Karthik Palanisamy created HBASE-23095:
------------------------------------------
Summary: Reuse FileStatus in StoreFileInfo
Key: HBASE-23095
URL: https://issues.apache.org/jira/browse/HBASE-23095
Project: HBase
Issue Type: Improvement
Components: mob, snapshots
Affects Versions: 2.2.1
Reporter: Karthik Palanisamy
Assignee: Karthik Palanisamy
Fix For: 3.0.0
Attachments: PerformanceComparision.pdf
The performance of create snapshot on large MOB table reasonably slow because
there are two unnecessary calls to namenode on each Hfile, this while we create
snapshot manifest. The first namenode call for getting StoreFile modification
time
[link|[https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreFileInfo.java#L139]]
which used for metrics and another namenode call for getting StoreFile size
[link|[https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/snapshot/SnapshotManifestV2.java#L132]]
which used in snapshot manifest. Both calls can be avoided and this info can
be fetched from existing FileStatus
[link|[https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreFileInfo.java#L155]].
PFA. 2x performance is seen after reusing existing FileStatus.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
