use of io.trino.spi.connector.Constraint in project trino by trinodb.
the class InformationSchemaMetadata method calculateRoles.
private Optional<Set<String>> calculateRoles(ConnectorSession connectorSession, TupleDomain<ColumnHandle> constraint, Optional<Predicate<Map<ColumnHandle, NullableValue>>> predicate) {
if (constraint.isNone()) {
return Optional.empty();
}
Optional<Set<String>> roles = filterString(constraint, ROLE_NAME_COLUMN_HANDLE);
if (roles.isPresent()) {
Set<String> result = roles.get().stream().filter(this::isLowerCase).filter(role -> predicate.isEmpty() || predicate.get().test(roleAsFixedValues(role))).collect(toImmutableSet());
if (result.isEmpty()) {
return Optional.empty();
}
if (result.size() <= MAX_ROLE_COUNT) {
return Optional.of(result);
}
}
if (predicate.isEmpty()) {
return Optional.empty();
}
Session session = ((FullConnectorSession) connectorSession).getSession();
return Optional.of(metadata.listRoles(session, Optional.of(catalogName)).stream().filter(role -> predicate.get().test(roleAsFixedValues(role))).collect(toImmutableSet()));
}
use of io.trino.spi.connector.Constraint in project trino by trinodb.
the class ColumnJdbcTable method applyFilter.
@Override
public TupleDomain<ColumnHandle> applyFilter(ConnectorSession connectorSession, Constraint constraint) {
TupleDomain<ColumnHandle> tupleDomain = constraint.getSummary();
if (tupleDomain.isNone() || constraint.predicate().isEmpty()) {
return tupleDomain;
}
Predicate<Map<ColumnHandle, NullableValue>> predicate = constraint.predicate().get();
Set<ColumnHandle> predicateColumns = constraint.getPredicateColumns().orElseThrow(() -> new VerifyException("columns not present for a predicate"));
boolean hasSchemaPredicate = predicateColumns.contains(TABLE_SCHEMA_COLUMN);
boolean hasTablePredicate = predicateColumns.contains(TABLE_NAME_COLUMN);
if (!hasSchemaPredicate && !hasTablePredicate) {
// No filter on schema name and table name at all.
return tupleDomain;
}
Session session = ((FullConnectorSession) connectorSession).getSession();
Optional<String> catalogFilter = tryGetSingleVarcharValue(tupleDomain, TABLE_CATALOG_COLUMN);
Optional<String> schemaFilter = tryGetSingleVarcharValue(tupleDomain, TABLE_SCHEMA_COLUMN);
Optional<String> tableFilter = tryGetSingleVarcharValue(tupleDomain, TABLE_NAME_COLUMN);
if (schemaFilter.isPresent() && tableFilter.isPresent()) {
// No need to narrow down the domain.
return tupleDomain;
}
List<String> catalogs = listCatalogs(session, metadata, accessControl, catalogFilter).keySet().stream().filter(catalogName -> predicate.test(ImmutableMap.of(TABLE_CATALOG_COLUMN, toNullableValue(catalogName)))).collect(toImmutableList());
List<CatalogSchemaName> schemas = catalogs.stream().flatMap(catalogName -> listSchemas(session, metadata, accessControl, catalogName, schemaFilter).stream().filter(schemaName -> !hasSchemaPredicate || predicate.test(ImmutableMap.of(TABLE_CATALOG_COLUMN, toNullableValue(catalogName), TABLE_SCHEMA_COLUMN, toNullableValue(schemaName)))).map(schemaName -> new CatalogSchemaName(catalogName, schemaName))).collect(toImmutableList());
if (!hasTablePredicate) {
return TupleDomain.withColumnDomains(ImmutableMap.<ColumnHandle, Domain>builder().put(TABLE_CATALOG_COLUMN, schemas.stream().map(CatalogSchemaName::getCatalogName).collect(toVarcharDomain()).simplify(MAX_DOMAIN_SIZE)).put(TABLE_SCHEMA_COLUMN, schemas.stream().map(CatalogSchemaName::getSchemaName).collect(toVarcharDomain()).simplify(MAX_DOMAIN_SIZE)).buildOrThrow());
}
List<CatalogSchemaTableName> tables = schemas.stream().flatMap(schema -> {
QualifiedTablePrefix tablePrefix = tableFilter.isPresent() ? new QualifiedTablePrefix(schema.getCatalogName(), schema.getSchemaName(), tableFilter.get()) : new QualifiedTablePrefix(schema.getCatalogName(), schema.getSchemaName());
return listTables(session, metadata, accessControl, tablePrefix).stream().filter(schemaTableName -> predicate.test(ImmutableMap.of(TABLE_CATALOG_COLUMN, toNullableValue(schema.getCatalogName()), TABLE_SCHEMA_COLUMN, toNullableValue(schemaTableName.getSchemaName()), TABLE_NAME_COLUMN, toNullableValue(schemaTableName.getTableName())))).map(schemaTableName -> new CatalogSchemaTableName(schema.getCatalogName(), schemaTableName.getSchemaName(), schemaTableName.getTableName()));
}).collect(toImmutableList());
return TupleDomain.withColumnDomains(ImmutableMap.<ColumnHandle, Domain>builder().put(TABLE_CATALOG_COLUMN, tables.stream().map(CatalogSchemaTableName::getCatalogName).collect(toVarcharDomain()).simplify(MAX_DOMAIN_SIZE)).put(TABLE_SCHEMA_COLUMN, tables.stream().map(catalogSchemaTableName -> catalogSchemaTableName.getSchemaTableName().getSchemaName()).collect(toVarcharDomain()).simplify(MAX_DOMAIN_SIZE)).put(TABLE_NAME_COLUMN, tables.stream().map(catalogSchemaTableName -> catalogSchemaTableName.getSchemaTableName().getTableName()).collect(toVarcharDomain()).simplify(MAX_DOMAIN_SIZE)).buildOrThrow());
}
use of io.trino.spi.connector.Constraint in project trino by trinodb.
the class MetadataManager method applyFilter.
@Override
public Optional<ConstraintApplicationResult<TableHandle>> applyFilter(Session session, TableHandle table, Constraint constraint) {
CatalogName catalogName = table.getCatalogName();
ConnectorMetadata metadata = getMetadata(session, catalogName);
ConnectorSession connectorSession = session.toConnectorSession(catalogName);
return metadata.applyFilter(connectorSession, table.getConnectorHandle(), constraint).map(result -> result.transform(handle -> new TableHandle(catalogName, handle, table.getTransaction())));
}
use of io.trino.spi.connector.Constraint in project trino by trinodb.
the class PushPredicateIntoTableScan method pushFilterIntoTableScan.
public static Optional<PlanNode> pushFilterIntoTableScan(FilterNode filterNode, TableScanNode node, boolean pruneWithPredicateExpression, Session session, SymbolAllocator symbolAllocator, PlannerContext plannerContext, TypeAnalyzer typeAnalyzer, StatsProvider statsProvider, DomainTranslator domainTranslator) {
if (!isAllowPushdownIntoConnectors(session)) {
return Optional.empty();
}
SplitExpression splitExpression = splitExpression(plannerContext, filterNode.getPredicate());
DomainTranslator.ExtractionResult decomposedPredicate = DomainTranslator.getExtractionResult(plannerContext, session, splitExpression.getDeterministicPredicate(), symbolAllocator.getTypes());
TupleDomain<ColumnHandle> newDomain = decomposedPredicate.getTupleDomain().transformKeys(node.getAssignments()::get).intersect(node.getEnforcedConstraint());
Map<NodeRef<Expression>, Type> remainingExpressionTypes = typeAnalyzer.getTypes(session, symbolAllocator.getTypes(), decomposedPredicate.getRemainingExpression());
Optional<ConnectorExpression> connectorExpression = new ConnectorExpressionTranslator.SqlToConnectorExpressionTranslator(session, remainingExpressionTypes, plannerContext).process(decomposedPredicate.getRemainingExpression());
Map<String, ColumnHandle> connectorExpressionAssignments = connectorExpression.map(ignored -> node.getAssignments().entrySet().stream().collect(toImmutableMap(entry -> entry.getKey().getName(), Map.Entry::getValue))).orElse(ImmutableMap.of());
Map<ColumnHandle, Symbol> assignments = ImmutableBiMap.copyOf(node.getAssignments()).inverse();
Constraint constraint;
// use evaluator only when there is some predicate which could not be translated into tuple domain
if (pruneWithPredicateExpression && !TRUE_LITERAL.equals(decomposedPredicate.getRemainingExpression())) {
LayoutConstraintEvaluator evaluator = new LayoutConstraintEvaluator(plannerContext, typeAnalyzer, session, symbolAllocator.getTypes(), node.getAssignments(), combineConjuncts(plannerContext.getMetadata(), splitExpression.getDeterministicPredicate(), // which would be expensive to evaluate in the call to isCandidate below.
domainTranslator.toPredicate(session, newDomain.simplify().transformKeys(assignments::get))));
constraint = new Constraint(newDomain, connectorExpression.orElse(TRUE), connectorExpressionAssignments, evaluator::isCandidate, evaluator.getArguments());
} else {
// Currently, invoking the expression interpreter is very expensive.
// TODO invoke the interpreter unconditionally when the interpreter becomes cheap enough.
constraint = new Constraint(newDomain, connectorExpression.orElse(TRUE), connectorExpressionAssignments);
}
// check if new domain is wider than domain already provided by table scan
if (constraint.predicate().isEmpty() && // TODO do we need to track enforced ConnectorExpression in TableScanNode?
TRUE.equals(connectorExpression.orElse(TRUE)) && newDomain.contains(node.getEnforcedConstraint())) {
Expression resultingPredicate = createResultingPredicate(plannerContext, session, symbolAllocator, typeAnalyzer, splitExpression.getDynamicFilter(), TRUE_LITERAL, splitExpression.getNonDeterministicPredicate(), decomposedPredicate.getRemainingExpression());
if (!TRUE_LITERAL.equals(resultingPredicate)) {
return Optional.of(new FilterNode(filterNode.getId(), node, resultingPredicate));
}
return Optional.of(node);
}
if (newDomain.isNone()) {
// to turn the subtree into a Values node
return Optional.of(new ValuesNode(node.getId(), node.getOutputSymbols(), ImmutableList.of()));
}
Optional<ConstraintApplicationResult<TableHandle>> result = plannerContext.getMetadata().applyFilter(session, node.getTable(), constraint);
if (result.isEmpty()) {
return Optional.empty();
}
TableHandle newTable = result.get().getHandle();
TableProperties newTableProperties = plannerContext.getMetadata().getTableProperties(session, newTable);
Optional<TablePartitioning> newTablePartitioning = newTableProperties.getTablePartitioning();
if (newTableProperties.getPredicate().isNone()) {
return Optional.of(new ValuesNode(node.getId(), node.getOutputSymbols(), ImmutableList.of()));
}
TupleDomain<ColumnHandle> remainingFilter = result.get().getRemainingFilter();
Optional<ConnectorExpression> remainingConnectorExpression = result.get().getRemainingExpression();
boolean precalculateStatistics = result.get().isPrecalculateStatistics();
verifyTablePartitioning(session, plannerContext.getMetadata(), node, newTablePartitioning);
TableScanNode tableScan = new TableScanNode(node.getId(), newTable, node.getOutputSymbols(), node.getAssignments(), computeEnforced(newDomain, remainingFilter), // TODO (https://github.com/trinodb/trino/issues/8144) distinguish between predicate pushed down and remaining
deriveTableStatisticsForPushdown(statsProvider, session, precalculateStatistics, filterNode), node.isUpdateTarget(), node.getUseConnectorNodePartitioning());
Expression remainingDecomposedPredicate;
if (remainingConnectorExpression.isEmpty() || remainingConnectorExpression.equals(connectorExpression)) {
remainingDecomposedPredicate = decomposedPredicate.getRemainingExpression();
} else {
Map<String, Symbol> variableMappings = assignments.values().stream().collect(toImmutableMap(Symbol::getName, Function.identity()));
Expression translatedExpression = ConnectorExpressionTranslator.translate(session, remainingConnectorExpression.get(), plannerContext, variableMappings, new LiteralEncoder(plannerContext));
if (connectorExpression.isEmpty()) {
remainingDecomposedPredicate = ExpressionUtils.combineConjuncts(plannerContext.getMetadata(), translatedExpression, decomposedPredicate.getRemainingExpression());
} else {
remainingDecomposedPredicate = translatedExpression;
}
}
Expression resultingPredicate = createResultingPredicate(plannerContext, session, symbolAllocator, typeAnalyzer, splitExpression.getDynamicFilter(), domainTranslator.toPredicate(session, remainingFilter.transformKeys(assignments::get)), splitExpression.getNonDeterministicPredicate(), remainingDecomposedPredicate);
if (!TRUE_LITERAL.equals(resultingPredicate)) {
return Optional.of(new FilterNode(filterNode.getId(), tableScan, resultingPredicate));
}
return Optional.of(tableScan);
}
use of io.trino.spi.connector.Constraint in project trino by trinodb.
the class HiveMetadata method getTableProperties.
@Override
public ConnectorTableProperties getTableProperties(ConnectorSession session, ConnectorTableHandle table) {
HiveTableHandle hiveTable = (HiveTableHandle) table;
List<ColumnHandle> partitionColumns = ImmutableList.copyOf(hiveTable.getPartitionColumns());
TupleDomain<ColumnHandle> predicate = TupleDomain.all();
Optional<DiscretePredicates> discretePredicates = Optional.empty();
// So computation of predicate and discretePredicates are not valid.
if (hiveTable.getPartitionNames().isEmpty()) {
Optional<List<HivePartition>> partitions = hiveTable.getPartitions().or(() -> {
// We load the partitions to compute the predicates enforced by the table.
// Note that the computation is not persisted in the table handle, so can be redone many times
// TODO: https://github.com/trinodb/trino/issues/10980.
HivePartitionResult partitionResult = partitionManager.getPartitions(metastore, table, new Constraint(hiveTable.getEnforcedConstraint()));
if (partitionManager.canPartitionsBeLoaded(partitionResult)) {
return Optional.of(partitionManager.getPartitionsAsList(partitionResult));
}
return Optional.empty();
});
if (partitions.isPresent()) {
List<HivePartition> hivePartitions = partitions.orElseThrow();
// Since the partitions are fully loaded now, we need to compute
predicate = createPredicate(partitionColumns, hivePartitions);
// this check allows us to ensure that table is partitioned
if (!partitionColumns.isEmpty()) {
// Do not create tuple domains for every partition at the same time!
// There can be a huge number of partitions so use an iterable so
// all domains do not need to be in memory at the same time.
Iterable<TupleDomain<ColumnHandle>> partitionDomains = Iterables.transform(hivePartitions, hivePartition -> TupleDomain.fromFixedValues(hivePartition.getKeys()));
discretePredicates = Optional.of(new DiscretePredicates(partitionColumns, partitionDomains));
}
}
}
Optional<ConnectorTablePartitioning> tablePartitioning = Optional.empty();
List<LocalProperty<ColumnHandle>> sortingProperties = ImmutableList.of();
if (hiveTable.getBucketHandle().isPresent()) {
if (isPropagateTableScanSortingProperties(session) && !hiveTable.getBucketHandle().get().getSortedBy().isEmpty()) {
// Populating SortingProperty guarantees to the engine that it is reading pre-sorted input.
// We detect compatibility between table and partition level sorted_by properties
// and fail the query if there is a mismatch in HiveSplitManager#getPartitionMetadata.
// This can lead to incorrect results if a sorted_by property is defined over unsorted files.
Map<String, ColumnHandle> columnHandles = getColumnHandles(session, table);
sortingProperties = hiveTable.getBucketHandle().get().getSortedBy().stream().map(sortingColumn -> new SortingProperty<>(columnHandles.get(sortingColumn.getColumnName()), sortingColumn.getOrder().getSortOrder())).collect(toImmutableList());
}
if (isBucketExecutionEnabled(session)) {
tablePartitioning = hiveTable.getBucketHandle().map(bucketing -> new ConnectorTablePartitioning(new HivePartitioningHandle(bucketing.getBucketingVersion(), bucketing.getReadBucketCount(), bucketing.getColumns().stream().map(HiveColumnHandle::getHiveType).collect(toImmutableList()), OptionalInt.empty(), false), bucketing.getColumns().stream().map(ColumnHandle.class::cast).collect(toImmutableList())));
}
}
return new ConnectorTableProperties(predicate, tablePartitioning, Optional.empty(), discretePredicates, sortingProperties);
}
Aggregations