Examples with SchemaColumn org.voltdb.plannodes.SchemaColumn used on opensource projects

Example 1 with SchemaColumn

use of org.voltdb.plannodes.SchemaColumn in project voltdb by VoltDB.

the class TupleValueExpression method setColumnIndexUsingSchema.

/**
     * Given an input schema, resolve this TVE expression.
     */
public int setColumnIndexUsingSchema(NodeSchema inputSchema) {
    int index = inputSchema.getIndexOfTve(this);
    if (index < 0) {
        //* enable to debug*/ System.out.println("DEBUG: setColumnIndex candidates: " + inputSchema);
        return index;
    }
    setColumnIndex(index);
    if (getValueType() == null) {
        // In case of sub-queries the TVE may not have its
        // value type and size resolved yet. Try to resolve it now
        SchemaColumn inputColumn = inputSchema.getColumns().get(index);
        setTypeSizeAndInBytes(inputColumn);
    }
    return index;
}

Example 2 with SchemaColumn

use of org.voltdb.plannodes.SchemaColumn in project voltdb by VoltDB.

the class InsertSubPlanAssembler method nextPlan.

@Override
AbstractPlanNode nextPlan() {
    if (m_bestAndOnlyPlanWasGenerated) {
        return null;
    }
    // We may generate a few different plans for the subquery, but by the time
    // we get here, we'll generate only one plan for the INSERT statement itself.
    // Mostly this method exists to check that we can find a valid partitioning
    // for the statement.
    m_bestAndOnlyPlanWasGenerated = true;
    ParsedInsertStmt insertStmt = (ParsedInsertStmt) m_parsedStmt;
    Table targetTable = insertStmt.m_tableList.get(0);
    targetTable.getTypeName();
    StmtSubqueryScan subquery = insertStmt.getSubqueryScan();
    boolean subqueryIsMultiFragment = subquery.getBestCostPlan().rootPlanGraph.hasAnyNodeOfType(PlanNodeType.SEND);
    if (targetTable.getIsreplicated()) {
        // setUpForNewPlans already validates this
        assert (!m_partitioning.wasSpecifiedAsSingle() && !m_partitioning.isInferredSingle());
        // Cannot access any partitioned tables in subquery for replicated table
        if (!subquery.getIsReplicated()) {
            throw new PlanningErrorException("Subquery in " + getSqlType() + " INTO ... SELECT statement may not access " + "partitioned data for insertion into replicated table " + targetTable.getTypeName() + ".");
        }
    } else if (!m_partitioning.wasSpecifiedAsSingle()) {
        if (subqueryIsMultiFragment) {
            // What is the appropriate level of detail for this message?
            m_recentErrorMsg = getSqlType() + " INTO ... SELECT statement subquery is too complex.  " + "Please either simplify the subquery or use a SELECT followed by an INSERT.";
            return null;
        }
        Column partitioningCol = targetTable.getPartitioncolumn();
        if (partitioningCol == null) {
            assert (m_targetIsExportTable);
            m_recentErrorMsg = "The target table for an INSERT INTO ... SELECT statement is an " + "stream with no partitioning column defined.  " + "This is not currently supported.  Please define a " + "partitioning column for this stream to use it with INSERT INTO ... SELECT.";
            return null;
        }
        List<StmtTableScan> tables = new ArrayList<>();
        StmtTargetTableScan stmtTargetTableScan = new StmtTargetTableScan(targetTable);
        tables.add(stmtTargetTableScan);
        tables.add(subquery);
        // Create value equivalence between the partitioning column of the target table
        // and the corresponding expression produced by the subquery.
        HashMap<AbstractExpression, Set<AbstractExpression>> valueEquivalence = new HashMap<>();
        int i = 0;
        boolean setEquivalenceForPartitioningCol = false;
        for (Column col : insertStmt.m_columns.keySet()) {
            if (partitioningCol.compareTo(col) == 0) {
                List<SchemaColumn> partitioningColumns = stmtTargetTableScan.getPartitioningColumns();
                assert (partitioningColumns.size() == 1);
                AbstractExpression targetPartitionColExpr = partitioningColumns.get(0).getExpression();
                TupleValueExpression selectedExpr = subquery.getOutputExpression(i);
                assert (!valueEquivalence.containsKey(targetPartitionColExpr));
                assert (!valueEquivalence.containsKey(selectedExpr));
                Set<AbstractExpression> equivSet = new HashSet<>();
                equivSet.add(targetPartitionColExpr);
                equivSet.add(selectedExpr);
                valueEquivalence.put(targetPartitionColExpr, equivSet);
                valueEquivalence.put(selectedExpr, equivSet);
                setEquivalenceForPartitioningCol = true;
            }
            ++i;
        }
        if (!setEquivalenceForPartitioningCol) {
            // partitioning column of target table is not being set from value produced by the subquery.
            m_recentErrorMsg = "Partitioning column must be assigned a value " + "produced by the subquery in an " + getSqlType() + " INTO ... SELECT statement.";
            return null;
        }
        m_partitioning.analyzeForMultiPartitionAccess(tables, valueEquivalence);
        if (!m_partitioning.isJoinValid()) {
            m_recentErrorMsg = "Partitioning could not be determined for " + getSqlType() + " INTO ... SELECT statement.  " + "Please ensure that statement does not attempt to copy row data from one partition to another, " + "which is unsupported.";
            return null;
        }
    }
    return subquery.getBestCostPlan().rootPlanGraph;
}

Example 3 with SchemaColumn

use of org.voltdb.plannodes.SchemaColumn in project voltdb by VoltDB.

the class StatementPartitioning method analyzeForMultiPartitionAccess.

/**
     * Given the query's list of tables and its collection(s) of equality-filtered columns and their equivalents,
     * determine whether all joins involving partitioned tables can be executed locally on a single partition.
     * This is only the case when they include equality comparisons between partition key columns.
     * VoltDB will reject joins of multiple partitioned tables unless all their partition keys are
     * constrained to be equal to each other.
     * Example: select * from T1, T2 where T1.ID = T2.ID
     * Additionally, in this case, there may be a constant equality filter on any of the columns,
     * which we want to extract as our SP partitioning parameter.
     *
     * @param tableAliasList The tables.
     * @param valueEquivalence Their column equality filters
     * @return the number of independently partitioned tables
     *         -- partitioned tables that aren't joined or filtered by the same value.
     *         The caller can raise an alarm if there is more than one.
     */
public void analyzeForMultiPartitionAccess(Collection<StmtTableScan> scans, HashMap<AbstractExpression, Set<AbstractExpression>> valueEquivalence) {
    //* enable to debug */ System.out.println("DEBUG: analyze4MPAccess w/ scans:" + scans.size() + " filters:" + valueEquivalence.size());
    TupleValueExpression tokenPartitionKey = null;
    Set<Set<AbstractExpression>> eqSets = new HashSet<Set<AbstractExpression>>();
    int unfilteredPartitionKeyCount = 0;
    // reset this flag to forget the last result of the multiple partition access path.
    // AdHoc with parameters will call this function at least two times
    // By default this flag should be true.
    setJoinValid(true);
    setJoinInvalidReason(null);
    boolean subqueryHasReceiveNode = false;
    boolean hasPartitionedTableJoin = false;
    // Iterate over the tables to collect partition columns.
    for (StmtTableScan tableScan : scans) {
        // Replicated tables don't need filter coverage.
        if (tableScan.getIsReplicated()) {
            continue;
        }
        // The partition column can be null in an obscure edge case.
        // The table is declared non-replicated yet specifies no partitioning column.
        // This can occur legitimately when views based on partitioned tables neglect to group by the partition column.
        // The interpretation of this edge case is that the table has "randomly distributed data".
        // In such a case, the table is valid for use by MP queries only and can only be joined with replicated tables
        // because it has no recognized partitioning join key.
        List<SchemaColumn> columnsNeedingCoverage = tableScan.getPartitioningColumns();
        if (tableScan instanceof StmtSubqueryScan) {
            StmtSubqueryScan subScan = (StmtSubqueryScan) tableScan;
            subScan.promoteSinglePartitionInfo(valueEquivalence, eqSets);
            CompiledPlan subqueryPlan = subScan.getBestCostPlan();
            if ((!subScan.canRunInOneFragment()) || ((subqueryPlan != null) && subqueryPlan.rootPlanGraph.hasAnyNodeOfClass(AbstractReceivePlanNode.class))) {
                if (subqueryHasReceiveNode) {
                    // Has found another subquery with receive node on the same level
                    // Not going to support this kind of subquery join with 2 fragment plan.
                    setJoinValid(false);
                    setJoinInvalidReason("This multipartition query is not plannable.  " + "It has a subquery which cannot be single partition.");
                    // Still needs to count the independent partition tables
                    break;
                }
                subqueryHasReceiveNode = true;
                if (subScan.isTableAggregate()) {
                    // Any process based on this subquery should require 1 fragment only.
                    continue;
                }
            } else {
                // this subquery partition table without receive node
                hasPartitionedTableJoin = true;
            }
        } else {
            // This table is a partition table
            hasPartitionedTableJoin = true;
        }
        boolean unfiltered = true;
        for (AbstractExpression candidateColumn : valueEquivalence.keySet()) {
            if (!(candidateColumn instanceof TupleValueExpression)) {
                continue;
            }
            TupleValueExpression candidatePartitionKey = (TupleValueExpression) candidateColumn;
            if (!canCoverPartitioningColumn(candidatePartitionKey, columnsNeedingCoverage)) {
                continue;
            }
            unfiltered = false;
            if (tokenPartitionKey == null) {
                tokenPartitionKey = candidatePartitionKey;
            }
            eqSets.add(valueEquivalence.get(candidatePartitionKey));
        }
        if (unfiltered) {
            ++unfilteredPartitionKeyCount;
        }
    }
    // end for each table StmtTableScan in the collection
    m_countOfIndependentlyPartitionedTables = eqSets.size() + unfilteredPartitionKeyCount;
    //* enable to debug */ System.out.println("DEBUG: analyze4MPAccess found: " + m_countOfIndependentlyPartitionedTables + " = " + eqSets.size() + " + " + unfilteredPartitionKeyCount);
    if (m_countOfIndependentlyPartitionedTables > 1) {
        setJoinValid(false);
        setJoinInvalidReason("This query is not plannable.  " + "The planner cannot guarantee that all rows would be in a single partition.");
    }
    // on outer level. Not going to support this kind of join.
    if (subqueryHasReceiveNode && hasPartitionedTableJoin) {
        setJoinValid(false);
        setJoinInvalidReason("This query is not plannable.  It has a subquery which needs cross-partition access.");
    }
    if ((unfilteredPartitionKeyCount == 0) && (eqSets.size() == 1)) {
        for (Set<AbstractExpression> partitioningValues : eqSets) {
            for (AbstractExpression constExpr : partitioningValues) {
                if (constExpr instanceof TupleValueExpression) {
                    continue;
                }
                VoltType valueType = tokenPartitionKey.getValueType();
                addPartitioningExpression(tokenPartitionKey.getTableName() + '.' + tokenPartitionKey.getColumnName(), constExpr, valueType);
                // Only need one constant value.
                break;
            }
        }
    }
}

Example 4 with SchemaColumn

use of org.voltdb.plannodes.SchemaColumn in project voltdb by VoltDB.

the class StatementPartitioning method canCoverPartitioningColumn.

private static boolean canCoverPartitioningColumn(TupleValueExpression candidatePartitionKey, List<SchemaColumn> columnsNeedingCoverage) {
    if (columnsNeedingCoverage == null)
        return false;
    for (SchemaColumn col : columnsNeedingCoverage) {
        String partitionedTableAlias = col.getTableAlias();
        String columnNeedingCoverage = col.getColumnAlias();
        assert (candidatePartitionKey.getTableAlias() != null);
        if (!candidatePartitionKey.getTableAlias().equals(partitionedTableAlias)) {
            continue;
        }
        String candidateColumnName = candidatePartitionKey.getColumnName();
        if (!candidateColumnName.equals(columnNeedingCoverage)) {
            continue;
        }
        // Maybe need more checkings
        return true;
    }
    return false;
}

Example 5 with SchemaColumn

use of org.voltdb.plannodes.SchemaColumn in project voltdb by VoltDB.

the class TestSelfJoins method testSelfJoin.

public void testSelfJoin() {
    AbstractPlanNode pn = compile("select * FROM R1 A JOIN R1 B ON A.C = B.C WHERE B.A > 0 AND A.C < 3");
    pn = pn.getChild(0).getChild(0);
    assertTrue(pn instanceof NestLoopPlanNode);
    assertEquals(4, pn.getOutputSchema().getColumns().size());
    assertEquals(2, pn.getChildCount());
    AbstractPlanNode c = pn.getChild(0);
    assertTrue(c instanceof SeqScanPlanNode);
    SeqScanPlanNode ss = (SeqScanPlanNode) c;
    assertEquals("R1", ss.getTargetTableName());
    assertEquals("A", ss.getTargetTableAlias());
    assertEquals(ExpressionType.COMPARE_LESSTHAN, ss.getPredicate().getExpressionType());
    c = pn.getChild(1);
    assertTrue(c instanceof SeqScanPlanNode);
    ss = (SeqScanPlanNode) c;
    assertEquals("R1", ss.getTargetTableName());
    assertEquals("B", ss.getTargetTableAlias());
    assertEquals(ExpressionType.COMPARE_GREATERTHAN, ss.getPredicate().getExpressionType());
    pn = compile("select * FROM R1 JOIN R1 B ON R1.C = B.C");
    pn = pn.getChild(0).getChild(0);
    assertTrue(pn instanceof NestLoopPlanNode);
    assertEquals(4, pn.getOutputSchema().getColumns().size());
    assertEquals(2, pn.getChildCount());
    c = pn.getChild(0);
    assertTrue(c instanceof SeqScanPlanNode);
    ss = (SeqScanPlanNode) c;
    assertEquals("R1", ss.getTargetTableName());
    assertEquals("R1", ss.getTargetTableAlias());
    c = pn.getChild(1);
    assertTrue(c instanceof SeqScanPlanNode);
    ss = (SeqScanPlanNode) c;
    assertEquals("R1", ss.getTargetTableName());
    assertEquals("B", ss.getTargetTableAlias());
    pn = compile("select A.A, A.C, B.A, B.C FROM R1 A JOIN R1 B ON A.C = B.C");
    pn = pn.getChild(0).getChild(0);
    assertTrue(pn instanceof NestLoopPlanNode);
    assertEquals(4, pn.getOutputSchema().getColumns().size());
    pn = compile("select A,B.C  FROM R1 A JOIN R2 B USING(A)");
    pn = pn.getChild(0);
    assertTrue(pn instanceof ProjectionPlanNode);
    NodeSchema ns = pn.getOutputSchema();
    for (SchemaColumn sc : ns.getColumns()) {
        AbstractExpression e = sc.getExpression();
        assertTrue(e instanceof TupleValueExpression);
        TupleValueExpression tve = (TupleValueExpression) e;
        assertNotSame(-1, tve.getColumnIndex());
    }
}