Uploaded image for project: 'Apache Drill'
  1. Apache Drill
  2. DRILL-540

Allow querying hive views in Drill

    XMLWordPrintableJSON

Details

    Description

      Currently hive views cannot be queried from drill.

      This Jira aims to add support for Hive views in Drill.

      Implementation details:

      1. Drill persists it's views metadata in file with suffix .view.drill using json format. For example: 
      {
       "name" : "view_from_calcite_1_4",
       "sql" : "SELECT * FROM `cp`.`store.json`WHERE `store_id` = 0",
       "fields" : [ {
       "name" : "*",
       "type" : "ANY",
       "isNullable" : true
       } ],
       "workspaceSchemaPath" : [ "dfs", "tmp" ]
      }
      

      Later Drill parses the metadata and uses it to treat view names in SQL as a subquery.

            2. In Apache Hive metadata about views is stored in similar way to tables. Below is example from metastore.TBLS :

       

      TBL_ID |CREATE_TIME |DB_ID |LAST_ACCESS_TIME |OWNER |RETENTION |SD_ID |TBL_NAME  |TBL_TYPE      |VIEW_EXPANDED_TEXT                         |
      -------|------------|------|-----------------|------|----------|------|----------|--------------|-------------------------------------------|
      2      |1542111078  |1     |0                |mapr  |0         |2     |cview     |VIRTUAL_VIEW  |SELECT COUNT(*) FROM `default`.`customers` |
      
      

            3. So in Hive metastore views are considered as tables of special type. And main benefit is that we also have expanded SQL definition of views (just like in view.drill files). Also reading of the metadata is already implemented in Drill with help of thrift Metastore API.

            4. To enable querying of Hive views we'll reuse existing code for Drill views as much as possible. First in HiveSchemaFactory.getDrillTable for HiveReadEntry we'll convert the metadata to instance of View (which is actually model for data persisted in .view.drill files) and then based on this instance return new DrillViewTable. Using this approach drill will handle hive views the same way as if it was initially defined in Drill and persisted in .view.drill file. 

           5. For conversion of Hive types: from FieldSchema to RelDataType we'll reuse existing code from DrillHiveTable, so the conversion functionality will be extracted and used for both (table and view) fields type conversions. 

      Security implications

      Consider simple example case where we have users, 

      user0  user1 user2
                 \ /
                group12
      

      and  sample db where object names contains user or group who should access them      

      db_all
          tbl_user0
          vw_user0
          tbl_group12
          vw_group12
      

      There are two Hive authorization modes supported  by Drill - SQL Standart and Strorage Based  authorization. For SQL Standart authorization permissions were granted using SQL: 

      SET ROLE admin;
      GRANT SELECT ON db_all.tbl_user0 TO USER user0;
      GRANT SELECT ON db_all.vw_user0 TO USER user0;
      CREATE ROLE group12;
      GRANT ROLE group12 TO USER user1;
      GRANT ROLE group12 TO USER user2;
      GRANT SELECT ON db_all.tbl_group12 TO ROLE group12;
      GRANT SELECT ON db_all.vw_group12 TO ROLE group12;
      

      And for Storage based authorization permissions were granted using commands: 

      hadoop fs -chown user0:user0 /user/hive/warehouse/db_all.db/tbl_user0
      hadoop fs -chmod 700 /user/hive/warehouse/db_all.db/tbl_user0
      hadoop fs -chmod 750 /user/hive/warehouse/db_all.db/tbl_group12
      hadoop fs -chown user1:group12 /user/hive/warehouse/db_all.db/tbl_group12

       Then the following table shows us results of queries for both authorization models. 

                                                                                                              SQL Standart     |            Storage Based Authorization

      SQL user0 user1 user2     user0 user1 user2
      Queries executed using Drill :              
      SHOW TABLES IN hive.db_all;    all     all    all   Accessibe tables + all views Accessibe tables + all views Accessibe tables + all views
      SELECT * FROM hive.db_all.tbl_user0;                                          
      SELECT * FROM hive.db_all.vw_user0;                                          
      SELECT * FROM hive.db_all.tbl_group12;                                          
      SELECT * FROM hive.db_all.vw_group12;                                          
      SELECT * FROM INFORMATION_SCHEMA.`TABLES`
      WHERE TABLE_SCHEMA = 'hive.db_all';
         all   all   all   Accessibe tables + all views Accessibe tables + all views Accessibe tables + all views
      DESCRIBE hive.db_all.tbl_user0;                                           
      DESCRIBE hive.db_all.vw_user0;                                            
      DESCRIBE hive.db_all.tbl_group12;                                            
      DESCRIBE hive.db_all.vw_group12;                                           
      Queries executed using Beeline and Hive CLI:              
      SHOW TABLES IN db_all;    all    all    all           all           all           all
      SELECT * FROM hive.db_all.tbl_user0;                                            
      SELECT * FROM db_all.vw_user0;                                           
      SELECT * FROM db_all.tbl_group12;                                           
      SELECT * FROM db_all.vw_group12;                                           
                     
      DESCRIBE db_all.tbl_user0;                                            
      DESCRIBE db_all.vw_user0;                                           
      DESCRIBE db_all.tbl_group12;                                           
      DESCRIBE db_all.vw_group12;                                            

         Warning:  Because views in Hive aren't present as physical files and access can't be granted using file system commands, then access to Hive views for Storage Based Authorization is based on the underlying tables used in view definition. For current example views were defined as selection over appropriate tables. 

      For documentation
      Remove note from https://drill.apache.org/docs/querying-the-information-schema/ that states:

      NOTE
      
      Currently, Drill only supports querying Drill views; Hive views are not yet supported.
      

      Attachments

        Issue Links

          Activity

            People

              ihuzenko Igor Guzenko
              inramana Ramana Inukonda Nagaraj
              Vitalii Diravka Vitalii Diravka
              Votes:
              0 Vote for this issue
              Watchers:
              6 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: