[HDDS-2382] Consider reducing number of file::exists() calls during write operation

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Component/s: Ozone Datanode
• Labels: Triaged performance
• Target Version/s: 1.0.0
• Assignee: Attila Doroszlai
• Reporter: Rajesh Balamohan
• Created: 30/Oct/19 09:15
• Updated: 02/Jun/20 18:24
• Resolved: 02/Jun/20 18:24

Description

When writing 100-200 MB files with multiple threads, observed lots of file::exists() checks.

For every 16 MB chunk, it ends up checking whether chunksLoc directory exists or not. (ref: https://github.com/apache/hadoop-ozone/blob/master/hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/keyvalue/helpers/ChunkUtils.java#L239)

Also, this check (ChunkUtils.getChunkFile) happens from 2 places.

1.org.apache.hadoop.ozone.container.common.transport.server.ratis.ContainerStateMachine.lambda$handleWriteChunk

2.org.apache.hadoop.ozone.container.common.transport.server.ratis.ContainerStateMachine.lambda$applyTransaction

Note that these are folders and not actual chunk filenames. It would be helpful to reduce this check, if we track create/delete of these folders.

Issue Links

• is superceded by: HDDS-3703 Consider avoiding file lookup calls in writeChunk hotpath