Search in sources :

Example 11 with RelOptPlanner

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.plan.RelOptPlanner in project drill by apache.

the class DefaultSqlHandler method transform.

/**
 * Transform RelNode to a new RelNode, targeting the provided set of traits. Also will log the outcome if asked.
 *
 * @param plannerType The type of Planner to use.
 * @param phase The transformation phase we're running.
 * @param input The original RelNode
 * @param targetTraits The traits we are targeting for output.
 * @param log Whether to log the planning phase.
 * @return The transformed relnode.
 */
protected RelNode transform(PlannerType plannerType, PlannerPhase phase, RelNode input, RelTraitSet targetTraits, boolean log) {
    final Stopwatch watch = Stopwatch.createStarted();
    final RuleSet rules = config.getRules(phase);
    final RelTraitSet toTraits = targetTraits.simplify();
    final RelNode output;
    switch(plannerType) {
        case HEP_BOTTOM_UP:
        case HEP:
            {
                final HepProgramBuilder hepPgmBldr = new HepProgramBuilder();
                if (plannerType == PlannerType.HEP_BOTTOM_UP) {
                    hepPgmBldr.addMatchOrder(HepMatchOrder.BOTTOM_UP);
                }
                for (RelOptRule rule : rules) {
                    hepPgmBldr.addRuleInstance(rule);
                }
                // Set noDAG = true to avoid caching problems which lead to incorrect Drill work.
                final HepPlanner planner = new HepPlanner(hepPgmBldr.build(), context.getPlannerSettings(), true, null, RelOptCostImpl.FACTORY);
                JaninoRelMetadataProvider relMetadataProvider = Utilities.registerJaninoRelMetadataProvider();
                // Modify RelMetaProvider for every RelNode in the SQL operator Rel tree.
                input.accept(new MetaDataProviderModifier(relMetadataProvider));
                planner.setRoot(input);
                if (!input.getTraitSet().equals(targetTraits)) {
                    planner.changeTraits(input, toTraits);
                }
                output = planner.findBestExp();
                break;
            }
        case VOLCANO:
        default:
            {
                // as weird as it seems, the cluster's only planner is the volcano planner.
                final RelOptPlanner planner = input.getCluster().getPlanner();
                final Program program = Programs.of(rules);
                Preconditions.checkArgument(planner instanceof VolcanoPlanner, "Cluster is expected to be constructed using VolcanoPlanner. Was actually of type %s.", planner.getClass().getName());
                output = program.run(planner, input, toTraits, ImmutableList.of(), ImmutableList.of());
                break;
            }
    }
    if (log) {
        log(plannerType, phase, output, logger, watch);
    }
    return output;
}
Also used : RuleSet(org.apache.calcite.tools.RuleSet) Program(org.apache.calcite.tools.Program) RelNode(org.apache.calcite.rel.RelNode) Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch) HepProgramBuilder(org.apache.calcite.plan.hep.HepProgramBuilder) JaninoRelMetadataProvider(org.apache.calcite.rel.metadata.JaninoRelMetadataProvider) VolcanoPlanner(org.apache.calcite.plan.volcano.VolcanoPlanner) RelTraitSet(org.apache.calcite.plan.RelTraitSet) HepPlanner(org.apache.calcite.plan.hep.HepPlanner) RelOptPlanner(org.apache.calcite.plan.RelOptPlanner) RelOptRule(org.apache.calcite.plan.RelOptRule)

Example 12 with RelOptPlanner

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.plan.RelOptPlanner 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 13 with RelOptPlanner

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.plan.RelOptPlanner in project hive by apache.

the class CalcitePlanner method createPlanner.

private static RelOptPlanner createPlanner(HiveConf conf, Set<RelNode> corrScalarRexSQWithAgg, StatsSource statsSource, boolean isExplainPlan) {
    final Double maxSplitSize = (double) HiveConf.getLongVar(conf, HiveConf.ConfVars.MAPREDMAXSPLITSIZE);
    final Double maxMemory = (double) HiveConf.getLongVar(conf, HiveConf.ConfVars.HIVECONVERTJOINNOCONDITIONALTASKTHRESHOLD);
    HiveAlgorithmsConf algorithmsConf = new HiveAlgorithmsConf(maxSplitSize, maxMemory);
    HiveRulesRegistry registry = new HiveRulesRegistry();
    Properties calciteConfigProperties = new Properties();
    calciteConfigProperties.setProperty(CalciteConnectionProperty.TIME_ZONE.camelName(), conf.getLocalTimeZone().getId());
    calciteConfigProperties.setProperty(CalciteConnectionProperty.MATERIALIZATIONS_ENABLED.camelName(), Boolean.FALSE.toString());
    CalciteConnectionConfig calciteConfig = new CalciteConnectionConfigImpl(calciteConfigProperties);
    boolean isCorrelatedColumns = HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_CBO_STATS_CORRELATED_MULTI_KEY_JOINS);
    boolean heuristicMaterializationStrategy = HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_MATERIALIZED_VIEW_REWRITING_SELECTION_STRATEGY).equals("heuristic");
    HivePlannerContext confContext = new HivePlannerContext(algorithmsConf, registry, calciteConfig, corrScalarRexSQWithAgg, new HiveConfPlannerContext(isCorrelatedColumns, heuristicMaterializationStrategy, isExplainPlan), statsSource);
    RelOptPlanner planner = HiveVolcanoPlanner.createPlanner(confContext);
    planner.addListener(new RuleEventLogger());
    return planner;
}
Also used : CalciteConnectionConfigImpl(org.apache.calcite.config.CalciteConnectionConfigImpl) HivePlannerContext(org.apache.hadoop.hive.ql.optimizer.calcite.HivePlannerContext) HiveRulesRegistry(org.apache.hadoop.hive.ql.optimizer.calcite.rules.HiveRulesRegistry) CalciteConnectionConfig(org.apache.calcite.config.CalciteConnectionConfig) HiveConfPlannerContext(org.apache.hadoop.hive.ql.optimizer.calcite.HiveConfPlannerContext) QueryProperties(org.apache.hadoop.hive.ql.QueryProperties) Properties(java.util.Properties) RelOptPlanner(org.apache.calcite.plan.RelOptPlanner) HiveAlgorithmsConf(org.apache.hadoop.hive.ql.optimizer.calcite.cost.HiveAlgorithmsConf) RuleEventLogger(org.apache.hadoop.hive.ql.optimizer.calcite.RuleEventLogger)

Example 14 with RelOptPlanner

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.plan.RelOptPlanner in project hive by apache.

the class TestRexNodeConverter method beforeClass.

@BeforeClass
public static void beforeClass() {
    smallIntegerType = TYPE_FACTORY.createSqlType(SqlTypeName.SMALLINT);
    integerType = TYPE_FACTORY.createSqlType(SqlTypeName.INTEGER);
    nullableSmallIntegerType = TYPE_FACTORY.createTypeWithNullability(smallIntegerType, true);
    RelDataType varcharType = TYPE_FACTORY.createSqlType(SqlTypeName.VARCHAR, 20);
    varChar34 = REX_BUILDER.makeLiteral("34", varcharType, true);
    varChar35 = REX_BUILDER.makeLiteral("35", varcharType, true);
    varCharNull = REX_BUILDER.makeLiteral(null, varcharType, true);
    tableType = TYPE_FACTORY.createStructType(ImmutableList.of(smallIntegerType, nullableSmallIntegerType), ImmutableList.of("f1", "f2"));
    RelOptPlanner planner = CalcitePlanner.createPlanner(new HiveConf());
    relOptCluster = RelOptCluster.create(planner, REX_BUILDER);
}
Also used : RelDataType(org.apache.calcite.rel.type.RelDataType) HiveConf(org.apache.hadoop.hive.conf.HiveConf) RelOptPlanner(org.apache.calcite.plan.RelOptPlanner) BeforeClass(org.junit.BeforeClass)

Example 15 with RelOptPlanner

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.plan.RelOptPlanner 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

RelOptPlanner (org.apache.calcite.plan.RelOptPlanner)26 RelNode (org.apache.calcite.rel.RelNode)16 RelOptCluster (org.apache.calcite.plan.RelOptCluster)9 RelTraitSet (org.apache.calcite.plan.RelTraitSet)9 RexBuilder (org.apache.calcite.rex.RexBuilder)7 RelOptMaterialization (org.apache.calcite.plan.RelOptMaterialization)5 RelDataType (org.apache.calcite.rel.type.RelDataType)5 ImmutableList (com.google.common.collect.ImmutableList)4 ArrayList (java.util.ArrayList)4 List (java.util.List)4 RelOptRule (org.apache.calcite.plan.RelOptRule)4 HepPlanner (org.apache.calcite.plan.hep.HepPlanner)4 VolcanoPlanner (org.apache.calcite.plan.volcano.VolcanoPlanner)4 JaninoRelMetadataProvider (org.apache.calcite.rel.metadata.JaninoRelMetadataProvider)4 Properties (java.util.Properties)3 DataContext (org.apache.calcite.DataContext)3 JavaTypeFactory (org.apache.calcite.adapter.java.JavaTypeFactory)3 JavaTypeFactoryImpl (org.apache.calcite.jdbc.JavaTypeFactoryImpl)3 HepProgram (org.apache.calcite.plan.hep.HepProgram)3 HepProgramBuilder (org.apache.calcite.plan.hep.HepProgramBuilder)3