use of org.apache.hadoop.hive.ql.parse.QueryTables 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;
}
Aggregations