Uploaded image for project: 'Apache Drill'
  1. Apache Drill
  2. DRILL-6465

Transitive closure is not working in Drill for Join with multiple local conditions

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Open
    • Minor
    • Resolution: Unresolved
    • 1.14.0
    • None
    • None
    • None

    Description

      For several SQL operators Transitive closure is not working during Partition Pruning and Filter Pushdown for the left table in Join.
      If I use several local conditions, then Drill scans full left table in Join.
      But if we move additional conditions to the WHERE statement, then Transitive closure works fine for all joined tables

      Query BETWEEN:

      EXPLAIN PLAN FOR
SELECT * FROM hive.`h_tab1` t1
JOIN hive.`h_tab2` t2
ON t1.y=t2.y
AND t2.y BETWEEN 1987 AND 1988;

      Expected result:

      Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab1), columns=[`**`], numPartitions=8, partitions= [Partition(values:[1987, 5, 1]), Partition(values:[1987, 5, 2]), Partition(values:[1987, 7, 1]), Partition(values:[1987, 7, 2]), Partition(values:[1988, 11, 1]), Partition(values:[1988, 11, 2]), Partition(values:[1988, 12, 1]), Partition(values:[1988, 12, 2])]

      Actual result:

      Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab1), columns=[`**`], numPartitions=16, partitions= [Partition(values:[1987, 5, 1]), Partition(values:[1987, 5, 2]), Partition(values:[1987, 7, 1]), Partition(values:[1987, 7, 2]), Partition(values:[1988, 11, 1]), Partition(values:[1988, 11, 2]), Partition(values:[1988, 12, 1]), Partition(values:[1988, 12, 2]), Partition(values:[1990, 4, 1]), Partition(values:[1990, 4, 2]), Partition(values:[1990, 5, 1]), Partition(values:[1990, 5, 2]), Partition(values:[1991, 3, 1]), Partition(values:[1991, 3, 2]), Partition(values:[1991, 3, 3]), Partition(values:[1991, 3, 4])
]

      There is the same Transitive closure behavior for this logical operators:

      • NOT IN
      • LIKE
      • NOT LIKE

      Also Transitive closure is not working during Partition Pruning and Filter Pushdown for this comparison operators:

      Query <

      EXPLAIN PLAN FOR
SELECT * FROM hive.`h_tab1` t1
JOIN hive.`h_tab2` t2
ON t1.y=t2.y
AND t2.y < 1988;

      Expected result:

      Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab1), columns=[`**`], numPartitions=4, partitions= [Partition(values:[1987, 5, 1]), Partition(values:[1987, 5, 2]), Partition(values:[1987, 7, 1]), Partition(values:[1987, 7, 2])]

      Actual result:

      00-00 Screen
00-01 Project(itm=[$0], y=[$1], m=[$2], category=[$3], itm0=[$4], category0=[$5], y0=[$6], m0=[$7])
00-02 Project(itm=[$0], y=[$1], m=[$2], category=[$3], itm0=[$4], category0=[$5], y0=[$6], m0=[$7])
00-03 HashJoin(condition=[=($1, $6)], joinType=[inner])
00-05 Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab1), columns=[`**`], numPartitions=16, partitions= [Partition(values:[1987, 5, 1]), Partition(values:[1987, 5, 2]), Partition(values:[1987, 7, 1]), Partition(values:[1987, 7, 2]), Partition(values:[1988, 11, 1]), Partition(values:[1988, 11, 2]), Partition(values:[1988, 12, 1]), Partition(values:[1988, 12, 2]), Partition(values:[1990, 4, 1]), Partition(values:[1990, 4, 2]), Partition(values:[1990, 5, 1]), Partition(values:[1990, 5, 2]), Partition(values:[1991, 3, 1]), Partition(values:[1991, 3, 2]), Partition(values:[1991, 3, 3]), Partition(values:[1991, 3, 4])], inputDirectories=[maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/1, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/2, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/3, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/4, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/5, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/6, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/7, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/8, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/9, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/10, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/11, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/12, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/13, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/14, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/15, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab1/16]]])
00-04 Project(itm0=[$0], category0=[$1], y0=[$2], m0=[$3], $f4=[$4])
00-06 Project(itm=[$0], category=[$1], y=[$2], m=[$3], $f4=[<($2, 1988)])
00-07 Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab2), columns=[`itm`, `category`, `y`, `m`], numPartitions=2, partitions= [Partition(values:[1987, 5]), Partition(values:[1987, 7])], inputDirectories=[maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab2/1, maprfs:/drill/testdata/ctas/parquet/DRILL_6173/tab2/2]]])

      There is the same Transitive closure behavior for this comparison operators:**

      >, <=, >=, <>

       But if I use the WHERE clause instead of local Join condition, Transitive closure works correct for all joined tables.

      Query:

      EXPLAIN PLAN FOR
SELECT * FROM hive.`h_tab1` t1
JOIN hive.`h_tab2` t2
ON t1.y=t2.y
WHERE t2.y < 1988;

      Result:

      Scan(groupscan=[HiveScan [table=Table(dbName:default, tableName:h_tab1), columns=[`**`], numPartitions=4, partitions= [Partition(values:[1987, 5, 1]), Partition(values:[1987, 5, 2]), Partition(values:[1987, 7, 1]), Partition(values:[1987, 7, 2])]

       

      It was examples for Drill Partition Pruning queries with hive. The same issues are in the Drill Filter Pushdown queries.

      Filter Pushdown 'BETWEEN' query example:

      EXPLAIN PLAN FOR
SELECT * FROM dfs.ctas_parquet.`/DRILL_6173_pushdown/tab1` t1
JOIN dfs.ctas_parquet.`/DRILL_6173_pushdown/tab2` t2
ON t1.y=t2.y
AND t2.y BETWEEN 1987 AND 1988;

      Expected result:

      Scan(groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath [path=/drill/testdata/ctas/parquet/DRILL_6173_pushdown/tab1/0_0_7.parquet], ReadEntryWithPath 
...
ReadEntryWithPath [path=/drill/testdata/ctas/parquet/DRILL_6173_pushdown/tab1/0_0_6.parquet]], selectionRoot=maprfs:/drill/testdata/ctas/parquet/DRILL_6173_pushdown/tab1, numFiles=8, numRowGroups=8

      Actual result:

      Scan(groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath [path=maprfs:///drill/testdata/ctas/parquet/DRILL_6173_pushdown/tab1]], selectionRoot=maprfs:/drill/testdata/ctas/parquet/DRILL_6173_pushdown/tab1, numFiles=1, numRowGroups=16

       

      Issues can be reproduced using parquet files in the attachment "drill.zip"

       

      Attachments

        1. drill.zip
          90 kB
          Denys Ordynskiy

        Issue Links

          Is contained by

          Improvement - An improvement or enhancement to an existing feature or task. DRILL-6350 Umbrella jira which tracks issues connected to Transitive Closure Inference

          • Major - Major loss of function.
          • Open

          Activity

            People

              Unassigned Unassigned
              denysord88 Denys Ordynskiy
              Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

              Dates

                Created:
                Updated: