep/12 06:05
• Updated: 13/Nov/14 22:20

Description

Like HDFS-2206 but for the BM. The BM currently calls into the FSN to acquire it's write lock, it looks like this is only needed for synchronization of it's local structures, in which case it can have it's own RW lock. This helps us further decouple block and namespace management, and also avoids scenarios like HDFS-3936 where the FSN holds the lock and calls into BM which may be blocked trying to acquire the FSN lock.

Per Todd's comment in HDFS-2206 we should do this as part of revisiting NN locking (HDFS-2184) vs just a local change.

Issue Links

• is related to: HDFS-3936 MiniDFSCluster shutdown races with BlocksMap usage
• relates to: HDFS-2206 Add read-write lock to DatanodeManager
• relates to: HDFS-2184 Revisit Namenode locking