use of io.trino.metadata.TableHandle in project trino by trinodb.
the class PushJoinIntoTableScan method apply.
@Override
public Result apply(JoinNode joinNode, Captures captures, Context context) {
if (joinNode.isCrossJoin()) {
return Result.empty();
}
TableScanNode left = captures.get(LEFT_TABLE_SCAN);
TableScanNode right = captures.get(RIGHT_TABLE_SCAN);
verify(!left.isUpdateTarget() && !right.isUpdateTarget(), "Unexpected Join over for-update table scan");
Expression effectiveFilter = getEffectiveFilter(joinNode);
FilterSplitResult filterSplitResult = splitFilter(effectiveFilter, left.getOutputSymbols(), right.getOutputSymbols(), context);
if (!filterSplitResult.getRemainingFilter().equals(BooleanLiteral.TRUE_LITERAL)) {
// TODO add extra filter node above join
return Result.empty();
}
if (left.getEnforcedConstraint().isNone() || right.getEnforcedConstraint().isNone()) {
// enforced constraint harder below.
return Result.empty();
}
Map<String, ColumnHandle> leftAssignments = left.getAssignments().entrySet().stream().collect(toImmutableMap(entry -> entry.getKey().getName(), Map.Entry::getValue));
Map<String, ColumnHandle> rightAssignments = right.getAssignments().entrySet().stream().collect(toImmutableMap(entry -> entry.getKey().getName(), Map.Entry::getValue));
/*
* We are (lazily) computing estimated statistics for join node and left and right table
* and passing those to connector via applyJoin.
*
* There are a couple reasons for this approach:
* - the engine knows how to estimate join and connector may not
* - the engine may have cached stats for the table scans (within context.getStatsProvider()), so can be able to provide information more inexpensively
* - in the future, the engine may be able to provide stats for table scan even in case when connector no longer can (see https://github.com/trinodb/trino/issues/6998)
* - the pushdown feasibility assessment logic may be different (or configured differently) for different connectors/catalogs.
*/
JoinStatistics joinStatistics = getJoinStatistics(joinNode, left, right, context);
Optional<JoinApplicationResult<TableHandle>> joinApplicationResult = metadata.applyJoin(context.getSession(), getJoinType(joinNode), left.getTable(), right.getTable(), filterSplitResult.getPushableConditions(), // TODO we could pass only subset of assignments here, those which are needed to resolve filterSplitResult.getPushableConditions
leftAssignments, rightAssignments, joinStatistics);
if (joinApplicationResult.isEmpty()) {
return Result.empty();
}
TableHandle handle = joinApplicationResult.get().getTableHandle();
Map<ColumnHandle, ColumnHandle> leftColumnHandlesMapping = joinApplicationResult.get().getLeftColumnHandles();
Map<ColumnHandle, ColumnHandle> rightColumnHandlesMapping = joinApplicationResult.get().getRightColumnHandles();
ImmutableMap.Builder<Symbol, ColumnHandle> assignmentsBuilder = ImmutableMap.builder();
assignmentsBuilder.putAll(left.getAssignments().entrySet().stream().collect(toImmutableMap(Map.Entry::getKey, entry -> leftColumnHandlesMapping.get(entry.getValue()))));
assignmentsBuilder.putAll(right.getAssignments().entrySet().stream().collect(toImmutableMap(Map.Entry::getKey, entry -> rightColumnHandlesMapping.get(entry.getValue()))));
Map<Symbol, ColumnHandle> assignments = assignmentsBuilder.buildOrThrow();
// convert enforced constraint
JoinNode.Type joinType = joinNode.getType();
TupleDomain<ColumnHandle> leftConstraint = deriveConstraint(left.getEnforcedConstraint(), leftColumnHandlesMapping, joinType == RIGHT || joinType == FULL);
TupleDomain<ColumnHandle> rightConstraint = deriveConstraint(right.getEnforcedConstraint(), rightColumnHandlesMapping, joinType == LEFT || joinType == FULL);
TupleDomain<ColumnHandle> newEnforcedConstraint = TupleDomain.withColumnDomains(ImmutableMap.<ColumnHandle, Domain>builder().putAll(leftConstraint.getDomains().orElseThrow()).putAll(rightConstraint.getDomains().orElseThrow()).buildOrThrow());
return Result.ofPlanNode(new ProjectNode(context.getIdAllocator().getNextId(), new TableScanNode(joinNode.getId(), handle, ImmutableList.copyOf(assignments.keySet()), assignments, newEnforcedConstraint, deriveTableStatisticsForPushdown(context.getStatsProvider(), context.getSession(), joinApplicationResult.get().isPrecalculateStatistics(), joinNode), false, Optional.empty()), Assignments.identity(joinNode.getOutputSymbols())));
}
use of io.trino.metadata.TableHandle in project trino by trinodb.
the class BaseHiveConnectorTest method getTableMetadata.
private TableMetadata getTableMetadata(String catalog, String schema, String tableName) {
Session session = getSession();
Metadata metadata = getDistributedQueryRunner().getCoordinator().getMetadata();
return transaction(getQueryRunner().getTransactionManager(), getQueryRunner().getAccessControl()).readOnly().execute(session, transactionSession -> {
Optional<TableHandle> tableHandle = metadata.getTableHandle(transactionSession, new QualifiedObjectName(catalog, schema, tableName));
assertTrue(tableHandle.isPresent());
return metadata.getTableMetadata(transactionSession, tableHandle.get());
});
}
use of io.trino.metadata.TableHandle in project trino by trinodb.
the class TestNodeLocalDynamicSplitPruning method createTestingPageSource.
private static ConnectorPageSource createTestingPageSource(HiveTransactionHandle transaction, HiveConfig hiveConfig, File outputFile, DynamicFilter dynamicFilter) {
Properties splitProperties = new Properties();
splitProperties.setProperty(FILE_INPUT_FORMAT, hiveConfig.getHiveStorageFormat().getInputFormat());
splitProperties.setProperty(SERIALIZATION_LIB, hiveConfig.getHiveStorageFormat().getSerde());
HiveSplit split = new HiveSplit(SCHEMA_NAME, TABLE_NAME, "", "file:///" + outputFile.getAbsolutePath(), 0, outputFile.length(), outputFile.length(), outputFile.lastModified(), splitProperties, ImmutableList.of(new HivePartitionKey(PARTITION_COLUMN.getName(), "42")), ImmutableList.of(), OptionalInt.of(1), 0, false, TableToPartitionMapping.empty(), Optional.empty(), Optional.empty(), false, Optional.empty(), 0, SplitWeight.standard());
TableHandle tableHandle = new TableHandle(new CatalogName(HIVE_CATALOG_NAME), new HiveTableHandle(SCHEMA_NAME, TABLE_NAME, ImmutableMap.of(), ImmutableList.of(), ImmutableList.of(BUCKET_HIVE_COLUMN_HANDLE), Optional.of(new HiveBucketHandle(ImmutableList.of(BUCKET_HIVE_COLUMN_HANDLE), BUCKETING_V1, 20, 20, ImmutableList.of()))), transaction);
HivePageSourceProvider provider = new HivePageSourceProvider(TESTING_TYPE_MANAGER, HDFS_ENVIRONMENT, hiveConfig, getDefaultHivePageSourceFactories(HDFS_ENVIRONMENT, hiveConfig), getDefaultHiveRecordCursorProviders(hiveConfig, HDFS_ENVIRONMENT), new GenericHiveRecordCursorProvider(HDFS_ENVIRONMENT, hiveConfig), Optional.empty());
return provider.createPageSource(transaction, getSession(hiveConfig), split, tableHandle.getConnectorHandle(), ImmutableList.of(BUCKET_HIVE_COLUMN_HANDLE, PARTITION_HIVE_COLUMN_HANDLE), dynamicFilter);
}
use of io.trino.metadata.TableHandle in project trino by trinodb.
the class ColumnReference method getAssignedSymbol.
@Override
public Optional<Symbol> getAssignedSymbol(PlanNode node, Session session, Metadata metadata, SymbolAliases symbolAliases) {
TableHandle tableHandle;
Map<Symbol, ColumnHandle> assignments;
if (node instanceof TableScanNode) {
TableScanNode tableScanNode = (TableScanNode) node;
tableHandle = tableScanNode.getTable();
assignments = tableScanNode.getAssignments();
} else if (node instanceof IndexSourceNode) {
IndexSourceNode indexSourceNode = (IndexSourceNode) node;
tableHandle = indexSourceNode.getTableHandle();
assignments = indexSourceNode.getAssignments();
} else {
return Optional.empty();
}
TableMetadata tableMetadata = metadata.getTableMetadata(session, tableHandle);
String actualTableName = tableMetadata.getTable().getTableName();
// Wrong table -> doesn't match.
if (!tableName.equalsIgnoreCase(actualTableName)) {
return Optional.empty();
}
Optional<ColumnHandle> columnHandle = getColumnHandle(tableHandle, session, metadata);
checkState(columnHandle.isPresent(), "Table %s doesn't have column %s. Typo in test?", tableName, columnName);
return getAssignedSymbol(assignments, columnHandle.get());
}
use of io.trino.metadata.TableHandle in project trino by trinodb.
the class TestPushProjectionIntoTableScan method createMockFactory.
private MockConnectorFactory createMockFactory(Map<String, ColumnHandle> assignments, Optional<MockConnectorFactory.ApplyProjection> applyProjection) {
List<ColumnMetadata> metadata = assignments.entrySet().stream().map(entry -> new ColumnMetadata(entry.getKey(), ((TpchColumnHandle) entry.getValue()).getType())).collect(toImmutableList());
MockConnectorFactory.Builder builder = MockConnectorFactory.builder().withListSchemaNames(connectorSession -> ImmutableList.of(TEST_SCHEMA)).withListTables((connectorSession, schema) -> TEST_SCHEMA.equals(schema) ? ImmutableList.of(TEST_SCHEMA_TABLE) : ImmutableList.of()).withGetColumns(schemaTableName -> metadata).withGetTableProperties((session, tableHandle) -> {
MockConnectorTableHandle mockTableHandle = (MockConnectorTableHandle) tableHandle;
if (mockTableHandle.getTableName().getTableName().equals(TEST_TABLE)) {
return new ConnectorTableProperties(TupleDomain.all(), Optional.of(new ConnectorTablePartitioning(PARTITIONING_HANDLE, ImmutableList.of(column("col", VARCHAR)))), Optional.empty(), Optional.empty(), ImmutableList.of());
}
return new ConnectorTableProperties();
});
if (applyProjection.isPresent()) {
builder = builder.withApplyProjection(applyProjection.get());
}
return builder.build();
}
Aggregations