Uploaded image for project: 'Apache Drill'
  1. Apache Drill
  2. DRILL-5374

Parquet filter pushdown does not prune partition with nulls when predicate uses float column

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Open
    • Major
    • Resolution: Unresolved
    • 1.9.0
    • None
    • None

    Description

      Drill does not prune enough partitions for this query when filter pushdown is used with metadata caching. The float column is being compared with a double value.

      0: jdbc:drill:zk=10.10.100.186:5181/drill/rho> select count(*) from orders_parts_metadata where float_id < 1100.0;
      

      To reproduce the problem, put the attached files into a directory. Then

      create the metadata:
      refresh table metadata dfs.`path_to_directory`;
      

      For example, if you put the files in /drill/testdata/filter/orders_parts_metadata, then run this sql command

      refresh table metadata dfs.`/drill/testdata/filter/orders_parts_metadata`;
      

      Attachments

        1. 0_0_1.parquet
          211 kB
          Robert Hou
        2. 0_0_2.parquet
          214 kB
          Robert Hou
        3. 0_0_3.parquet
          206 kB
          Robert Hou
        4. 0_0_4.parquet
          111 kB
          Robert Hou
        5. 0_0_5.parquet
          205 kB
          Robert Hou
        6. drill.parquet_metadata
          16 kB
          Robert Hou

        Activity

          People

            jni Jinfeng Ni
            rhou Robert Hou
            Robert Hou Robert Hou
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

            Dates

              Created:
              Updated: