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

Example 6 with GroupScan

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

the class ParquetPushDownFilter method doOnMatch.

protected void doOnMatch(RelOptRuleCall call, FilterPrel filter, ProjectPrel project, ScanPrel scan) {
    ParquetGroupScan groupScan = (ParquetGroupScan) scan.getGroupScan();
    if (groupScan.getFilter() != null && !groupScan.getFilter().equals(ValueExpressions.BooleanExpression.TRUE)) {
        return;
    }
    RexNode condition = null;
    if (project == null) {
        condition = filter.getCondition();
    } else {
        // get the filter as if it were below the projection.
        condition = RelOptUtil.pushFilterPastProject(filter.getCondition(), project);
    }
    if (condition == null || condition.equals(ValueExpressions.BooleanExpression.TRUE)) {
        return;
    }
    // get a conjunctions of the filter condition. For each conjunction, if it refers to ITEM or FLATTEN expression
    // then we could not pushed down. Otherwise, it's qualified to be pushed down.
    final List<RexNode> predList = RelOptUtil.conjunctions(condition);
    final List<RexNode> qualifiedPredList = Lists.newArrayList();
    for (final RexNode pred : predList) {
        if (DrillRelOptUtil.findItemOrFlatten(pred, ImmutableList.<RexNode>of()) == null) {
            qualifiedPredList.add(pred);
        }
    }
    final RexNode qualifedPred = RexUtil.composeConjunction(filter.getCluster().getRexBuilder(), qualifiedPredList, true);
    if (qualifedPred == null) {
        return;
    }
    LogicalExpression conditionExp = DrillOptiq.toDrill(new DrillParseContext(PrelUtil.getPlannerSettings(call.getPlanner())), scan, qualifedPred);
    Stopwatch timer = Stopwatch.createStarted();
    final GroupScan newGroupScan = groupScan.applyFilter(conditionExp, optimizerContext, optimizerContext.getFunctionRegistry(), optimizerContext.getPlannerSettings().getOptions());
    logger.info("Took {} ms to apply filter on parquet row groups. ", timer.elapsed(TimeUnit.MILLISECONDS));
    if (newGroupScan == null) {
        return;
    }
    final ScanPrel newScanRel = ScanPrel.create(scan, scan.getTraitSet(), newGroupScan, scan.getRowType());
    RelNode inputRel = newScanRel;
    if (project != null) {
        inputRel = project.copy(project.getTraitSet(), ImmutableList.of(inputRel));
    }
    final RelNode newFilter = filter.copy(filter.getTraitSet(), ImmutableList.<RelNode>of(inputRel));
    call.transformTo(newFilter);
}

Example 7 with GroupScan

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

the class ParquetPartitionDescriptor method createTableScan.

@Override
public TableScan createTableScan(List<PartitionLocation> newPartitionLocation, String cacheFileRoot, boolean wasAllPartitionsPruned, MetadataContext metaContext) throws Exception {
    List<String> newFiles = Lists.newArrayList();
    for (final PartitionLocation location : newPartitionLocation) {
        newFiles.add(location.getEntirePartitionLocation());
    }
    final GroupScan newGroupScan = createNewGroupScan(newFiles, cacheFileRoot, wasAllPartitionsPruned, metaContext);
    return new DrillScanRel(scanRel.getCluster(), scanRel.getTraitSet().plus(DrillRel.DRILL_LOGICAL), scanRel.getTable(), newGroupScan, scanRel.getRowType(), scanRel.getColumns(), true);
}

Example 8 with GroupScan

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

the class ParquetPruneScanRule method getFilterOnProjectParquet.

public static final RelOptRule getFilterOnProjectParquet(OptimizerRulesContext optimizerRulesContext) {
    return new PruneScanRule(RelOptHelper.some(DrillFilterRel.class, RelOptHelper.some(DrillProjectRel.class, RelOptHelper.any(DrillScanRel.class))), "PruneScanRule:Filter_On_Project_Parquet", optimizerRulesContext) {

        @Override
        public PartitionDescriptor getPartitionDescriptor(PlannerSettings settings, TableScan scanRel) {
            return new ParquetPartitionDescriptor(settings, (DrillScanRel) scanRel);
        }

        @Override
        public boolean matches(RelOptRuleCall call) {
            final DrillScanRel scan = call.rel(2);
            GroupScan groupScan = scan.getGroupScan();
            // this rule is applicable only for parquet based partition pruning
            if (PrelUtil.getPlannerSettings(scan.getCluster().getPlanner()).isHepPartitionPruningEnabled()) {
                return groupScan instanceof ParquetGroupScan && groupScan.supportsPartitionFilterPushdown() && !scan.partitionFilterPushdown();
            } else {
                return groupScan instanceof ParquetGroupScan && groupScan.supportsPartitionFilterPushdown();
            }
        }

        @Override
        public void onMatch(RelOptRuleCall call) {
            final DrillFilterRel filterRel = call.rel(0);
            final DrillProjectRel projectRel = call.rel(1);
            final DrillScanRel scanRel = call.rel(2);
            doOnMatch(call, filterRel, projectRel, scanRel);
        }
    };
}

Example 9 with GroupScan

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

the class ConvertCountToDirectScan method onMatch.

@Override
public void onMatch(RelOptRuleCall call) {
    final DrillAggregateRel agg = (DrillAggregateRel) call.rel(0);
    final DrillScanRel scan = (DrillScanRel) call.rel(call.rels.length - 1);
    final DrillProjectRel proj = call.rels.length == 3 ? (DrillProjectRel) call.rel(1) : null;
    final GroupScan oldGrpScan = scan.getGroupScan();
    final PlannerSettings settings = PrelUtil.getPlannerSettings(call.getPlanner());
    //    4) No distinct agg call.
    if (!(oldGrpScan.getScanStats(settings).getGroupScanProperty().hasExactRowCount() && agg.getGroupCount() == 0 && agg.getAggCallList().size() == 1 && !agg.containsDistinctCall())) {
        return;
    }
    AggregateCall aggCall = agg.getAggCallList().get(0);
    if (aggCall.getAggregation().getName().equals("COUNT")) {
        long cnt = 0;
        //  count(Not-null-input) ==> rowCount
        if (aggCall.getArgList().isEmpty() || (aggCall.getArgList().size() == 1 && !agg.getInput().getRowType().getFieldList().get(aggCall.getArgList().get(0).intValue()).getType().isNullable())) {
            cnt = (long) oldGrpScan.getScanStats(settings).getRecordCount();
        } else if (aggCall.getArgList().size() == 1) {
            // count(columnName) ==> Agg ( Scan )) ==> columnValueCount
            int index = aggCall.getArgList().get(0);
            if (proj != null) {
                if (proj.getProjects().get(index) instanceof RexInputRef) {
                    index = ((RexInputRef) proj.getProjects().get(index)).getIndex();
                } else {
                    // do not apply for all other cases.
                    return;
                }
            }
            String columnName = scan.getRowType().getFieldNames().get(index).toLowerCase();
            cnt = oldGrpScan.getColumnValueCount(SchemaPath.getSimplePath(columnName));
            if (cnt == GroupScan.NO_COLUMN_STATS) {
                // if column stats are not available don't apply this rule
                return;
            }
        } else {
            // do nothing.
            return;
        }
        RelDataType scanRowType = getCountDirectScanRowType(agg.getCluster().getTypeFactory());
        final ScanPrel newScan = ScanPrel.create(scan, scan.getTraitSet().plus(Prel.DRILL_PHYSICAL).plus(DrillDistributionTrait.SINGLETON), getCountDirectScan(cnt), scanRowType);
        List<RexNode> exprs = Lists.newArrayList();
        exprs.add(RexInputRef.of(0, scanRowType));
        final ProjectPrel newProj = new ProjectPrel(agg.getCluster(), agg.getTraitSet().plus(Prel.DRILL_PHYSICAL).plus(DrillDistributionTrait.SINGLETON), newScan, exprs, agg.getRowType());
        call.transformTo(newProj);
    }
}