org.apache.calcite.rel.metadata

Interface BuiltInMetadata.Parallelism

All Superinterfaces:

Metadata

Enclosing class:

BuiltInMetadata

public static interface BuiltInMetadata.Parallelism
extends Metadata

Metadata about the degree of parallelism of a relational expression, and how its operators are assigned to processes with independent resource pools.

Method Summary

Methods

Modifier and Type	Method and Description
Boolean	isPhaseTransition() Returns whether each physical operator implementing this relational expression belongs to a different process than its inputs.
Integer	splitCount() Returns the number of distinct splits of the data.

Methods inherited from interface org.apache.calcite.rel.metadata.Metadata

rel

Method Detail

isPhaseTransition

Boolean isPhaseTransition()

Returns whether each physical operator implementing this relational expression belongs to a different process than its inputs.

A collection of operators processing all of the splits of a particular stage in the query pipeline is called a "phase". A phase starts with a leaf node such as a TableScan, or with a phase-change node such as an Exchange. Hadoop's shuffle operator (a form of sort-exchange) causes data to be sent across the network.

splitCount

Integer splitCount()

Returns the number of distinct splits of the data.

Note that splits must be distinct. For broadcast, where each copy is the same, returns 1.

Thus the split count is the proportion of the data seen by each operator instance.