Examples with DbGroupScan org.apache.drill.exec.physical.base.DbGroupScan used on opensource projects

Example 11 with DbGroupScan

use of org.apache.drill.exec.physical.base.DbGroupScan in project drill by apache.

the class AbstractIndexPlanGenerator method convertRowType.

protected RelDataType convertRowType(RelDataType origRowType, RelDataTypeFactory typeFactory) {
    if (getRowKeyIndex(origRowType, origScan) >= 0) {
        // row key already present
        return origRowType;
    }
    List<RelDataTypeField> fields = new ArrayList<>();
    fields.addAll(origRowType.getFieldList());
    fields.add(new RelDataTypeFieldImpl(((DbGroupScan) IndexPlanUtils.getGroupScan(origScan)).getRowKeyName(), fields.size(), typeFactory.createSqlType(SqlTypeName.ANY)));
    return new RelRecordType(fields);
}

Example 12 with DbGroupScan

use of org.apache.drill.exec.physical.base.DbGroupScan in project drill by apache.

the class MapRDBStatistics method initialize.

public boolean initialize(RexNode condition, DrillScanRelBase scanRel, IndexCallContext context) {
    GroupScan scan = IndexPlanUtils.getGroupScan(scanRel);
    PlannerSettings settings = PrelUtil.getPlannerSettings(scanRel.getCluster().getPlanner());
    rowKeyJoinBackIOFactor = settings.getIndexRowKeyJoinCostFactor();
    if (scan instanceof DbGroupScan) {
        String conditionAsStr = convertRexToString(condition, scanRel.getRowType());
        if (statsCache.get(conditionAsStr) == null) {
            IndexCollection indexes = ((DbGroupScan) scan).getSecondaryIndexCollection(scanRel);
            populateStats(condition, indexes, scanRel, context);
            logger.info("index_plan_info: initialize: scanRel #{} and groupScan {} got fulltable {}, statsCache: {}, fiStatsCache: {}", scanRel.getId(), System.identityHashCode(scan), fullTableScanPayload, statsCache, fIStatsCache);
            return true;
        }
    }
    return false;
}

Example 13 with DbGroupScan

use of org.apache.drill.exec.physical.base.DbGroupScan in project drill by apache.

the class MapRDBIndexDescriptor method getCost.

@Override
public RelOptCost getCost(IndexProperties indexProps, RelOptPlanner planner, int numProjectedFields, GroupScan primaryTableGroupScan) {
    Preconditions.checkArgument(primaryTableGroupScan instanceof DbGroupScan);
    DbGroupScan dbGroupScan = (DbGroupScan) primaryTableGroupScan;
    DrillCostFactory costFactory = (DrillCostFactory) planner.getCostFactory();
    double totalRows = indexProps.getTotalRows();
    double leadRowCount = indexProps.getLeadingSelectivity() * totalRows;
    double avgRowSize = indexProps.getAvgRowSize();
    if (indexProps.isCovering()) {
        // covering index
        // int numIndexCols = allFields.size();
        // for disk i/o, all index columns are going to be read into memory
        double numBlocks = Math.ceil((leadRowCount * avgRowSize) / pluginCost.getBlockSize(primaryTableGroupScan));
        double diskCost = numBlocks * pluginCost.getSequentialBlockReadCost(primaryTableGroupScan);
        // cpu cost is cost of filter evaluation for the remainder condition
        double cpuCost = 0.0;
        if (indexProps.getTotalRemainderFilter() != null) {
            cpuCost = leadRowCount * DrillCostBase.COMPARE_CPU_COST;
        }
        // TODO: add network cost once full table scan also considers network cost
        double networkCost = 0.0;
        return costFactory.makeCost(leadRowCount, cpuCost, diskCost, networkCost);
    } else {
        // non-covering index
        // int numIndexCols = allFields.size();
        double numBlocksIndex = Math.ceil((leadRowCount * avgRowSize) / pluginCost.getBlockSize(primaryTableGroupScan));
        double diskCostIndex = numBlocksIndex * pluginCost.getSequentialBlockReadCost(primaryTableGroupScan);
        // for the primary table join-back each row may belong to a different block, so in general num_blocks = num_rows;
        // however, num_blocks cannot exceed the total number of blocks of the table
        double totalBlocksPrimary = Math.ceil((dbGroupScan.getColumns().size() * pluginCost.getAverageColumnSize(primaryTableGroupScan) * totalRows) / pluginCost.getBlockSize(primaryTableGroupScan));
        double diskBlocksPrimary = Math.min(totalBlocksPrimary, leadRowCount);
        double diskCostPrimary = diskBlocksPrimary * pluginCost.getRandomBlockReadCost(primaryTableGroupScan);
        double diskCostTotal = diskCostIndex + diskCostPrimary;
        // cpu cost of remainder condition evaluation over the selected rows
        double cpuCost = 0.0;
        if (indexProps.getTotalRemainderFilter() != null) {
            cpuCost = leadRowCount * DrillCostBase.COMPARE_CPU_COST;
        }
        // TODO: add network cost once full table scan also considers network cost
        double networkCost = 0.0;
        return costFactory.makeCost(leadRowCount, cpuCost, diskCostTotal, networkCost);
    }
}

Example 14 with DbGroupScan

use of org.apache.drill.exec.physical.base.DbGroupScan in project drill by apache.

the class FunctionalIndexHelper method convertRowTypeForIndexScan.

/**
 * For IndexScan in non-covering case, rowType to return contains only row_key('_id') of primary table.
 * so the rowType for IndexScan should be converted from [Primary_table.row_key, primary_table.indexed_col]
 * to [indexTable.row_key(primary_table.indexed_col), indexTable.<primary_key.row_key> (Primary_table.row_key)]
 * This will impact the columns of scan, the rowType of ScanRel
 *
 * @param origScan
 * @param idxMarker  the IndexableExprMarker that has analyzed original index condition on top of index scan
 * @param idxScan
 * @return
 */
public static RelDataType convertRowTypeForIndexScan(DrillScanRelBase origScan, IndexableExprMarker idxMarker, IndexGroupScan idxScan, FunctionalIndexInfo functionInfo) {
    RelDataTypeFactory typeFactory = origScan.getCluster().getTypeFactory();
    List<RelDataTypeField> fields = new ArrayList<>();
    Set<SchemaPath> rowPaths = new LinkedHashSet<>();
    // row_key in the rowType of scan on primary table
    RelDataTypeField rowkey_primary;
    RelRecordType newRowType = null;
    DbGroupScan scan = (DbGroupScan) IndexPlanUtils.getGroupScan(origScan);
    // first add row_key of primary table,
    rowkey_primary = new RelDataTypeFieldImpl(scan.getRowKeyName(), fields.size(), typeFactory.createSqlType(SqlTypeName.ANY));
    fields.add(rowkey_primary);
    Map<RexNode, LogicalExpression> idxExprMap = idxMarker.getIndexableExpression();
    for (LogicalExpression indexedExpr : idxExprMap.values()) {
        if (indexedExpr instanceof SchemaPath) {
            rowPaths.add((SchemaPath) indexedExpr);
        } else if (indexedExpr instanceof CastExpression) {
            SchemaPath newPath = functionInfo.getNewPathFromExpr(indexedExpr);
            if (newPath != null) {
                rowPaths.add(newPath);
            }
        }
    }
    for (SchemaPath newPath : rowPaths) {
        fields.add(new RelDataTypeFieldImpl(newPath.getRootSegmentPath(), fields.size(), typeFactory.createSqlType(SqlTypeName.ANY)));
    }
    // update columns of groupscan accordingly
    Set<RelDataTypeField> rowfields = Sets.newLinkedHashSet();
    final List<SchemaPath> columns = Lists.newArrayList();
    for (RelDataTypeField f : fields) {
        SchemaPath path = SchemaPath.parseFromString(f.getName());
        rowfields.add(new RelDataTypeFieldImpl(path.getRootSegmentPath(), rowfields.size(), typeFactory.createSqlType(SqlTypeName.ANY)));
        columns.add(path);
    }
    idxScan.setColumns(columns);
    // rowtype does not take the whole path, but only the rootSegment of the SchemaPath
    newRowType = new RelRecordType(Lists.newArrayList(rowfields));
    return newRowType;
}