org.apache.cassandra.io.sstable

Class Downsampling

java.lang.Object

org.apache.cassandra.io.sstable.Downsampling

public class Downsampling
extends java.lang.Object

Field Summary

Fields

Modifier and Type	Field and Description
static int	BASE_SAMPLING_LEVEL The base (down)sampling level determines the granularity at which we can down/upsample.

Constructor Summary

Constructors

Constructor and Description
Downsampling()

Method Summary

Modifier and Type	Method and Description
static int	getEffectiveIndexIntervalAfterIndex(int index, int samplingLevel, int minIndexInterval) Calculates the effective index interval after the entry at `index` in an IndexSummary.
static java.util.List<java.lang.Integer>	getOriginalIndexes(int samplingLevel) Returns a list that can be used to translate current index summary indexes to their original index before downsampling.
static java.util.List<java.lang.Integer>	getSamplingPattern(int samplingLevel) Gets a list L of starting indices for downsampling rounds: the first round should start with the offset given by L[0], the second by the offset in L[1], etc.
static int[]	getStartPoints(int currentSamplingLevel, int newSamplingLevel)

Methods inherited from class java.lang.Object

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

Field Detail

BASE_SAMPLING_LEVEL

public static final int BASE_SAMPLING_LEVEL

The base (down)sampling level determines the granularity at which we can down/upsample. A higher number allows us to approximate more closely the ideal sampling. (It could also mean we do a lot of expensive almost-no-op resamplings from N to N-1, but the thresholds in IndexSummaryManager prevent that.) BSL must be a power of two in order to have good sampling patterns. This cannot be changed without rebuilding all index summaries at full sampling; for now we treat it as a constant.

See Also:

Constant Field Values

Constructor Detail

Downsampling

public Downsampling()

Method Detail

getSamplingPattern

public static java.util.List<java.lang.Integer> getSamplingPattern(int samplingLevel)

Gets a list L of starting indices for downsampling rounds: the first round should start with the offset given by L[0], the second by the offset in L[1], etc.

Parameters:

samplingLevel - the base sampling level

Returns:

A list of `samplingLevel` unique indices between 0 and `samplingLevel`

getOriginalIndexes

public static java.util.List<java.lang.Integer> getOriginalIndexes(int samplingLevel)

Returns a list that can be used to translate current index summary indexes to their original index before downsampling. (This repeats every `samplingLevel`, so that's how many entries we return.) For example, if [0, 64] is returned, the current index summary entry at index 0 was originally at index 0, and the current index 1 was originally at index 64.

Parameters:

samplingLevel - the current sampling level for the index summary

Returns:

a list of original indexes for current summary entries

getEffectiveIndexIntervalAfterIndex

public static int getEffectiveIndexIntervalAfterIndex(int index,
                                                      int samplingLevel,
                                                      int minIndexInterval)

Calculates the effective index interval after the entry at `index` in an IndexSummary. In other words, this returns the number of partitions in the primary on-disk index before the next partition that has an entry in the index summary. If samplingLevel == BASE_SAMPLING_LEVEL, this will be equal to the index interval.

Parameters:

index - an index into an IndexSummary

samplingLevel - the current sampling level for that IndexSummary

minIndexInterval - the min index interval (effective index interval at full sampling)

Returns:

the number of partitions before the next index summary entry, inclusive on one end

getStartPoints

public static int[] getStartPoints(int currentSamplingLevel,
                                   int newSamplingLevel)