phoenix-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Hadoop QA (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-5105) Push Filter through Sort for SortMergeJoin
Date Wed, 23 Jan 2019 16:49:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-5105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16750206#comment-16750206
] 

Hadoop QA commented on PHOENIX-5105:
------------------------------------

{color:red}-1 overall{color}.  Here are the results of testing the latest attachment 
  http://issues.apache.org/jira/secure/attachment/12955974/PHOENIX-5015_v2-4.x-HBase-1.4.patch
  against 4.x-HBase-1.4 branch at commit c8686ce73093b879bbda5f08f3e7933a9708e110.
  ATTACHMENT ID: 12955974

    {color:green}+1 @author{color}.  The patch does not contain any @author tags.

    {color:green}+1 tests included{color}.  The patch appears to include 13 new or modified
tests.

    {color:green}+1 javac{color}.  The applied patch does not increase the total number of
javac compiler warnings.

    {color:red}-1 release audit{color}.  The applied patch generated 1 release audit warnings
(more than the master's current 0 warnings).

    {color:red}-1 lineLengths{color}.  The patch introduces the following lines longer than
100:
    +            sql="select /*+ USE_SORT_MERGE_JOIN */ a.aid,b.code from (select aid,age
from "+tableName1+" where age >=11 and age<=33) a inner join "+
+                "(select bid,code from "+tableName2+" order by code limit 2) b on a.aid=b.bid
where b.code > 50";
+            sql="select /*+ USE_SORT_MERGE_JOIN */ a.aid,b.codesum from (select aid,sum(age)
agesum from "+tableName1+" where age >=11 and age<=33 group by aid order by agesum limit
3) a inner join "+
+                "(select bid,sum(code) codesum from "+tableName2+" group by bid order by
codesum limit 2) b on a.aid=b.bid where b.codesum > 50";
+                "(select a.aid,b.code from "+tableName1+" a inner join "+tableName2+" b on
a.aid=b.bid where b.code >=44 and b.code<=66 order by b.code limit 3) t1 inner join
"+
+                "(select a.aid,c.region from "+tableName1+" a inner join "+tableName3+" c
on a.aid=c.cid where c.region>=77 and c.region<=99 order by c.region desc limit 1) t2
on t1.aid=t2.aid "+
+                "(select a.aid,sum(b.code) codesum from "+tableName1+" a inner join "+tableName2+"
b on a.aid=b.bid where b.code >=44 and b.code<=66 group by a.aid order by codesum limit
3) t1 inner join "+
+                "(select a.aid,sum(c.region) regionsum from "+tableName1+" a inner join "+tableName3+"
c on a.aid=c.cid where c.region>=77 and c.region<=99 group by a.aid order by regionsum
desc limit 2) t2 on t1.aid=t2.aid "+
+                "(select a.aid,sum(b.code) codesum from "+tableName1+" a inner join "+tableName2+"
b on a.aid=b.bid where b.code >=44 and b.code<=66 group by a.aid order by a.aid,codesum
limit 3) t1 inner join "+
+                "(select a.aid,sum(c.region) regionsum from "+tableName1+" a inner join "+tableName3+"
c on a.aid=c.cid where c.region>=77 and c.region<=99 group by a.aid order by a.aid desc,regionsum
desc limit 2) t2 on t1.aid=t2.aid "+

     {color:red}-1 core tests{color}.  The patch failed these unit tests:
     ./phoenix-core/target/failsafe-reports/TEST-org.apache.phoenix.end2end.index.MutableIndexIT

Test results: https://builds.apache.org/job/PreCommit-PHOENIX-Build/2269//testReport/
Release audit warnings: https://builds.apache.org/job/PreCommit-PHOENIX-Build/2269//artifact/patchprocess/patchReleaseAuditWarnings.txt
Console output: https://builds.apache.org/job/PreCommit-PHOENIX-Build/2269//console

This message is automatically generated.

> Push Filter through Sort for SortMergeJoin
> ------------------------------------------
>
>                 Key: PHOENIX-5105
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-5105
>             Project: Phoenix
>          Issue Type: Improvement
>    Affects Versions: 4.14.1
>            Reporter: chenglei
>            Assignee: chenglei
>            Priority: Major
>             Fix For: 4.15.0
>
>         Attachments: PHOENIX-5015-4.x-HBase-1.4.patch, PHOENIX-5015_v2-4.x-HBase-1.4.patch
>
>          Time Spent: 3h 50m
>  Remaining Estimate: 0h
>
> Given two tables:
> {code:java}
>           CREATE TABLE merge1 ( 
>                     aid INTEGER PRIMARY KEY,
>                     age INTEGER)
>           
>           CREATE TABLE merge2  ( 
>                     bid INTEGER PRIMARY KEY,
>                     code INTEGER)
> {code}
> for following sql :
> {code:java}
> select /*+ USE_SORT_MERGE_JOIN */ a.aid,b.code from 
> (select aid,age from merge1 where age >=11 and age<=33 order by age limit 3) a
inner join 
> (select bid,code from merge2 order by code limit 1) b on a.aid=b.bid where b.code >
50
> {code}
> For the RHS of SortMergeJoin, at first the where condition {{b.code > 50}} is pushed
down to RHS as its {{JoinCompiler.Table.postFilters}},  then {{order by b.bid}} is appended
to RHS and it is rewritten as 
>  {{select bid,code from (select bid,code from merge2 order by code limit 1) order by
bid}}
> by following line 211 in {{QueryCompiler.compileJoinQuery}}.
> Next the above rewritten sql is compiled to ClientScanPlan by following line 221 ,and
previously pushed down {{b.code > 50}} is compiled by {{table.compilePostFilterExpression}}
method in following line 224 to filter the result of the preceding ClientScanPlan. The problem
here is that we execute the {{order by bid}} first and then the postFilter {{b.code > 50}},
obviously it is inefficient. In fact, we can directly rewrite the RHS as 
>  {{select bid,code from (select bid,code from merge2 order by code limit 1) order by
bid where code > 50}} 
>  to first filter {{b.code > 50}} and then execute the {{order by bid}} .
> {code:java}
> 208    protected QueryPlan compileJoinQuery(StatementContext context, List<Object>
binds, JoinTable joinTable, boolean asSubquery, boolean projectPKColumns, List<OrderByNode>
orderBy) throws SQLException {
> 209         if (joinTable.getJoinSpecs().isEmpty()) {
> 210              Table table = joinTable.getTable();
> 211               SelectStatement subquery = table.getAsSubquery(orderBy);
> 212              if (!table.isSubselect()) {
> 213                  context.setCurrentTable(table.getTableRef());
> 214                  PTable projectedTable = table.createProjectedTable(!projectPKColumns,
context);
> 215                  TupleProjector projector = new TupleProjector(projectedTable);
> 216                  TupleProjector.serializeProjectorIntoScan(context.getScan(), projector);
> 217                  context.setResolver(FromCompiler.getResolverForProjectedTable(projectedTable,
context.getConnection(), subquery.getUdfParseNodes()));
> 218                  table.projectColumns(context.getScan());
> 219                  return compileSingleFlatQuery(context, subquery, binds, asSubquery,
!asSubquery, null, projectPKColumns ? projector : null, true);
> 220            }
> 221            QueryPlan plan = compileSubquery(subquery, false);
> 222            PTable projectedTable = table.createProjectedTable(plan.getProjector());
> 223            context.setResolver(FromCompiler.getResolverForProjectedTable(projectedTable,
context.getConnection(), subquery.getUdfParseNodes()));
> 224            return new TupleProjectionPlan(plan, new TupleProjector(plan.getProjector()),
table.compilePostFilterExpression(context));
> 225        }
> {code}



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message