Package org.apache.drill.exec.planner.logical

Examples of org.apache.drill.exec.planner.logical.DrillAggregateRel

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
    return PrelUtil.getPlannerSettings(call.getPlanner()).isHashAggEnabled();
  }

  @Override
  public void onMatch(RelOptRuleCall call) {
    final DrillAggregateRel aggregate = (DrillAggregateRel) call.rel(0);
    final RelNode input = call.rel(1);

    if (aggregate.containsDistinctCall() || aggregate.getGroupCount() == 0) {
      // currently, don't use HashAggregate if any of the logical aggrs contains DISTINCT or
      // if there are no grouping keys
      return;
    }

    RelTraitSet traits = null;

    try {
      if (aggregate.getGroupSet().isEmpty()) {
        DrillDistributionTrait singleDist = DrillDistributionTrait.SINGLETON;
        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(singleDist);
        createTransformRequest(call, aggregate, input, traits);
      } else {
        // hash distribute on all grouping keys
View Full Code Here
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
    super(rule, "ConvertCountToDirectScan:" + id);
  }

  @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;

    GroupScan oldGrpScan = scan.getGroupScan();

    // Only apply the rule when :
    //    1) scan knows the exact row count in getSize() call,
    //    2) No GroupBY key,
    //    3) only one agg function (Check if it's count(*) below).
    //    4) No distinct agg call.
    if (! (oldGrpScan.getScanStats().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(*)  == >  empty arg  ==>  rowCount
      //  count(Not-null-input) ==> rowCount
      if (aggCall.getArgList().isEmpty() ||
          (aggCall.getArgList().size() == 1 &&
           ! agg.getChild().getRowType().getFieldList().get(aggCall.getArgList().get(0).intValue()).getType().isNullable())) {
        cnt = (long) oldGrpScan.getScanStats().getRecordCount();
      } else if (aggCall.getArgList().size() == 1) {
      // count(columnName) ==> Agg ( Scan )) ==> columnValueCount
        int index = aggCall.getArgList().get(0);
        String columnName = scan.getRowType().getFieldNames().get(index).toLowerCase();

        cnt = oldGrpScan.getColumnValueCount(SchemaPath.getSimplePath(columnName));
      } else {
        return; // do nothing.
      }

      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);
    }

  }
View Full Code Here
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
    return PrelUtil.getPlannerSettings(call.getPlanner()).isStreamAggEnabled();
  }

  @Override
  public void onMatch(RelOptRuleCall call) {
    final DrillAggregateRel aggregate = (DrillAggregateRel) call.rel(0);
    final RelNode input = aggregate.getChild();
    final RelCollation collation = getCollation(aggregate);
    RelTraitSet traits = null;

    if (aggregate.containsDistinctCall()) {
      // currently, don't use StreamingAggregate if any of the logical aggrs contains DISTINCT
      return;
    }

    try {
      if (aggregate.getGroupSet().isEmpty()) {
        DrillDistributionTrait singleDist = DrillDistributionTrait.SINGLETON;
        final RelTraitSet singleDistTrait = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(singleDist);

        if (create2PhasePlan(call, aggregate)) {
          traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL) ;

          RelNode convertedInput = convert(input, traits);
          new SubsetTransformer<DrillAggregateRel, InvalidRelException>(call){

            public RelNode convertChild(final DrillAggregateRel join, final RelNode rel) throws InvalidRelException {
              DrillDistributionTrait toDist = rel.getTraitSet().getTrait(DrillDistributionTraitDef.INSTANCE);
              RelTraitSet traits = newTraitSet(Prel.DRILL_PHYSICAL, toDist);
              RelNode newInput = convert(input, traits);

              StreamAggPrel phase1Agg = new StreamAggPrel(aggregate.getCluster(), traits, newInput,
                  aggregate.getGroupSet(),
                  aggregate.getAggCallList(),
                  OperatorPhase.PHASE_1of2);

              UnionExchangePrel exch =
                  new UnionExchangePrel(phase1Agg.getCluster(), singleDistTrait, phase1Agg);

              return  new StreamAggPrel(aggregate.getCluster(), singleDistTrait, exch,
                  aggregate.getGroupSet(),
                  phase1Agg.getPhase2AggCalls(),
                  OperatorPhase.PHASE_2of2);
            }
          }.go(aggregate, convertedInput);

        } else {
          createTransformRequest(call, aggregate, input, singleDistTrait);
        }
      } else {
        // hash distribute on all grouping keys
        final DrillDistributionTrait distOnAllKeys =
            new DrillDistributionTrait(DrillDistributionTrait.DistributionType.HASH_DISTRIBUTED,
                                       ImmutableList.copyOf(getDistributionField(aggregate, true)));

        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(collation).plus(distOnAllKeys);
        createTransformRequest(call, aggregate, input, traits);

        // hash distribute on one grouping key
        DrillDistributionTrait distOnOneKey =
            new DrillDistributionTrait(DrillDistributionTrait.DistributionType.HASH_DISTRIBUTED,
                                       ImmutableList.copyOf(getDistributionField(aggregate, false)));

        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(collation).plus(distOnOneKey);
        // Temporarily commenting out the single distkey plan since a few tpch queries (e.g 01.sql) get stuck
        // in VolcanoPlanner.canonize() method. Note that the corresponding single distkey plan for HashAggr works
        // ok.  One possibility is that here we have dist on single key but collation on all keys, so that
        // might be causing some problem.
        /// TODO: re-enable this plan after resolving the issue.
        // createTransformRequest(call, aggregate, input, traits);

        if (create2PhasePlan(call, aggregate)) {
          traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL) ;
          RelNode convertedInput = convert(input, traits);

          new SubsetTransformer<DrillAggregateRel, InvalidRelException>(call){

            public RelNode convertChild(final DrillAggregateRel aggregate, final RelNode rel) throws InvalidRelException {
              DrillDistributionTrait toDist = rel.getTraitSet().getTrait(DrillDistributionTraitDef.INSTANCE);
              RelTraitSet traits = newTraitSet(Prel.DRILL_PHYSICAL, collation, toDist);
              RelNode newInput = convert(input, traits);

              StreamAggPrel phase1Agg = new StreamAggPrel(aggregate.getCluster(), traits, newInput,
                  aggregate.getGroupSet(),
                  aggregate.getAggCallList(),
                  OperatorPhase.PHASE_1of2);

              int numEndPoints = PrelUtil.getSettings(phase1Agg.getCluster()).numEndPoints();

              HashToMergeExchangePrel exch =
                  new HashToMergeExchangePrel(phase1Agg.getCluster(), phase1Agg.getTraitSet().plus(Prel.DRILL_PHYSICAL).plus(distOnAllKeys),
                      phase1Agg, ImmutableList.copyOf(getDistributionField(aggregate, true)),
                      collation,
                      numEndPoints);

              return new StreamAggPrel(aggregate.getCluster(), exch.getTraitSet(), exch,
                  aggregate.getGroupSet(),
                  phase1Agg.getPhase2AggCalls(),
                  OperatorPhase.PHASE_2of2);
            }
          }.go(aggregate, convertedInput);
        }
View Full Code Here
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
  public void onMatch(RelOptRuleCall call) {
    if (!PrelUtil.getPlannerSettings(call.getPlanner()).isHashAggEnabled()) {
      return;
    }

    final DrillAggregateRel aggregate = (DrillAggregateRel) call.rel(0);
    final RelNode input = call.rel(1);

    if (aggregate.containsDistinctCall() || aggregate.getGroupCount() == 0) {
      // currently, don't use HashAggregate if any of the logical aggrs contains DISTINCT or
      // if there are no grouping keys
      return;
    }

    RelTraitSet traits = null;

    try {
      if (aggregate.getGroupSet().isEmpty()) {
        DrillDistributionTrait singleDist = DrillDistributionTrait.SINGLETON;
        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(singleDist);
        createTransformRequest(call, aggregate, input, traits);
      } else {
        // hash distribute on all grouping keys
View Full Code Here
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
    return PrelUtil.getPlannerSettings(call.getPlanner()).isStreamAggEnabled();
  }

  @Override
  public void onMatch(RelOptRuleCall call) {
    final DrillAggregateRel aggregate = (DrillAggregateRel) call.rel(0);
    final RelNode input = aggregate.getChild();
    final RelCollation collation = getCollation(aggregate);
    RelTraitSet traits = null;

    if (aggregate.containsDistinctCall()) {
      // currently, don't use StreamingAggregate if any of the logical aggrs contains DISTINCT
      return;
    }

    try {
      if (aggregate.getGroupSet().isEmpty()) {
        DrillDistributionTrait singleDist = DrillDistributionTrait.SINGLETON;
        final RelTraitSet singleDistTrait = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(singleDist);

        if (create2PhasePlan(call, aggregate)) {
          traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL) ;

          RelNode convertedInput = convert(input, traits);
          new SubsetTransformer<DrillAggregateRel, InvalidRelException>(call){

            @Override
            public RelNode convertChild(final DrillAggregateRel join, final RelNode rel) throws InvalidRelException {
              DrillDistributionTrait toDist = rel.getTraitSet().getTrait(DrillDistributionTraitDef.INSTANCE);
              RelTraitSet traits = newTraitSet(Prel.DRILL_PHYSICAL, toDist);
              RelNode newInput = convert(input, traits);

              StreamAggPrel phase1Agg = new StreamAggPrel(aggregate.getCluster(), traits, newInput,
                  aggregate.getGroupSet(),
                  aggregate.getAggCallList(),
                  OperatorPhase.PHASE_1of2);

              UnionExchangePrel exch =
                  new UnionExchangePrel(phase1Agg.getCluster(), singleDistTrait, phase1Agg);

              return  new StreamAggPrel(aggregate.getCluster(), singleDistTrait, exch,
                  aggregate.getGroupSet(),
                  phase1Agg.getPhase2AggCalls(),
                  OperatorPhase.PHASE_2of2);
            }
          }.go(aggregate, convertedInput);

        } else {
          createTransformRequest(call, aggregate, input, singleDistTrait);
        }
      } else {
        // hash distribute on all grouping keys
        final DrillDistributionTrait distOnAllKeys =
            new DrillDistributionTrait(DrillDistributionTrait.DistributionType.HASH_DISTRIBUTED,
                                       ImmutableList.copyOf(getDistributionField(aggregate, true)));

        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(collation).plus(distOnAllKeys);
        createTransformRequest(call, aggregate, input, traits);

        // hash distribute on one grouping key
        DrillDistributionTrait distOnOneKey =
            new DrillDistributionTrait(DrillDistributionTrait.DistributionType.HASH_DISTRIBUTED,
                                       ImmutableList.copyOf(getDistributionField(aggregate, false)));

        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(collation).plus(distOnOneKey);
        // Temporarily commenting out the single distkey plan since a few tpch queries (e.g 01.sql) get stuck
        // in VolcanoPlanner.canonize() method. Note that the corresponding single distkey plan for HashAggr works
        // ok.  One possibility is that here we have dist on single key but collation on all keys, so that
        // might be causing some problem.
        /// TODO: re-enable this plan after resolving the issue.
        // createTransformRequest(call, aggregate, input, traits);

        if (create2PhasePlan(call, aggregate)) {
          traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL) ;
          RelNode convertedInput = convert(input, traits);

          new SubsetTransformer<DrillAggregateRel, InvalidRelException>(call){

            @Override
            public RelNode convertChild(final DrillAggregateRel aggregate, final RelNode rel) throws InvalidRelException {
              DrillDistributionTrait toDist = rel.getTraitSet().getTrait(DrillDistributionTraitDef.INSTANCE);
              RelTraitSet traits = newTraitSet(Prel.DRILL_PHYSICAL, collation, toDist);
              RelNode newInput = convert(input, traits);

              StreamAggPrel phase1Agg = new StreamAggPrel(aggregate.getCluster(), traits, newInput,
                  aggregate.getGroupSet(),
                  aggregate.getAggCallList(),
                  OperatorPhase.PHASE_1of2);

              int numEndPoints = PrelUtil.getSettings(phase1Agg.getCluster()).numEndPoints();

              HashToMergeExchangePrel exch =
                  new HashToMergeExchangePrel(phase1Agg.getCluster(), phase1Agg.getTraitSet().plus(Prel.DRILL_PHYSICAL).plus(distOnAllKeys),
                      phase1Agg, ImmutableList.copyOf(getDistributionField(aggregate, true)),
                      collation,
                      numEndPoints);

              return new StreamAggPrel(aggregate.getCluster(), exch.getTraitSet(), exch,
                  aggregate.getGroupSet(),
                  phase1Agg.getPhase2AggCalls(),
                  OperatorPhase.PHASE_2of2);
            }
          }.go(aggregate, convertedInput);
        }
View Full Code Here
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
    super(rule, "ConvertCountToDirectScan:" + id);
  }

  @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;

    GroupScan oldGrpScan = scan.getGroupScan();

    // Only apply the rule when :
    //    1) scan knows the exact row count in getSize() call,
    //    2) No GroupBY key,
    //    3) only one agg function (Check if it's count(*) below).
    //    4) No distinct agg call.
    if (! (oldGrpScan.getScanStats().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(*)  == >  empty arg  ==>  rowCount
      //  count(Not-null-input) ==> rowCount
      if (aggCall.getArgList().isEmpty() ||
          (aggCall.getArgList().size() == 1 &&
           ! agg.getChild().getRowType().getFieldList().get(aggCall.getArgList().get(0).intValue()).getType().isNullable())) {
        cnt = (long) oldGrpScan.getScanStats().getRecordCount();
      } else if (aggCall.getArgList().size() == 1) {
      // count(columnName) ==> Agg ( Scan )) ==> columnValueCount
        int index = aggCall.getArgList().get(0);
        String columnName = scan.getRowType().getFieldNames().get(index).toLowerCase();

        cnt = oldGrpScan.getColumnValueCount(SchemaPath.getSimplePath(columnName));
      } else {
        return; // do nothing.
      }

      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);
    }

  }
View Full Code Here
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
  @Override
  public void onMatch(RelOptRuleCall call) {
    if (!PrelUtil.getPlannerSettings(call.getPlanner()).isHashAggEnabled()) return;

    final DrillAggregateRel aggregate = (DrillAggregateRel) call.rel(0);
    final RelNode input = call.rel(1);

    if (aggregate.containsDistinctCall() || aggregate.getGroupCount() == 0) {
      // currently, don't use HashAggregate if any of the logical aggrs contains DISTINCT or
      // if there are no grouping keys
      return;
    }

    RelTraitSet traits = null;

    try {
      if (aggregate.getGroupSet().isEmpty()) {
        DrillDistributionTrait singleDist = DrillDistributionTrait.SINGLETON;
        traits = call.getPlanner().emptyTraitSet().plus(Prel.DRILL_PHYSICAL).plus(singleDist);
        createTransformRequest(call, aggregate, input, traits);
      } else {
        // hash distribute on all grouping keys
View Full Code Here
TOP

Related Classes of org.apache.drill.exec.planner.logical.DrillAggregateRel

  • org.apache.drill.common.expression.FieldReference
  • org.apache.drill.common.expression.LogicalExpression
  • org.apache.drill.exec.planner.physical.ConvertCountToDirectScan
  • org.apache.drill.exec.planner.physical.HashAggPrule
  • org.apache.drill.exec.planner.physical.StreamAggPrule
    • Copyright © 2018 www.massapicom. All rights reserved.
    All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.