public class BeamUnionRel extends org.apache.calcite.rel.core.Union implements BeamRelNode
BeamRelNode
to replace a Union
.
BeamUnionRel
needs the input of it have the same WindowFn
. From the SQL
perspective, two cases are supported:
1) Do not use grouped window function
:
select * from person UNION select * from person
2) Use the same grouped window function
, with the same param:
select id, count(*) from person
group by id, TUMBLE(order_time, INTERVAL '1' HOUR)
UNION
select * from person
group by id, TUMBLE(order_time, INTERVAL '1' HOUR)
Inputs with different group functions are NOT supported:
select id, count(*) from person
group by id, TUMBLE(order_time, INTERVAL '1' HOUR)
UNION
select * from person
group by id, TUMBLE(order_time, INTERVAL '2' HOUR)
Constructor and Description |
---|
BeamUnionRel(org.apache.calcite.plan.RelOptCluster cluster,
org.apache.calcite.plan.RelTraitSet traits,
java.util.List<org.apache.calcite.rel.RelNode> inputs,
boolean all) |
Modifier and Type | Method and Description |
---|---|
BeamCostModel |
beamComputeSelfCost(org.apache.calcite.plan.RelOptPlanner planner,
org.apache.calcite.rel.metadata.RelMetadataQuery mq)
This method is called by
org.apache.beam.sdk.extensions.sql.impl.CalciteQueryPlanner.NonCumulativeCostImpl . |
PTransform<PCollectionList<Row>,PCollection<Row>> |
buildPTransform() |
org.apache.calcite.rel.core.SetOp |
copy(org.apache.calcite.plan.RelTraitSet traitSet,
java.util.List<org.apache.calcite.rel.RelNode> inputs,
boolean all) |
NodeStats |
estimateNodeStats(org.apache.calcite.rel.metadata.RelMetadataQuery mq)
This method is called by
org.apache.beam.sdk.extensions.sql.impl.planner.RelMdNodeStats . |
copy, deriveRowType, explainTerms, getInputs, isHomogeneous, replaceInput
accept, accept, childrenAccept, collectVariablesSet, collectVariablesUsed, computeDigest, computeSelfCost, computeSelfCost, explain, getChildExps, getCluster, getCollationList, getConvention, getCorrelVariable, getDescription, getDigest, getExpectedInputRowType, getId, getInput, getQuery, getRelTypeName, getRows, getRowType, getTable, getTraitSet, getVariablesSet, getVariablesStopped, isDistinct, isKey, isValid, isValid, metadata, onRegister, recomputeDigest, register, sole, toString
clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait
getPCollectionInputs, getPipelineOptions, isBounded
accept, accept, childrenAccept, collectVariablesSet, collectVariablesUsed, computeSelfCost, computeSelfCost, copy, estimateRowCount, explain, getChildExps, getCollationList, getConvention, getCorrelVariable, getExpectedInputRowType, getInput, getInputs, getQuery, getRelTypeName, getRows, getRowType, getTable, getVariablesSet, getVariablesStopped, isDistinct, isKey, isValid, isValid, metadata, onRegister, recomputeDigest, register, replaceInput
public BeamUnionRel(org.apache.calcite.plan.RelOptCluster cluster, org.apache.calcite.plan.RelTraitSet traits, java.util.List<org.apache.calcite.rel.RelNode> inputs, boolean all)
public org.apache.calcite.rel.core.SetOp copy(org.apache.calcite.plan.RelTraitSet traitSet, java.util.List<org.apache.calcite.rel.RelNode> inputs, boolean all)
copy
in class org.apache.calcite.rel.core.SetOp
public PTransform<PCollectionList<Row>,PCollection<Row>> buildPTransform()
buildPTransform
in interface BeamRelNode
public NodeStats estimateNodeStats(org.apache.calcite.rel.metadata.RelMetadataQuery mq)
BeamRelNode
org.apache.beam.sdk.extensions.sql.impl.planner.RelMdNodeStats
. This is currently only used in
SQLTransform Path (and not JDBC path). When a RelNode wants to calculate its BeamCost or
estimate its NodeStats, it may need NodeStat of its inputs. However, it should not call this
directly (because maybe its inputs are not physical yet). It should call BeamSqlRelUtils.getNodeStats(org.apache.calcite.rel.RelNode,
org.apache.calcite.rel.metadata.RelMetadataQuery)
instead.estimateNodeStats
in interface BeamRelNode
public BeamCostModel beamComputeSelfCost(org.apache.calcite.plan.RelOptPlanner planner, org.apache.calcite.rel.metadata.RelMetadataQuery mq)
BeamRelNode
org.apache.beam.sdk.extensions.sql.impl.CalciteQueryPlanner.NonCumulativeCostImpl
. This is
currently only used in SQLTransform Path (and not JDBC path). This is needed when Calcite Query
Planner wants to get the cost of a plan. Instead of calling this directly for a node, if we
needed that it should be obtained by calling mq.getNonCumulativeCost. This way RelMetadataQuery
will call this method instead of ComputeSelfCost if the handler is set correctly (see org.apache.beam.sdk.extensions.sql.impl.CalciteQueryPlanner#convertToBeamRel(String)
)beamComputeSelfCost
in interface BeamRelNode