[
https://issues.apache.org/jira/browse/HBASE-18555?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16125114#comment-16125114
]
Xiang Li commented on HBASE-18555:
----------------------------------
Thanks Jerry, Chia-Ping Tsai and Mike for the review!
> Remove redundant familyMap.put() from addxxx() of sub-classes of Mutation and
> Query
> -----------------------------------------------------------------------------------
>
> Key: HBASE-18555
> URL: https://issues.apache.org/jira/browse/HBASE-18555
> Project: HBase
> Issue Type: Improvement
> Components: Client
> Reporter: Xiang Li
> Assignee: Xiang Li
> Priority: Minor
> Fix For: 3.0.0, 1.5.0, 2.0.0-alpha-2
>
> Attachments: HBASE-18555.master.000.patch,
> HBASE-18555.master.001.patch, HBASE-18555.master.002.patch
>
>
> In addxxx() functions of Mutation(Append, Delete, Increment and Put) and
> Query(Get and Scan), there are redundant Map#put() calls which could be
> removed to improve the performance.
> For example, in Put#addColumn() and addImmutable(), after getting the cell
> list of the given family and add the cell into the list, the code puts
> (key=family, value=list) into familyMap.
> In addColumn(), it is like
> {code}
> List<Cell> list = getCellList(family);
> KeyValue kv = createPutKeyValue(family, qualifier, ts, value);
> list.add(kv);
> familyMap.put(CellUtil.cloneFamily(kv), list); // <-- here
> return this;
> {code}
> In addImmutable(), it is like
> {code}
> List<Cell> list = getCellList(family);
> KeyValue kv = createPutKeyValue(family, qualifier, ts, value, tag);
> list.add(kv);
> familyMap.put(family, list); // <-- here
> return this;
> {code}
> I think those put() for Map only take effect when getCellList(family) returns
> a new allocated ArrayList. When the list for a family already exist, put()
> for Map will update the value to the reference of the list, but actually, the
> reference of the list is not changed.
> Those put() do not do any harm in terms of the correctness when they are
> here. But it could be removed to improve the performance. familyMap searches
> for key and set the new value and return the old value. Those operation take
> some time but actually are not needed.
> The put() could be moved into Mutation#getCellList(family)
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
