Uploaded image for project: 'Hive'
  1. Hive
  2. HIVE-28363

Improve heuristics of FilterStatsRule without column stats

    XMLWordPrintableJSON

Details

    Description

      HIVE-19097 changed the formula to estimate # of rows selected by FilterOperator. This ticket would try to improve the case where column stats are unavailable.

       

      This is an example. The table has ten rows and no column stats on `id`.

      0: jdbc:hive2://hive-hiveserver2:10000/defaul> DESCRIBE FORMATTED users id;
      ...
      +------------------------+-----------------------------+
      |    column_property     |            value            |
      +------------------------+-----------------------------+
      | col_name               | id                          |
      | data_type              | int                         |
      | min                    |                             |
      | max                    |                             |
      | num_nulls              |                             |
      | distinct_count         |                             |
      | avg_col_len            |                             |
      | max_col_len            |                             |
      | num_trues              |                             |
      | num_falses             |                             |
      | bit_vector             |                             |
      | comment                | from deserializer           |
      | COLUMN_STATS_ACCURATE  | {\"BASIC_STATS\":\"true\"}  |
      +------------------------+-----------------------------+

      With a single needle, the estimated number becomes 10 * 0.5 = 5 because of the fallback heuristics.

      0: jdbc:hive2://hive-hiveserver2:10000/defaul> EXPLAIN SELECT * FROM users WHERE id IN (1);
      ...
      |                 TableScan                          |
      |                   alias: users                     |
      |                   filterExpr: (id = 1) (type: boolean) |
      |                   Statistics: Num rows: 10 Data size: 11 Basic stats: COMPLETE Column stats: NONE |
      |                   Filter Operator                  |
      |                     predicate: (id = 1) (type: boolean) |
      |                     Statistics: Num rows: 5 Data size: 5 Basic stats: COMPLETE Column stats: NONE | 

      The size is estimated to be the original size with two or more needles. The heuristics estimate the size as min(10, 10 * 0.5 * N) = 10. However, I believe users expect to observe some reduction when using IN.

      0: jdbc:hive2://hive-hiveserver2:10000/defaul> EXPLAIN SELECT * FROM users WHERE id IN (1, 2);
      |                 TableScan                          |
      |                   alias: users                     |
      |                   filterExpr: (id) IN (1, 2) (type: boolean) |
      |                   Statistics: Num rows: 10 Data size: 11 Basic stats: COMPLETE Column stats: NONE |
      |                   Filter Operator                  |
      |                     predicate: (id) IN (1, 2) (type: boolean) |
      |                     Statistics: Num rows: 10 Data size: 11 Basic stats: COMPLETE Column stats: NONE | 

       

      Attachments

        Issue Links

          Activity

            People

              okumin Shohei Okumiya
              okumin Shohei Okumiya
              Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: