You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@calcite.apache.org by "Julian Hyde (Jira)" <ji...@apache.org> on 2020/10/07 19:48:00 UTC
[jira] [Commented] (CALCITE-4317) Some rules fail to handle
Aggregate node if RelFieldTrimmer trims all the fields
[ https://issues.apache.org/jira/browse/CALCITE-4317?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17209816#comment-17209816 ]
Julian Hyde commented on CALCITE-4317:
--------------------------------------
We don't allow {{RelNode}} with zero fields, so {{RelFieldTrimmer}} is at fault here.
A zero-field {{Aggregate}} would return one row for all possible inputs, even if the input is empty and/or has zero columns. Is that the semantic we want here? (We had to work around this problem in {{AggregateProjectPullUpConstantsRule}}.)
Should Calcite allow {{RelNode}} with zero fields? I think it might cause a lot of subtle problems. If we want to change that, it would be bigger discussion.
> Some rules fail to handle Aggregate node if RelFieldTrimmer trims all the fields
> --------------------------------------------------------------------------------
>
> Key: CALCITE-4317
> URL: https://issues.apache.org/jira/browse/CALCITE-4317
> Project: Calcite
> Issue Type: Bug
> Components: core
> Reporter: Rafay A
> Priority: Major
>
> Consider this query:
> {code:java}
> select o_orderkey from
> (SELECT count(*) cnt_star, count(l_orderkey) cnt_ok FROM lineitem)
> cross join orders
> limit 10
> {code}
> which generates this plan:
> {code:java}
> LogicalSort(fetch=[10])
> LogicalProject(o_orderkey=[$2])
> LogicalJoin(condition=[true], joinType=[inner])
> LogicalAggregate(group=[{}], cnt_star=[COUNT()], cnt_ok=[COUNT($0)])
> LogicalProject(l_orderkey=[$0])
> ScanCrel(table=[lineitem], columns=[`l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment`])
> ScanCrel(table=[orders], columns=[`o_orderkey`, `o_custkey`, `o_orderstatus`, `o_totalprice`, `o_orderdate`, `o_orderpriority`, `o_clerk`, `o_shippriority`, `o_comment`])
> {code}
> and after we apply RelFieldTrimmer:
> {code:java}
> LogicalSort(fetch=[10])
> LogicalProject(o_orderkey=[$0])
> LogicalJoin(condition=[true], joinType=[inner])
> LogicalAggregate(group=[{}])
> LogicalProject(l_orderkey=[$0])
> ScanCrel(table=[lineitem], columns=[`l_orderkey`])
> ScanCrel(table=[orders], columns=[`o_orderkey`])
> {code}
>
> We see that the LogicalAggregate has no groups/functions after applying the trimmer. When we apply ProjectJoinTransposeRule or LoptOptimizeJoinRule after trimming, we are seeing these exceptions:
> For LoptOptimizeJoinRule:
> {code:java}
> (java.lang.ArrayIndexOutOfBoundsException) 0
> com.google.common.collect.RegularImmutableList.get():75
> org.apache.calcite.rel.metadata.RelMdColumnOrigins.getColumnOrigins():77
> sun.reflect.NativeMethodAccessorImpl.invoke0():-2
> sun.reflect.NativeMethodAccessorImpl.invoke():62
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> org.apache.calcite.rel.metadata.ReflectiveRelMetadataProvider$1$1.invoke():178
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> sun.reflect.GeneratedMethodAccessor144.invoke():-1
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> org.apache.calcite.rel.metadata.ChainedRelMetadataProvider$ChainedInvocationHandler.invoke():139
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> sun.reflect.GeneratedMethodAccessor144.invoke():-1
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> sun.reflect.GeneratedMethodAccessor144.invoke():-1
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> org.apache.calcite.rel.metadata.ChainedRelMetadataProvider$ChainedInvocationHandler.invoke():139
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> sun.reflect.GeneratedMethodAccessor144.invoke():-1
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> sun.reflect.GeneratedMethodAccessor144.invoke():-1
> sun.reflect.DelegatingMethodAccessorImpl.invoke():43
> java.lang.reflect.Method.invoke():498
> org.apache.calcite.rel.metadata.RelMetadataQuery$MetadataInvocationHandler.invoke():110
> com.sun.proxy.$Proxy175.getColumnOrigins():-1
> org.apache.calcite.rel.metadata.RelMetadataQuery.getColumnOrigins():293
> org.apache.calcite.rel.metadata.RelMetadataQuery.getTableOrigin():348
> org.apache.calcite.rel.rules.LoptOptimizeJoinRule.getSimpleFactors():377
> org.apache.calcite.rel.rules.LoptOptimizeJoinRule.findRemovableSelfJoins():297
> org.apache.calcite.rel.rules.LoptOptimizeJoinRule.onMatch():126
> org.apache.calcite.plan.AbstractRelOptPlanner.fireRule():317
> org.apache.calcite.plan.hep.HepPlanner.applyRule():556
> org.apache.calcite.plan.hep.HepPlanner.applyRules():415
> org.apache.calcite.plan.hep.HepPlanner.executeInstruction():280
> org.apache.calcite.plan.hep.HepInstruction$RuleCollection.execute():74
> org.apache.calcite.plan.hep.HepPlanner.executeProgram():211
> org.apache.calcite.plan.hep.HepPlanner.findBestExp():198
> {code}
> For ProjectJoinTransposeRule:
> {code:java}
> (java.lang.ArrayIndexOutOfBoundsException) 0
> com.google.common.collect.RegularImmutableList.get():75
> org.apache.calcite.rel.rules.PushProjector.createProjectRefsAndExprs():477
> org.apache.calcite.rel.rules.ProjectJoinTransposeRule.onMatch():138
> org.apache.calcite.plan.AbstractRelOptPlanner.fireRule():317
> org.apache.calcite.plan.hep.HepPlanner.applyRule():556
> org.apache.calcite.plan.hep.HepPlanner.applyRules():415
> org.apache.calcite.plan.hep.HepPlanner.executeInstruction():280
> org.apache.calcite.plan.hep.HepInstruction$RuleCollection.execute():74
> org.apache.calcite.plan.hep.HepPlanner.executeProgram():211
> org.apache.calcite.plan.hep.HepPlanner.findBestExp():198
> {code}
> I think we should either fix the trimmer, or the rules to handle these cases.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)