Search in sources :

Example 1 with DruidTable

use of org.apache.calcite.adapter.druid.DruidTable in project hive by apache.

the class HiveMaterializedViewsRegistry method createTableScan.

private static RelNode createTableScan(Table viewTable) {
    // 0. Recreate cluster
    final RelOptPlanner planner = HiveVolcanoPlanner.createPlanner(null);
    final RexBuilder rexBuilder = new RexBuilder(new JavaTypeFactoryImpl());
    final RelOptCluster cluster = RelOptCluster.create(planner, rexBuilder);
    // 1. Create column schema
    final RowResolver rr = new RowResolver();
    // 1.1 Add Column info for non partion cols (Object Inspector fields)
    StructObjectInspector rowObjectInspector;
    try {
        rowObjectInspector = (StructObjectInspector) viewTable.getDeserializer().getObjectInspector();
    } catch (SerDeException e) {
        // Bail out
        return null;
    }
    List<? extends StructField> fields = rowObjectInspector.getAllStructFieldRefs();
    ColumnInfo colInfo;
    String colName;
    ArrayList<ColumnInfo> cInfoLst = new ArrayList<ColumnInfo>();
    for (int i = 0; i < fields.size(); i++) {
        colName = fields.get(i).getFieldName();
        colInfo = new ColumnInfo(fields.get(i).getFieldName(), TypeInfoUtils.getTypeInfoFromObjectInspector(fields.get(i).getFieldObjectInspector()), null, false);
        rr.put(null, colName, colInfo);
        cInfoLst.add(colInfo);
    }
    ArrayList<ColumnInfo> nonPartitionColumns = new ArrayList<ColumnInfo>(cInfoLst);
    // 1.2 Add column info corresponding to partition columns
    ArrayList<ColumnInfo> partitionColumns = new ArrayList<ColumnInfo>();
    for (FieldSchema part_col : viewTable.getPartCols()) {
        colName = part_col.getName();
        colInfo = new ColumnInfo(colName, TypeInfoFactory.getPrimitiveTypeInfo(part_col.getType()), null, true);
        rr.put(null, colName, colInfo);
        cInfoLst.add(colInfo);
        partitionColumns.add(colInfo);
    }
    // 1.3 Build row type from field <type, name>
    RelDataType rowType;
    try {
        rowType = TypeConverter.getType(cluster, rr, null);
    } catch (CalciteSemanticException e) {
        // Bail out
        return null;
    }
    // 2. Build RelOptAbstractTable
    String fullyQualifiedTabName = viewTable.getDbName();
    if (fullyQualifiedTabName != null && !fullyQualifiedTabName.isEmpty()) {
        fullyQualifiedTabName = fullyQualifiedTabName + "." + viewTable.getTableName();
    } else {
        fullyQualifiedTabName = viewTable.getTableName();
    }
    RelOptHiveTable optTable = new RelOptHiveTable(null, fullyQualifiedTabName, rowType, viewTable, nonPartitionColumns, partitionColumns, new ArrayList<VirtualColumn>(), SessionState.get().getConf(), new HashMap<String, PrunedPartitionList>(), new AtomicInteger());
    RelNode tableRel;
    // 3. Build operator
    if (obtainTableType(viewTable) == TableType.DRUID) {
        // Build Druid query
        String address = HiveConf.getVar(SessionState.get().getConf(), HiveConf.ConfVars.HIVE_DRUID_BROKER_DEFAULT_ADDRESS);
        String dataSource = viewTable.getParameters().get(Constants.DRUID_DATA_SOURCE);
        Set<String> metrics = new HashSet<>();
        List<RelDataType> druidColTypes = new ArrayList<>();
        List<String> druidColNames = new ArrayList<>();
        for (RelDataTypeField field : rowType.getFieldList()) {
            druidColTypes.add(field.getType());
            druidColNames.add(field.getName());
            if (field.getName().equals(DruidTable.DEFAULT_TIMESTAMP_COLUMN)) {
                // timestamp
                continue;
            }
            if (field.getType().getSqlTypeName() == SqlTypeName.VARCHAR) {
                // dimension
                continue;
            }
            metrics.add(field.getName());
        }
        List<Interval> intervals = Arrays.asList(DruidTable.DEFAULT_INTERVAL);
        DruidTable druidTable = new DruidTable(new DruidSchema(address, address, false), dataSource, RelDataTypeImpl.proto(rowType), metrics, DruidTable.DEFAULT_TIMESTAMP_COLUMN, intervals);
        final TableScan scan = new HiveTableScan(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), optTable, viewTable.getTableName(), null, false, false);
        tableRel = DruidQuery.create(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), optTable, druidTable, ImmutableList.<RelNode>of(scan));
    } else {
        // Build Hive Table Scan Rel
        tableRel = new HiveTableScan(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), optTable, viewTable.getTableName(), null, false, false);
    }
    return tableRel;
}
Also used : RelOptCluster(org.apache.calcite.plan.RelOptCluster) FieldSchema(org.apache.hadoop.hive.metastore.api.FieldSchema) ArrayList(java.util.ArrayList) ColumnInfo(org.apache.hadoop.hive.ql.exec.ColumnInfo) DruidTable(org.apache.calcite.adapter.druid.DruidTable) RelDataType(org.apache.calcite.rel.type.RelDataType) RowResolver(org.apache.hadoop.hive.ql.parse.RowResolver) RelOptPlanner(org.apache.calcite.plan.RelOptPlanner) PrunedPartitionList(org.apache.hadoop.hive.ql.parse.PrunedPartitionList) JavaTypeFactoryImpl(org.apache.calcite.jdbc.JavaTypeFactoryImpl) RexBuilder(org.apache.calcite.rex.RexBuilder) CalciteSemanticException(org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException) SerDeException(org.apache.hadoop.hive.serde2.SerDeException) HashSet(java.util.HashSet) HiveTableScan(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveTableScan) TableScan(org.apache.calcite.rel.core.TableScan) DruidSchema(org.apache.calcite.adapter.druid.DruidSchema) RelDataTypeField(org.apache.calcite.rel.type.RelDataTypeField) RelOptHiveTable(org.apache.hadoop.hive.ql.optimizer.calcite.RelOptHiveTable) HiveRelNode(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveRelNode) RelNode(org.apache.calcite.rel.RelNode) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) HiveTableScan(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveTableScan) StructObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector) Interval(org.joda.time.Interval)

Example 2 with DruidTable

use of org.apache.calcite.adapter.druid.DruidTable in project hive by apache.

the class HiveMaterializedViewsRegistry method createMaterializedViewScan.

private static RelNode createMaterializedViewScan(HiveConf conf, Table viewTable) {
    // 0. Recreate cluster
    final RelOptPlanner planner = CalcitePlanner.createPlanner(conf);
    final RexBuilder rexBuilder = new RexBuilder(new JavaTypeFactoryImpl(new HiveTypeSystemImpl()));
    final RelOptCluster cluster = RelOptCluster.create(planner, rexBuilder);
    // 1. Create column schema
    final RowResolver rr = new RowResolver();
    // 1.1 Add Column info for non partion cols (Object Inspector fields)
    StructObjectInspector rowObjectInspector;
    try {
        rowObjectInspector = (StructObjectInspector) viewTable.getDeserializer().getObjectInspector();
    } catch (SerDeException e) {
        // Bail out
        return null;
    }
    List<? extends StructField> fields = rowObjectInspector.getAllStructFieldRefs();
    ColumnInfo colInfo;
    String colName;
    ArrayList<ColumnInfo> cInfoLst = new ArrayList<>();
    for (StructField structField : fields) {
        colName = structField.getFieldName();
        colInfo = new ColumnInfo(structField.getFieldName(), TypeInfoUtils.getTypeInfoFromObjectInspector(structField.getFieldObjectInspector()), null, false);
        rr.put(null, colName, colInfo);
        cInfoLst.add(colInfo);
    }
    ArrayList<ColumnInfo> nonPartitionColumns = new ArrayList<ColumnInfo>(cInfoLst);
    // 1.2 Add column info corresponding to partition columns
    ArrayList<ColumnInfo> partitionColumns = new ArrayList<ColumnInfo>();
    for (FieldSchema part_col : viewTable.getPartCols()) {
        colName = part_col.getName();
        colInfo = new ColumnInfo(colName, TypeInfoFactory.getPrimitiveTypeInfo(part_col.getType()), null, true);
        rr.put(null, colName, colInfo);
        cInfoLst.add(colInfo);
        partitionColumns.add(colInfo);
    }
    // 1.3 Build row type from field <type, name>
    RelDataType rowType;
    try {
        rowType = TypeConverter.getType(cluster, rr, null);
    } catch (CalciteSemanticException e) {
        // Bail out
        return null;
    }
    // 2. Build RelOptAbstractTable
    List<String> fullyQualifiedTabName = new ArrayList<>();
    if (viewTable.getDbName() != null && !viewTable.getDbName().isEmpty()) {
        fullyQualifiedTabName.add(viewTable.getDbName());
    }
    fullyQualifiedTabName.add(viewTable.getTableName());
    RelNode tableRel;
    // 3. Build operator
    if (obtainTableType(viewTable) == TableType.DRUID) {
        // Build Druid query
        String address = HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_DRUID_BROKER_DEFAULT_ADDRESS);
        String dataSource = viewTable.getParameters().get(Constants.DRUID_DATA_SOURCE);
        Set<String> metrics = new HashSet<>();
        List<RelDataType> druidColTypes = new ArrayList<>();
        List<String> druidColNames = new ArrayList<>();
        // @NOTE this code is very similar to the code at org/apache/hadoop/hive/ql/parse/CalcitePlanner.java:2362
        // @TODO it will be nice to refactor it
        RelDataTypeFactory dtFactory = cluster.getRexBuilder().getTypeFactory();
        for (RelDataTypeField field : rowType.getFieldList()) {
            if (DruidTable.DEFAULT_TIMESTAMP_COLUMN.equals(field.getName())) {
                // Druid's time column is always not null.
                druidColTypes.add(dtFactory.createTypeWithNullability(field.getType(), false));
            } else {
                druidColTypes.add(field.getType());
            }
            druidColNames.add(field.getName());
            if (field.getName().equals(DruidTable.DEFAULT_TIMESTAMP_COLUMN)) {
                // timestamp
                continue;
            }
            if (field.getType().getSqlTypeName() == SqlTypeName.VARCHAR) {
                // dimension
                continue;
            }
            metrics.add(field.getName());
        }
        List<Interval> intervals = Collections.singletonList(DruidTable.DEFAULT_INTERVAL);
        rowType = dtFactory.createStructType(druidColTypes, druidColNames);
        // We can pass null for Hive object because it is only used to retrieve tables
        // if constraints on a table object are existing, but constraints cannot be defined
        // for materialized views.
        RelOptHiveTable optTable = new RelOptHiveTable(null, cluster.getTypeFactory(), fullyQualifiedTabName, rowType, viewTable, nonPartitionColumns, partitionColumns, new ArrayList<>(), conf, null, new QueryTables(true), new HashMap<>(), new HashMap<>(), new AtomicInteger());
        DruidTable druidTable = new DruidTable(new DruidSchema(address, address, false), dataSource, RelDataTypeImpl.proto(rowType), metrics, DruidTable.DEFAULT_TIMESTAMP_COLUMN, intervals, null, null);
        final TableScan scan = new HiveTableScan(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), optTable, viewTable.getTableName(), null, false, false);
        tableRel = DruidQuery.create(cluster, cluster.traitSetOf(BindableConvention.INSTANCE), optTable, druidTable, ImmutableList.<RelNode>of(scan), ImmutableMap.of());
    } else {
        // Build Hive Table Scan Rel.
        // We can pass null for Hive object because it is only used to retrieve tables
        // if constraints on a table object are existing, but constraints cannot be defined
        // for materialized views.
        RelOptHiveTable optTable = new RelOptHiveTable(null, cluster.getTypeFactory(), fullyQualifiedTabName, rowType, viewTable, nonPartitionColumns, partitionColumns, new ArrayList<>(), conf, null, new QueryTables(true), new HashMap<>(), new HashMap<>(), new AtomicInteger());
        tableRel = new HiveTableScan(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), optTable, viewTable.getTableName(), null, false, false);
    }
    return tableRel;
}
Also used : RelOptCluster(org.apache.calcite.plan.RelOptCluster) FieldSchema(org.apache.hadoop.hive.metastore.api.FieldSchema) ArrayList(java.util.ArrayList) ColumnInfo(org.apache.hadoop.hive.ql.exec.ColumnInfo) DruidTable(org.apache.calcite.adapter.druid.DruidTable) RelDataType(org.apache.calcite.rel.type.RelDataType) RowResolver(org.apache.hadoop.hive.ql.parse.RowResolver) QueryTables(org.apache.hadoop.hive.ql.parse.QueryTables) RelOptPlanner(org.apache.calcite.plan.RelOptPlanner) StructField(org.apache.hadoop.hive.serde2.objectinspector.StructField) JavaTypeFactoryImpl(org.apache.calcite.jdbc.JavaTypeFactoryImpl) RelDataTypeFactory(org.apache.calcite.rel.type.RelDataTypeFactory) RexBuilder(org.apache.calcite.rex.RexBuilder) CalciteSemanticException(org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException) SerDeException(org.apache.hadoop.hive.serde2.SerDeException) HashSet(java.util.HashSet) HiveTableScan(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveTableScan) TableScan(org.apache.calcite.rel.core.TableScan) DruidSchema(org.apache.calcite.adapter.druid.DruidSchema) RelDataTypeField(org.apache.calcite.rel.type.RelDataTypeField) RelOptHiveTable(org.apache.hadoop.hive.ql.optimizer.calcite.RelOptHiveTable) HiveRelNode(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveRelNode) RelNode(org.apache.calcite.rel.RelNode) AtomicInteger(java.util.concurrent.atomic.AtomicInteger) HiveTableScan(org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveTableScan) HiveTypeSystemImpl(org.apache.hadoop.hive.ql.optimizer.calcite.HiveTypeSystemImpl) StructObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector) Interval(org.joda.time.Interval)

Aggregations

ArrayList (java.util.ArrayList)2 HashSet (java.util.HashSet)2 AtomicInteger (java.util.concurrent.atomic.AtomicInteger)2 DruidSchema (org.apache.calcite.adapter.druid.DruidSchema)2 DruidTable (org.apache.calcite.adapter.druid.DruidTable)2 JavaTypeFactoryImpl (org.apache.calcite.jdbc.JavaTypeFactoryImpl)2 RelOptCluster (org.apache.calcite.plan.RelOptCluster)2 RelOptPlanner (org.apache.calcite.plan.RelOptPlanner)2 RelNode (org.apache.calcite.rel.RelNode)2 TableScan (org.apache.calcite.rel.core.TableScan)2 RelDataType (org.apache.calcite.rel.type.RelDataType)2 RelDataTypeField (org.apache.calcite.rel.type.RelDataTypeField)2 RexBuilder (org.apache.calcite.rex.RexBuilder)2 FieldSchema (org.apache.hadoop.hive.metastore.api.FieldSchema)2 ColumnInfo (org.apache.hadoop.hive.ql.exec.ColumnInfo)2 CalciteSemanticException (org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException)2 RelOptHiveTable (org.apache.hadoop.hive.ql.optimizer.calcite.RelOptHiveTable)2 HiveRelNode (org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveRelNode)2 HiveTableScan (org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveTableScan)2 RowResolver (org.apache.hadoop.hive.ql.parse.RowResolver)2