Huang Kuan Hao created HDDS-16327:
-------------------------------------
Summary: Read the block file length once per WriteChunk instead of
three fstat calls
Key: HDDS-16327
URL: https://issues.apache.org/jira/browse/HDDS-16327
Project: Apache Ozone
Issue Type: Improvement
Reporter: Huang Kuan Hao
Assignee: Huang Kuan Hao
FilePerBlockStrategy.writeChunk resolves the block file length via
FileChannel.size() three times per WriteChunk (twice on an overwrite), each an
uncached fstat:
1. validateChunkForOverwrite(channel, info)
-> ChunkUtils.isOverWriteRequested (ChunkUtils.java:419)
2. ChunkUtils.validateChunkSize(channel, info, ...) (ChunkUtils.java:537),
skipped when it is an overwrite
3. fileLengthBeforeWrite = channel.size() (FilePerBlockStrategy.java:181)
All three run before ChunkUtils.writeData, and WriteChunk apply for a given
block file is serialized, so the length is stable across them. WriteChunk is
the hottest datanode write path (once per chunk of every write, default
FILE_PER_BLOCK layout).
Fix: read channel.size() once after obtaining the channel and pass it into
isOverWriteRequested / validateChunkSize / fileLengthBeforeWrite (adding an
overload that takes the already-read length) instead of re-stat-ing. Behavior
unchanged.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
