[CALCITE-1805] Druid adapter incorrectly pushes down "COUNT(c)"; Druid only supports "COUNT(*)"

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.12.0
• Fix Version/s: 1.13.0
• Component/s: druid-adapter
• Assignee: Slim Bouguerra
• Reporter: Slim Bouguerra
• Created: 24/May/17 20:53
• Updated: 27/Feb/24 22:24
• Resolved: 20/Jun/17 06:41

Description

Currently queries like

select count(column) from table

is pushed to druid as timeseries with an aggregator

 {"type":"count","name":"EXPR$0","fieldName":"countryName"} 

Such an aggregator does not exist in druid. The count aggregator does only work as

 count(*) 

.
here is a test case that summarize the issue.

  @Test public void testCount() {

    final String sql = "SELECT count(\"countryName\") FROM (SELECT \"countryName\" FROM \"wikiticker\" WHERE \"countryName\"  IS NOT NULL) as a"; // correct count
    sql(sql, WIKI_AUTO2).returnsUnordered("EXPR$0=3799");

    final String sql2 = "SELECT count(\"countryName\") FROM (SELECT \"countryName\" FROM \"wikiticker\") as a";
    sql(sql2, WIKI_AUTO2).returnsUnordered("EXPR$0=3799"); // it will fail
  }

First test will pass while the second will not.

Issue Links

• depends upon: CALCITE-1436 AggregateNode NPE for aggregators other than SUM/COUNT