Examples with TableEstimates org.voltdb.compiler.DatabaseEstimates.TableEstimates used on opensource projects

Example 1 with TableEstimates

use of org.voltdb.compiler.DatabaseEstimates.TableEstimates in project voltdb by VoltDB.

the class SeqScanPlanNode method computeCostEstimates.

@Override
public void computeCostEstimates(long childOutputTupleCountEstimate, DatabaseEstimates estimates, ScalarValueHints[] paramHints) {
    if (m_isSubQuery) {
        // Get estimates from the sub-query
        // @TODO For the sub-query the cost estimates will be calculated separately
        // At the moment its contribution to the parent's cost plan is irrelevant because
        // all parent plans have the same best cost plan for the sub-query
        m_estimatedProcessedTupleCount = SUBQUERY_TABLE_ESTIMATES_HACK.minTuples;
        m_estimatedOutputTupleCount = SUBQUERY_TABLE_ESTIMATES_HACK.minTuples;
        return;
    }
    Table target = ((StmtTargetTableScan) m_tableScan).getTargetTable();
    TableEstimates tableEstimates = estimates.getEstimatesForTable(target.getTypeName());
    // This maxTuples value estimates the number of tuples fetched from the sequential scan.
    // It's a vague measure of the cost of the scan.
    // Its accuracy depends a lot on what kind of post-filtering or projection needs to happen, if any.
    // The tuplesRead value is also used to estimate the number of RESULT rows, regardless of
    // how effective post-filtering might be -- as if all rows passed the filters.
    // This is at least semi-consistent with the ignoring of post-filter effects in IndexScanPlanNode.
    // In effect, though, it gives index scans an "unfair" advantage when they reduce the estimated result size
    // by taking into account the indexed filters -- follow-on plan steps, sorts (etc.), are costed lower
    // as if they are operating on fewer rows than would have come out of the seqscan,
    // though that's nonsense.
    // In any case, it's important to keep an eye on any changes (discounts) to SeqScanPlanNode's costing
    // here to make sure that SeqScanPlanNode never gains an unfair advantage over IndexScanPlanNode.
    m_estimatedProcessedTupleCount = tableEstimates.maxTuples;
    m_estimatedOutputTupleCount = tableEstimates.maxTuples;
}