[IMPALA-11843] IndexOutOfBoundsException in analytic limit pushdown

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: Impala 4.0.0, Impala 4.1.0, Impala 4.2.0, Impala 4.1.1
• Fix Version/s: Impala 4.1.2, Impala 4.3.0
• Component/s: Frontend
• Target Version: Impala 4.1.2
• Epic Color: ghx-label-11
• Assignee: Quanlong Huang
• Reporter: Quanlong Huang
• Created: 16/Jan/23 08:05
• Updated: 25/May/23 07:09
• Resolved: 17/Jan/23 02:40

Description

The following query fails with IndexOutOfBoundsException:

create table tbl (id int);

select id from (
  select id, 
    row_number() over (order by id) rn, 
    max(id) over () max_id
  from tbl
) t
where id = max_id and rn < 10;

ERROR: IndexOutOfBoundsException: Index: 0, Size: 0

The stacktrace in logs:

I0116 15:55:46.766265 23944 Frontend.java:2062] be402cb92ecc5490:11cbe79000000000] Analysis and authorization finished.
I0116 15:55:46.803608 23944 jni-util.cc:288] be402cb92ecc5490:11cbe79000000000] java.lang.IndexOutOfBoundsException: Index: 0, Size: 0
        at java.util.ArrayList.rangeCheck(ArrayList.java:659)
        at java.util.ArrayList.get(ArrayList.java:435)
        at org.apache.impala.planner.AnalyticPlanner.inferPartitionLimits(AnalyticPlanner.java:914)
        at org.apache.impala.planner.AnalyticPlanner.createSingleNodePlan(AnalyticPlanner.java:115)
        at org.apache.impala.planner.SingleNodePlanner.createQueryPlan(SingleNodePlanner.java:295)
        at org.apache.impala.planner.SingleNodePlanner.createInlineViewPlan(SingleNodePlanner.java:1244)
        at org.apache.impala.planner.SingleNodePlanner.createTableRefNode(SingleNodePlanner.java:2208)
        at org.apache.impala.planner.SingleNodePlanner.createTableRefsPlan(SingleNodePlanner.java:931)
        at org.apache.impala.planner.SingleNodePlanner.createSelectPlan(SingleNodePlanner.java:750)
        at org.apache.impala.planner.SingleNodePlanner.createQueryPlan(SingleNodePlanner.java:278)
        at org.apache.impala.planner.SingleNodePlanner.createSingleNodePlan(SingleNodePlanner.java:170)
        at org.apache.impala.planner.Planner.createPlanFragments(Planner.java:120)
        at org.apache.impala.planner.Planner.createPlans(Planner.java:249)
        at org.apache.impala.service.Frontend.createExecRequest(Frontend.java:1733)
        at org.apache.impala.service.Frontend.getPlannedExecRequest(Frontend.java:2344)
        at org.apache.impala.service.Frontend.doCreateExecRequest(Frontend.java:2181)
        at org.apache.impala.service.Frontend.getTExecRequest(Frontend.java:1967)
        at org.apache.impala.service.Frontend.createExecRequest(Frontend.java:1789)
        at org.apache.impala.service.JniFrontend.createExecRequest(JniFrontend.java:164)

There is a predicate "rn < 10" on the row_number() results so the limit is considered push down into the inline view to make it a TopN query. While considering the conjuncts, the other predicate "id = max_id" is also checked. It fails the following code in AnalyticPlanner.inferPartitionLimits():

      List<Expr> lhsSourceExprs = ((SlotRef) lhs).getDesc().getSourceExprs();
      if (lhsSourceExprs.size() > 1 ||
            !(lhsSourceExprs.get(0) instanceof AnalyticExpr)) {
        continue;
      }

https://github.com/apache/impala/blob/f2f6b4b5804df036a5a7dc8ff23f8a0537b5bf97/fe/src/main/java/org/apache/impala/planner/AnalyticPlanner.java#L912-L916

'lhsSourceExprs' is empty since "id" is a slot ref for a source table column. Thus lhsSourceExprs.get(0) throws the IndexOutOfBoundsException.

To workaround this bug, users can disable this optimization by query option:

set ANALYTIC_RANK_PUSHDOWN_THRESHOLD=0;

Issue Links

• is caused by: IMPALA-9979 Backend partitioned top-n operator
• is related to: IMPALA-12164 The query fails with "IllegalStateException: Illegal reference to non-materialized slot: tid=x sid=x"
• links to: Code Review