/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hive.metastore;
import java.util.Random;
import javolution.util.FastBitSet;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.hive.common.type.HiveDecimal;
import org.apache.hadoop.io.Text;
/*
* https://en.wikipedia.org/wiki/Flajolet%E2%80%93Martin_algorithm
* We implement Flajolet–Martin algorithm in this class.
* The Flajolet–Martin algorithm is an algorithm for approximating the number of distinct elements
* in a stream with a single pass and space-consumption which is logarithmic in the maximum number
* of possible distinct elements in the stream. The algorithm was introduced by Philippe Flajolet
* and G. Nigel Martin in their 1984 paper "Probabilistic Counting Algorithms for Data Base Applications".
* Later it has been refined in the papers "LogLog counting of large cardinalities" by Marianne Durand
* and Philippe Flajolet, and "HyperLogLog: The analysis of a near-optimal cardinality estimation
* algorithm" by Philippe Flajolet et al.
*/
/*
* The algorithm works like this.
* (1) Set the number of bit vectors, i.e., numBitVectors, based on the precision.
* (2) For each bit vector, generate hash value of the long value and mod it by 2^bitVectorSize-1. (addToEstimator)
* (3) Set the index (addToEstimator)
* (4) Take the average of the index for all the bit vectors and get the estimated NDV (estimateNumDistinctValues).
*/
public class NumDistinctValueEstimator {
static final Log LOG = LogFactory.getLog(NumDistinctValueEstimator.class.getName());
/* We want a,b,x to come from a finite field of size 0 to k, where k is a prime number.
* 2^p - 1 is prime for p = 31. Hence bitvectorSize has to be 31. Pick k to be 2^p -1.
* If a,b,x didn't come from a finite field ax1 + b mod k and ax2 + b mod k will not be pair wise
* independent. As a consequence, the hash values will not distribute uniformly from 0 to 2^p-1
* thus introducing errors in the estimates.
*/
private static final int BIT_VECTOR_SIZE = 31;
private final int numBitVectors;
// Refer to Flajolet-Martin'86 for the value of phi
private static final double PHI = 0.77351;
private final int[] a;
private final int[] b;
private final FastBitSet[] bitVector;
private final Random aValue;
private final Random bValue;
/* Create a new distinctValueEstimator
*/
public NumDistinctValueEstimator(int numBitVectors) {
this.numBitVectors = numBitVectors;
bitVector = new FastBitSet[numBitVectors];
for (int i=0; i< numBitVectors; i++) {
bitVector[i] = new FastBitSet(BIT_VECTOR_SIZE);
}
a = new int[numBitVectors];
b = new int[numBitVectors];
/* Use a large prime number as a seed to the random number generator.
* Java's random number generator uses the Linear Congruential Generator to generate random
* numbers using the following recurrence relation,
*
* X(n+1) = (a X(n) + c ) mod m
*
* where X0 is the seed. Java implementation uses m = 2^48. This is problematic because 2^48
* is not a prime number and hence the set of numbers from 0 to m don't form a finite field.
* If these numbers don't come from a finite field any give X(n) and X(n+1) may not be pair
* wise independent.
*
* However, empirically passing in prime numbers as seeds seems to work better than when passing
* composite numbers as seeds. Ideally Java's Random should pick m such that m is prime.
*
*/
aValue = new Random(99397);
bValue = new Random(9876413);
for (int i = 0; i < numBitVectors; i++) {
int randVal;
/* a and b shouldn't be even; If a and b are even, then none of the values
* will set bit 0 thus introducing errors in the estimate. Both a and b can be even
* 25% of the times and as a result 25% of the bit vectors could be inaccurate. To avoid this
* always pick odd values for a and b.
*/
do {
randVal = aValue.nextInt();
} while (randVal % 2 == 0);
a[i] = randVal;
do {
randVal = bValue.nextInt();
} while (randVal % 2 == 0);
b[i] = randVal;
if (a[i] < 0) {
a[i] = a[i] + (1 << BIT_VECTOR_SIZE - 1);
}
if (b[i] < 0) {
b[i] = b[i] + (1 << BIT_VECTOR_SIZE - 1);
}
}
}
public NumDistinctValueEstimator(String s, int numBitVectors) {
this.numBitVectors = numBitVectors;
FastBitSet bitVectorDeser[] = deserialize(s, numBitVectors);
bitVector = new FastBitSet[numBitVectors];
for(int i=0; i <numBitVectors; i++) {
bitVector[i] = new FastBitSet(BIT_VECTOR_SIZE);
bitVector[i].clear();
bitVector[i].or(bitVectorDeser[i]);
}
a = null;
b = null;
aValue = null;
bValue = null;
}
/**
* Resets a distinctValueEstimator object to its original state.
*/
public void reset() {
for (int i=0; i< numBitVectors; i++) {
bitVector[i].clear();
}
}
public FastBitSet getBitVector(int index) {
return bitVector[index];
}
public int getnumBitVectors() {
return numBitVectors;
}
public int getBitVectorSize() {
return BIT_VECTOR_SIZE;
}
public void printNumDistinctValueEstimator() {
String t = new String();
LOG.debug("NumDistinctValueEstimator");
LOG.debug("Number of Vectors:");
LOG.debug(numBitVectors);
LOG.debug("Vector Size: ");
LOG.debug(BIT_VECTOR_SIZE);
for (int i=0; i < numBitVectors; i++) {
t = t + bitVector[i].toString();
}
LOG.debug("Serialized Vectors: ");
LOG.debug(t);
}
/* Serializes a distinctValueEstimator object to Text for transport.
*
*/
public Text serialize() {
String s = new String();
for(int i=0; i < numBitVectors; i++) {
s = s + (bitVector[i].toString());
}
return new Text(s);
}
/* Deserializes from string to FastBitSet; Creates a NumDistinctValueEstimator object and
* returns it.
*/
private FastBitSet[] deserialize(String s, int numBitVectors) {
FastBitSet[] b = new FastBitSet[numBitVectors];
for (int j=0; j < numBitVectors; j++) {
b[j] = new FastBitSet(BIT_VECTOR_SIZE);
b[j].clear();
}
int vectorIndex =0;
/* Parse input string to obtain the indexes that are set in the bitvector.
* When a toString() is called on a FastBitSet object to serialize it, the serialization
* adds { and } to the beginning and end of the return String.
* Skip "{", "}", ",", " " in the input string.
*/
for(int i=1; i < s.length()-1;) {
char c = s.charAt(i);
i = i + 1;
// Move on to the next bit vector
if (c == '}') {
vectorIndex = vectorIndex + 1;
}
// Encountered a numeric value; Extract out the entire number
if (c >= '0' && c <= '9') {
String t = new String();
t = t + c;
c = s.charAt(i);
i = i + 1;
while (c != ',' && c!= '}') {
t = t + c;
c = s.charAt(i);
i = i + 1;
}
int bitIndex = Integer.parseInt(t);
assert(bitIndex >= 0);
assert(vectorIndex < numBitVectors);
b[vectorIndex].set(bitIndex);
if (c == '}') {
vectorIndex = vectorIndex + 1;
}
}
}
return b;
}
private int generateHash(long v, int hashNum) {
int mod = (1<<BIT_VECTOR_SIZE) - 1;
long tempHash = a[hashNum] * v + b[hashNum];
tempHash %= mod;
int hash = (int) tempHash;
/* Hash function should map the long value to 0...2^L-1.
* Hence hash value has to be non-negative.
*/
if (hash < 0) {
hash = hash + mod;
}
return hash;
}
private int generateHashForPCSA(long v) {
return generateHash(v, 0);
}
public void addToEstimator(long v) {
/* Update summary bitVector :
* Generate hash value of the long value and mod it by 2^bitVectorSize-1.
* In this implementation bitVectorSize is 31.
*/
for (int i = 0; i<numBitVectors; i++) {
int hash = generateHash(v,i);
int index;
// Find the index of the least significant bit that is 1
for (index=0; index<BIT_VECTOR_SIZE; index++) {
if (hash % 2 != 0) {
break;
}
hash = hash >> 1;
}
// Set bitvector[index] := 1
bitVector[i].set(index);
}
}
public void addToEstimatorPCSA(long v) {
int hash = generateHashForPCSA(v);
int rho = hash/numBitVectors;
int index;
// Find the index of the least significant bit that is 1
for (index=0; index<BIT_VECTOR_SIZE; index++) {
if (rho % 2 != 0) {
break;
}
rho = rho >> 1;
}
// Set bitvector[index] := 1
bitVector[hash%numBitVectors].set(index);
}
public void addToEstimator(double d) {
int v = new Double(d).hashCode();
addToEstimator(v);
}
public void addToEstimatorPCSA(double d) {
int v = new Double(d).hashCode();
addToEstimatorPCSA(v);
}
public void addToEstimator(HiveDecimal decimal) {
int v = decimal.hashCode();
addToEstimator(v);
}
public void addToEstimatorPCSA(HiveDecimal decimal) {
int v = decimal.hashCode();
addToEstimatorPCSA(v);
}
public void mergeEstimators(NumDistinctValueEstimator o) {
// Bitwise OR the bitvector with the bitvector in the agg buffer
for (int i=0; i<numBitVectors; i++) {
bitVector[i].or(o.getBitVector(i));
}
}
public long estimateNumDistinctValuesPCSA() {
double numDistinctValues = 0.0;
long S = 0;
for (int i=0; i < numBitVectors; i++) {
int index = 0;
while (bitVector[i].get(index) && index < BIT_VECTOR_SIZE) {
index = index + 1;
}
S = S + index;
}
numDistinctValues = ((numBitVectors/PHI) * Math.pow(2.0, S/numBitVectors));
return ((long)numDistinctValues);
}
/* We use the Flajolet-Martin estimator to estimate the number of distinct values.FM uses the
* location of the least significant zero as an estimate of log2(phi*ndvs).
*/
public long estimateNumDistinctValues() {
int sumLeastSigZero = 0;
double avgLeastSigZero;
double numDistinctValues;
for (int i=0; i< numBitVectors; i++) {
int leastSigZero = bitVector[i].nextClearBit(0);
sumLeastSigZero += leastSigZero;
}
avgLeastSigZero =
(double)(sumLeastSigZero/(numBitVectors * 1.0)) - (Math.log(PHI)/Math.log(2.0));
numDistinctValues = Math.pow(2.0, avgLeastSigZero);
return ((long)(numDistinctValues));
}
}