ZanderXu created HDFS-17481:
-------------------------------
Summary: [FGL] ComputeFileSize in INodeFile should avoid BM lock
Key: HDFS-17481
URL: https://issues.apache.org/jira/browse/HDFS-17481
Project: Hadoop HDFS
Issue Type: Sub-task
Reporter: ZanderXu
Assignee: ZanderXu
GetFileInfo is a very common used operations. The result FileStatus contains
the file size, which is computed by computeFileSize in InodeFile.
Since it involves the states of block, it needs to hold the BM readLock. But
holding the BM readLock will impact the performance, since getFileInfo is very
common used RPC.
As we all known, block size may be changed only by the following operations:
# fsync
# updatePipeline
# commit block
# commitBlockSynchronization
# truncate
# updates block in FSEditLogLoader
# forceCompleteBlock in FSEditLogLoader
All these above operations will be locked by the directory tree or global write
lock, so computeFileSize can be handled without holding BM Lock.
But the BR and IBM may change the block status from COMMIT to COMPLETE even
through the block size is not changed. So computeFileSize cannot using
isComplete to compute the size of last block.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org
