com.bigdata.bop.joinGraph.rto

Class SampleBase

java.lang.Object

com.bigdata.bop.joinGraph.rto.SampleBase

Direct Known Subclasses:

EdgeSample, VertexSample

public abstract class SampleBase
extends Object

Base class for a sample taken from a vertex (access path) or edge (cutoff join).

Version:

$Id$ TODO Large samples should be buffered on the IMemoryManager so they do not pose a burden on the heap. This will require us to manage the allocation contexts so we can release samples in a timely manner once they are no longer used and always release samples by the time the RTO is finished. [There is an additional twist if we have fully materialized some part of the join since we no longer need to evaluate that path segment. If the RTO can interleave query evaluation with exploration then we can take advantage of these materialized solutions.]

Author:

Bryan Thompson

Field Summary

Fields

Modifier and Type	Field and Description
long	estCard The total estimated cardinality of the underlying access path (for a vertex) or the join path segment (for a cutoff join).
EstimateEnum	estimateEnum Indicates whether the estimate is exact, an upper bound, or a lower bound.
int	limit The limit used to produce the sample.

Constructor Summary

Constructors

Constructor and Description
SampleBase(long estimatedCardinality, int limit, EstimateEnum estimateEnum, IBindingSet[] sample)

Method Summary

Methods

Modifier and Type	Method and Description
IBindingSet[]	getSample() The sampled solution set.
boolean	isExact() Return true iff this sample is the fully materialized solution for the vertex or join path segment.
boolean	isUnderflow() Return true iff this sample has cardinality underflow (the sample is empty).
String	toString()
protected void	toString(StringBuilder sb) Hook for extending toString().

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

estCard

public final long estCard

The total estimated cardinality of the underlying access path (for a vertex) or the join path segment (for a cutoff join). TODO When using a non-perfect index, the estimated cardinality is only part of the cost. The #of tuples scanned is also important. Even when scanning and filtering in key order this could trigger random IOs unless the file has index order (an IndexSegment file has index order but a BTree on a journal does not).

limit

public final int limit

The limit used to produce the sample.

estimateEnum

public final EstimateEnum estimateEnum

Indicates whether the estimate is exact, an upper bound, or a lower bound. TODO When the input to a cutoff join is EstimateEnum.Exact, we could run the join against the sample rather than the disk by wrapping the sample as an inline access path. TODO This field should be used to avoid needless re-computation of a join whose exact solution is already known. We already do this within the runtime optimizer. To go further than that we need to do the partial evaluation of the join graph.

Constructor Detail

SampleBase

public SampleBase(long estimatedCardinality,
          int limit,
          EstimateEnum estimateEnum,
          IBindingSet[] sample)

Parameters:

estimatedCardinality - The estimated cardinality.

limit - The cutoff limit used to make that cardinality estimate.

estimateEnum - Type safe enumeration indication various edge conditions which can arise when making a cardinality estimate.

sample - The sample.

Method Detail

isExact

public boolean isExact()

Return true iff this sample is the fully materialized solution for the vertex or join path segment.

isUnderflow

public boolean isUnderflow()

Return true iff this sample has cardinality underflow (the sample is empty). Cardinality underflow occurs when the sampling process was unable to find any solutions. Underflow is typically addressed by increasing the sample size, but sometimes underflow indicates that an access path (if it has filters) or a join may not have any solutions in the data.

getSample

public IBindingSet[] getSample()

The sampled solution set.

Returns:

The sampled solution set -or- null if it has been released. TODO Wrap up as an IChunkMessage so we can store this on the native heap?

toString

protected void toString(StringBuilder sb)

Hook for extending toString().

Parameters:

sb - The buffer into which the implementation can write additional information.

toString

public String toString()

Overrides:

toString in class Object