Examples with TupleValueExpression org.voltdb.expressions.TupleValueExpression used on opensource projects

Example 31 with TupleValueExpression

use of org.voltdb.expressions.TupleValueExpression in project voltdb by VoltDB.

the class TestIndexSelection method checkIndexPredicateContains.

private void checkIndexPredicateContains(AbstractPlanNode pn, String... columns) {
    assertEquals(1, pn.getChildCount());
    pn = pn.getChild(0);
    assertEquals(PlanNodeType.INDEXSCAN, pn.getPlanNodeType());
    IndexScanPlanNode ipn = (IndexScanPlanNode) pn;
    AbstractExpression pred = ipn.getPredicate();
    assertNotNull(pred);
    List<TupleValueExpression> tves = pred.findAllTupleValueSubexpressions();
    for (TupleValueExpression tve : tves) {
        boolean match = false;
        for (String column : columns) {
            if (tve.getColumnName().equals(column)) {
                match = true;
                break;
            }
        }
        assertTrue(match);
    }
}

Example 32 with TupleValueExpression

use of org.voltdb.expressions.TupleValueExpression in project voltdb by VoltDB.

the class MaterializedViewProcessor method isIndexOptimalForMinMax.

private static boolean isIndexOptimalForMinMax(MatViewIndexMatchingGroupby matchingCase, AbstractExpression minMaxAggExpr, List<ColumnRef> indexedColRefs, List<AbstractExpression> indexedExprs, List<Column> srcColumnArray) {
    // We have minMaxAggExpr and the index also has one extra column
    switch(matchingCase) {
        case GB_COL_IDX_COL:
            if (!(minMaxAggExpr instanceof TupleValueExpression)) {
                // so the minMaxAggExpr must be TupleValueExpression.
                return false;
            }
            int aggSrcColIdx = ((TupleValueExpression) minMaxAggExpr).getColumnIndex();
            Column aggSrcCol = srcColumnArray.get(aggSrcColIdx);
            Column lastIndexCol = indexedColRefs.get(indexedColRefs.size() - 1).getColumn();
            // Compare the two columns, if they are equal as well, then this is the optimal index! Congrats!
            if (aggSrcCol.equals(lastIndexCol)) {
                return true;
            }
            break;
        case GB_COL_IDX_EXP:
        case GB_EXP_IDX_EXP:
            if (indexedExprs.get(indexedExprs.size() - 1).equals(minMaxAggExpr)) {
                return true;
            }
            break;
        default:
            assert (false);
    }
    // this is not the optimal index candidate for now
    return false;
}

Example 33 with TupleValueExpression

use of org.voltdb.expressions.TupleValueExpression in project voltdb by VoltDB.

the class PlanAssembler method addCoordinatorToDMLNode.

/**
     * Add a receive node, a sum or limit node, and a send node to the given DML node.
     * If the DML target is a replicated table, it will add a limit node,
     * otherwise it adds a sum node.
     *
     * @param dmlRoot
     * @param isReplicated Whether or not the target table is a replicated table.
     * @return
     */
private static AbstractPlanNode addCoordinatorToDMLNode(AbstractPlanNode dmlRoot, boolean isReplicated) {
    dmlRoot = SubPlanAssembler.addSendReceivePair(dmlRoot);
    AbstractPlanNode sumOrLimitNode;
    if (isReplicated) {
        // Replicated table DML result doesn't need to be summed. All partitions should
        // modify the same number of tuples in replicated table, so just pick the result from
        // any partition.
        LimitPlanNode limitNode = new LimitPlanNode();
        sumOrLimitNode = limitNode;
        limitNode.setLimit(1);
    } else {
        // create the nodes being pushed on top of dmlRoot.
        AggregatePlanNode countNode = new AggregatePlanNode();
        sumOrLimitNode = countNode;
        // configure the count aggregate (sum) node to produce a single
        // output column containing the result of the sum.
        // Create a TVE that should match the tuple count input column
        // This TVE is magic.
        // really really need to make this less hard-wired
        TupleValueExpression count_tve = new TupleValueExpression(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "modified_tuples", "modified_tuples", 0);
        count_tve.setValueType(VoltType.BIGINT);
        count_tve.setValueSize(VoltType.BIGINT.getLengthInBytesForFixedTypes());
        countNode.addAggregate(ExpressionType.AGGREGATE_SUM, false, 0, count_tve);
        // The output column. Not really based on a TVE (it is really the
        // count expression represented by the count configured above). But
        // this is sufficient for now.  This looks identical to the above
        // TVE but it's logically different so we'll create a fresh one.
        TupleValueExpression tve = new TupleValueExpression(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "modified_tuples", "modified_tuples", 0);
        tve.setValueType(VoltType.BIGINT);
        tve.setValueSize(VoltType.BIGINT.getLengthInBytesForFixedTypes());
        NodeSchema count_schema = new NodeSchema();
        count_schema.addColumn(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "modified_tuples", "modified_tuples", tve);
        countNode.setOutputSchema(count_schema);
    }
    // connect the nodes to build the graph
    sumOrLimitNode.addAndLinkChild(dmlRoot);
    SendPlanNode sendNode = new SendPlanNode();
    sendNode.addAndLinkChild(sumOrLimitNode);
    return sendNode;
}

Example 34 with TupleValueExpression

use of org.voltdb.expressions.TupleValueExpression in project voltdb by VoltDB.

the class PlanAssembler method handleAggregationOperators.

private AbstractPlanNode handleAggregationOperators(AbstractPlanNode root) {
    /*
         * "Select A from T group by A" is grouped but has no aggregate operator
         * expressions. Catch that case by checking the grouped flag
         */
    if (m_parsedSelect.hasAggregateOrGroupby()) {
        AggregatePlanNode aggNode = null;
        // i.e., on the coordinator
        AggregatePlanNode topAggNode = null;
        IndexGroupByInfo gbInfo = new IndexGroupByInfo();
        if (root instanceof AbstractReceivePlanNode) {
            // for distinct that does not group by partition column
            if (!m_parsedSelect.hasAggregateDistinct() || m_parsedSelect.hasPartitionColumnInGroupby()) {
                AbstractPlanNode candidate = root.getChild(0).getChild(0);
                gbInfo.m_multiPartition = true;
                switchToIndexScanForGroupBy(candidate, gbInfo);
            }
        } else if (switchToIndexScanForGroupBy(root, gbInfo)) {
            root = gbInfo.m_indexAccess;
        }
        boolean needHashAgg = gbInfo.needHashAggregator(root, m_parsedSelect);
        // Construct the aggregate nodes
        if (needHashAgg) {
            if (m_parsedSelect.m_mvFixInfo.needed()) {
                // TODO: may optimize this edge case in future
                aggNode = new HashAggregatePlanNode();
            } else {
                if (gbInfo.isChangedToSerialAggregate()) {
                    assert (root instanceof ReceivePlanNode);
                    aggNode = new AggregatePlanNode();
                } else if (gbInfo.isChangedToPartialAggregate()) {
                    aggNode = new PartialAggregatePlanNode(gbInfo.m_coveredGroupByColumns);
                } else {
                    aggNode = new HashAggregatePlanNode();
                }
                topAggNode = new HashAggregatePlanNode();
            }
        } else {
            aggNode = new AggregatePlanNode();
            if (!m_parsedSelect.m_mvFixInfo.needed()) {
                topAggNode = new AggregatePlanNode();
            }
        }
        NodeSchema agg_schema = new NodeSchema();
        NodeSchema top_agg_schema = new NodeSchema();
        for (int outputColumnIndex = 0; outputColumnIndex < m_parsedSelect.m_aggResultColumns.size(); outputColumnIndex += 1) {
            ParsedColInfo col = m_parsedSelect.m_aggResultColumns.get(outputColumnIndex);
            AbstractExpression rootExpr = col.expression;
            AbstractExpression agg_input_expr = null;
            SchemaColumn schema_col = null;
            SchemaColumn top_schema_col = null;
            if (rootExpr instanceof AggregateExpression) {
                ExpressionType agg_expression_type = rootExpr.getExpressionType();
                agg_input_expr = rootExpr.getLeft();
                // A bit of a hack: ProjectionNodes after the
                // aggregate node need the output columns here to
                // contain TupleValueExpressions (effectively on a temp table).
                // So we construct one based on the output of the
                // aggregate expression, the column alias provided by HSQL,
                // and the offset into the output table schema for the
                // aggregate node that we're computing.
                // Oh, oh, it's magic, you know..
                TupleValueExpression tve = new TupleValueExpression(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "", col.alias, rootExpr, outputColumnIndex);
                tve.setDifferentiator(col.differentiator);
                boolean is_distinct = ((AggregateExpression) rootExpr).isDistinct();
                aggNode.addAggregate(agg_expression_type, is_distinct, outputColumnIndex, agg_input_expr);
                schema_col = new SchemaColumn(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "", col.alias, tve, outputColumnIndex);
                top_schema_col = new SchemaColumn(AbstractParsedStmt.TEMP_TABLE_NAME, AbstractParsedStmt.TEMP_TABLE_NAME, "", col.alias, tve, outputColumnIndex);
                /*
                     * Special case count(*), count(), sum(), min() and max() to
                     * push them down to each partition. It will do the
                     * push-down if the select columns only contains the listed
                     * aggregate operators and other group-by columns. If the
                     * select columns includes any other aggregates, it will not
                     * do the push-down. - nshi
                     */
                if (topAggNode != null) {
                    ExpressionType top_expression_type = agg_expression_type;
                    /*
                         * For count(*), count() and sum(), the pushed-down
                         * aggregate node doesn't change. An extra sum()
                         * aggregate node is added to the coordinator to sum up
                         * the numbers from all the partitions. The input schema
                         * and the output schema of the sum() aggregate node is
                         * the same as the output schema of the push-down
                         * aggregate node.
                         *
                         * If DISTINCT is specified, don't do push-down for
                         * count() and sum() when not group by partition column.
                         * An exception is the aggregation arguments are the
                         * partition column (ENG-4980).
                         */
                    if (agg_expression_type == ExpressionType.AGGREGATE_COUNT_STAR || agg_expression_type == ExpressionType.AGGREGATE_COUNT || agg_expression_type == ExpressionType.AGGREGATE_SUM) {
                        if (is_distinct && !(m_parsedSelect.hasPartitionColumnInGroupby() || canPushDownDistinctAggregation((AggregateExpression) rootExpr))) {
                            topAggNode = null;
                        } else {
                            // for aggregate distinct when group by
                            // partition column, the top aggregate node
                            // will be dropped later, thus there is no
                            // effect to assign the top_expression_type.
                            top_expression_type = ExpressionType.AGGREGATE_SUM;
                        }
                    } else /*
                         * For min() and max(), the pushed-down aggregate node
                         * doesn't change. An extra aggregate node of the same
                         * type is added to the coordinator. The input schema
                         * and the output schema of the top aggregate node is
                         * the same as the output schema of the pushed-down
                         * aggregate node.
                         *
                         * APPROX_COUNT_DISTINCT can be similarly pushed down, but
                         * must be split into two different functions, which is
                         * done later, from pushDownAggregate().
                         */
                    if (agg_expression_type != ExpressionType.AGGREGATE_MIN && agg_expression_type != ExpressionType.AGGREGATE_MAX && agg_expression_type != ExpressionType.AGGREGATE_APPROX_COUNT_DISTINCT) {
                        /*
                             * Unsupported aggregate for push-down (AVG for example).
                             */
                        topAggNode = null;
                    }
                    if (topAggNode != null) {
                        /*
                             * Input column of the top aggregate node is the
                             * output column of the push-down aggregate node
                             */
                        boolean topDistinctFalse = false;
                        topAggNode.addAggregate(top_expression_type, topDistinctFalse, outputColumnIndex, tve);
                    }
                }
            // end if we have a top agg node
            } else {
                // has already been broken down.
                assert (!rootExpr.hasAnySubexpressionOfClass(AggregateExpression.class));
                /*
                     * These columns are the pass through columns that are not being
                     * aggregated on. These are the ones from the SELECT list. They
                     * MUST already exist in the child node's output. Find them and
                     * add them to the aggregate's output.
                     */
                schema_col = new SchemaColumn(col.tableName, col.tableAlias, col.columnName, col.alias, col.expression, outputColumnIndex);
                AbstractExpression topExpr = null;
                if (col.groupBy) {
                    topExpr = m_parsedSelect.m_groupByExpressions.get(col.alias);
                } else {
                    topExpr = col.expression;
                }
                top_schema_col = new SchemaColumn(col.tableName, col.tableAlias, col.columnName, col.alias, topExpr, outputColumnIndex);
            }
            agg_schema.addColumn(schema_col);
            top_agg_schema.addColumn(top_schema_col);
        }
        for (ParsedColInfo col : m_parsedSelect.groupByColumns()) {
            aggNode.addGroupByExpression(col.expression);
            if (topAggNode != null) {
                topAggNode.addGroupByExpression(m_parsedSelect.m_groupByExpressions.get(col.alias));
            }
        }
        aggNode.setOutputSchema(agg_schema);
        if (topAggNode != null) {
            if (m_parsedSelect.hasComplexGroupby()) {
                topAggNode.setOutputSchema(top_agg_schema);
            } else {
                topAggNode.setOutputSchema(agg_schema);
            }
        }
        // Never push down aggregation for MV fix case.
        root = pushDownAggregate(root, aggNode, topAggNode, m_parsedSelect);
    }
    return handleDistinctWithGroupby(root);
}

Example 35 with TupleValueExpression

use of org.voltdb.expressions.TupleValueExpression in project voltdb by VoltDB.

the class PlanAssembler method calculateGroupbyColumnsCovered.

private List<Integer> calculateGroupbyColumnsCovered(Index index, String fromTableAlias, List<AbstractExpression> bindings) {
    List<Integer> coveredGroupByColumns = new ArrayList<>();
    List<ParsedColInfo> groupBys = m_parsedSelect.groupByColumns();
    String exprsjson = index.getExpressionsjson();
    if (exprsjson.isEmpty()) {
        List<ColumnRef> indexedColRefs = CatalogUtil.getSortedCatalogItems(index.getColumns(), "index");
        for (int j = 0; j < indexedColRefs.size(); j++) {
            String indexColumnName = indexedColRefs.get(j).getColumn().getName();
            // ignore order of keys in GROUP BY expr
            int ithCovered = 0;
            boolean foundPrefixedColumn = false;
            for (; ithCovered < groupBys.size(); ithCovered++) {
                AbstractExpression gbExpr = groupBys.get(ithCovered).expression;
                if (!(gbExpr instanceof TupleValueExpression)) {
                    continue;
                }
                TupleValueExpression gbTVE = (TupleValueExpression) gbExpr;
                // TVE column index has not been resolved currently
                if (fromTableAlias.equals(gbTVE.getTableAlias()) && indexColumnName.equals(gbTVE.getColumnName())) {
                    foundPrefixedColumn = true;
                    break;
                }
            }
            if (!foundPrefixedColumn) {
                // no prefix match any more
                break;
            }
            coveredGroupByColumns.add(ithCovered);
            if (coveredGroupByColumns.size() == groupBys.size()) {
                // covered all group by columns already
                break;
            }
        }
    } else {
        StmtTableScan fromTableScan = m_parsedSelect.getStmtTableScanByAlias(fromTableAlias);
        // either pure expression index or mix of expressions and simple columns
        List<AbstractExpression> indexedExprs = null;
        try {
            indexedExprs = AbstractExpression.fromJSONArrayString(exprsjson, fromTableScan);
        } catch (JSONException e) {
            e.printStackTrace();
            // This case sounds impossible
            return coveredGroupByColumns;
        }
        for (AbstractExpression indexExpr : indexedExprs) {
            // ignore order of keys in GROUP BY expr
            List<AbstractExpression> binding = null;
            for (int ithCovered = 0; ithCovered < groupBys.size(); ithCovered++) {
                AbstractExpression gbExpr = groupBys.get(ithCovered).expression;
                binding = gbExpr.bindingToIndexedExpression(indexExpr);
                if (binding != null) {
                    bindings.addAll(binding);
                    coveredGroupByColumns.add(ithCovered);
                    break;
                }
            }
            // no prefix match any more or covered all group by columns already
            if (binding == null || coveredGroupByColumns.size() == groupBys.size()) {
                break;
            }
        }
    }
    return coveredGroupByColumns;
}