[CALCITE-2907] AggregateExpandDistinctAggregatesRule produces a wrong relational algebra

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Invalid
• Affects Version/s: 1.18.0
• Component/s: core
• Assignee: Unassigned
• Reporter: Lai Zhou
• Created: 11/Mar/19 06:43
• Updated: 11/Mar/19 11:44
• Resolved: 11/Mar/19 11:44

Description

In my usecase:

an Aggregate which contains distinct call was converted improperly to  a wrong relational algebra.

 

SELECT user_id,
                       order_id,
                       product_id,
                            count(DISTINCT secured_libs.u51decrypt(phone)) AS contact_count,
                            stddev_pop(secured_libs.u51decrypt(phone)) AS seg3stddev,
                            entropy(secured_libs.u51decrypt(phone)) AS seg3entropy

                FROM dw_risk__mygravitation_v_snap_contacts_contacts
                GROUP BY user_id,
                         order_id,
                         product_id

 

 

After digging into the code,I found at the line 444 of the AggregateExpandDistinctAggregatesRule.java :

 

int x = groupCount;
final List<AggregateCall> newCalls = new ArrayList<>();
for (AggregateCall aggCall : aggregate.getAggCallList()) {
  final int newFilterArg;
  final List<Integer> newArgList;
  final SqlAggFunction aggregation;
  if (!aggCall.isDistinct()) {
    aggregation = SqlStdOperatorTable.MIN;
    newArgList = ImmutableIntList.of(x++);
    newFilterArg = filters.get(aggregate.getGroupSet());
  } else {

the undistinct aggregate call `stddev_pop` and `entropy` was converted to a 

 SqlStdOperatorTable.MIN,（actually all undistinct aggregate call here will be converted to a SqlStdOperatorTable.MIN ）.

I don't understand how it works.

I guess someone made a faulty assumption here. julianhyde ,can someone help me ASAP ?

It’s very important for my business.