搜云库技术团队
一、Mycat分片路由原理
我们先来看下面的一个SQL在Mycat里面是如何执行的：
select * from travelrecord where id in(5000001, 10000001);
有3个分片dn1,dn2,dn3, id=5000001这条数据在dn2上，id=10000001这条数据在dn3上。
查询时可能有出现的问题：
1）全部扫描一遍dn1 dn2 dn3，结果导致性能浪费。
2）只扫描某个片。漏掉数据的情况。
总结：
不能多扫——>性能不足
也不能少——>漏掉数据
那么Mycat是如何解决上面的问题的呢？
Mycat使用Druid的DruidParser作为分析器/解析器，解析的过程主要有Visitor和Statement两个阶段
说明：
1）Visitor过程，解析出如下属性：
哪一张表
字段列表
条件信息
什么样的SQL
解析出以上信息以后就可以根据schema.xml和rule.xml文件确认要去哪个分片上进行DML操作了
2）Statement过程转化：转化后知道执行的是什么样的SQL（增删改查）
3）改写SQL
通过查询条件可以知道要查询的数据都在哪些分片上
Dn2， id= 5000001
Dn3， id= 100000001
所以SQL被改写成以下的形式：
select * from travelrecord where id = 5000001；（dn2执行）
select * from travelrecord where id = 10000001；（dn3执行）
4）分别在分片dn2，dn3上执行第 3）步改写的SQL，然后把从dn2，dn3上得到的结果进行拼装就是最终的结果了
备注：
多表关联查询的时候，visitor会遍历出所有的表，然后去各个分片上去获取结果，同时把结果缓存起来，最后根据关联查询计算出结果。
确定分片的过程：首先看where条件里面是否含有分片字段，有就根据分片字段的值结合schema.xml、rule.xml的值确定是哪个分片。当不能确定在哪一个分片上的时候，mycat会到所有的分片上去找
二、Mycat常用分片规则
1、 时间类：按天分片、自然月分片、单月小时分片
2、 哈希类：Hash固定分片、日期范围Hash分片、截取数字Hash求模范围分片、截取数字Hash分片、一致性Hash分片
3、 取模类：取模分片、取模范围分片、范围求模分片
4、 其他类：枚举分片、范围约定分片、应用指定分片、冷热数据分片
下面基于源码来介绍Mycat的常用分片规则，源码地址
三、Mycat常用分片规则介绍
说明：分片规则都定义在rule.xml文件里面
<!--
tableRule标签:定义table分片策略
-->
<tableRule name="rule1">
<!--
rule标签:策略定义标签
-->
<rule>
<!--
columns标签:对应的分片字段
-->
<columns>id</columns>
<!--
algorithm标签:tableRule分片策略对应的function名称
-->
<algorithm>func1</algorithm>
</rule>
</tableRule>
<!-- 定义分片函数 -->
<function name="func1" class="io.mycat.route.function.PartitionByLong">
<property name="partitionCount">1,1,2,3,1</property><!-- 分片数 -->
<property name="partitionLength">128,128,128,128,128</property><!-- 分片长度 -->
</function>
1. 自动范围分片
在rule.xml里面的配置：
<tableRule name="auto-sharding-long">
<rule>
<columns>id</columns>
<algorithm>rang-long</algorithm>
</rule>
</tableRule>
<function name="rang-long"
class="io.mycat.route.function.AutoPartitionByLong">
<property name="mapFile">autopartition-long.txt</property>
</function>
说明：
有3个分片，第1个分片存储的是1-500000的数据，第2个分片存储的是500001-1000000的数据，第3个分片存储的是1000001-1500000的数据
insert into employee(id, name) value(1,Tom);在第1个分片
insert into employee(id, name) value(500002,Jack);在第2个分片
insert into employee(id, name) value(1000002,Lucy);在第3个分片
对应代码：
package io.mycat.route.function;
import java.io.BufferedReader;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.util.HashSet;
import java.util.LinkedList;
import java.util.Set;
import io.mycat.config.model.rule.RuleAlgorithm;
/**
* auto partition by Long ,can be used in auto increment primary key partition
*
* @author wuzhi
*/
public class AutoPartitionByLong extends AbstractPartitionAlgorithm implements RuleAlgorithm{
private String mapFile;
private LongRange[] longRongs;
private int defaultNode = -1;
@Override
public void init() {
initialize();
}
public void setMapFile(String mapFile) {
this.mapFile = mapFile;
}
@Override
public Integer calculate(String columnValue) {
// columnValue = NumberParseUtil.eliminateQoute(columnValue);
try {
long value = Long.parseLong(columnValue);
Integer rst = null;
for (LongRange longRang : this.longRongs) {
if (value <= longRang.valueEnd && value >= longRang.valueStart) {
return longRang.nodeIndx;
}
}
//数据超过范围，暂时使用配置的默认节点
if (rst == null && defaultNode >= 0) {
return defaultNode;
}
return rst;
} catch (NumberFormatException e){
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please eliminate any quote and non number within it.").toString(),e);
}
}
@Override
public Integer[] calculateRange(String beginValue, String endValue) {
return AbstractPartitionAlgorithm.calculateSequenceRange(this, beginValue, endValue);
}
@Override
public int getPartitionNum() {
// int nPartition = longRongs.length;
/*
* fix #1284 这里的统计应该统计Range的nodeIndex的distinct总数
*/
Set<Integer> distNodeIdxSet = new HashSet<Integer>();
for(LongRange range : longRongs) {
distNodeIdxSet.add(range.nodeIndx);
}
int nPartition = distNodeIdxSet.size();
return nPartition;
}
private void initialize() {
BufferedReader in = null;
try {
// FileInputStream fin = new FileInputStream(new File(fileMapPath));
InputStream fin = this.getClass().getClassLoader()
.getResourceAsStream(mapFile);
if (fin == null) {
throw new RuntimeException("can't find class resource file "
+ mapFile);
}
in = new BufferedReader(new InputStreamReader(fin));
LinkedList<LongRange> longRangeList = new LinkedList<LongRange>();
for (String line = null; (line = in.readLine()) != null;) {
line = line.trim();
if (line.startsWith("#") || line.startsWith("//")) {
continue;
}
int ind = line.indexOf('=');
if (ind < 0) {
System.out.println(" warn: bad line int " + mapFile + " :"
+ line);
continue;
}
String pairs[] = line.substring(0, ind).trim().split("-");
long longStart = NumberParseUtil.parseLong(pairs[0].trim());
long longEnd = NumberParseUtil.parseLong(pairs[1].trim());
int nodeId = Integer.parseInt(line.substring(ind + 1)
.trim());
longRangeList
.add(new LongRange(nodeId, longStart, longEnd));
}
longRongs = longRangeList.toArray(new LongRange[longRangeList
.size()]);
} catch (Exception e) {
if (e instanceof RuntimeException) {
throw (RuntimeException) e;
} else {
throw new RuntimeException(e);
}
} finally {
try {
in.close();
} catch (Exception e2) {
}
}
}
public int getDefaultNode() {
return defaultNode;
}
public void setDefaultNode(int defaultNode) {
this.defaultNode = defaultNode;
}
static class LongRange {
public final int nodeIndx;
public final long valueStart;
public final long valueEnd;
public LongRange(int nodeIndx, long valueStart, long valueEnd) {
super();
this.nodeIndx = nodeIndx;
this.valueStart = valueStart;
this.valueEnd = valueEnd;
}
}
}
View Code
2. 枚举分片
把数据分类存储
在rule.xml里面的配置：
<tableRule name="sharding-by-intfile">
<rule>
<columns>sharding_id</columns>
<algorithm>hash-int</algorithm>
</rule>
</tableRule>
<function name="hash-int"
class="io.mycat.route.function.PartitionByFileMap">
<property name="mapFile">partition-hash-int.txt</property>
<property name="defaultNode">0</property> <!-- 找不到分片时设置容错规则，把数据插入到默认分片0里面 -->
</function>
说明：找不到分片时设置容错规则，把数据插入到默认分片0里面
对应代码：
package io.mycat.route.function;
import java.io.BufferedReader;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Map;
import java.util.Set;
import io.mycat.config.model.rule.RuleAlgorithm;
/**
*
* @author mycat
*/
public class PartitionByFileMap extends AbstractPartitionAlgorithm implements RuleAlgorithm {
private String mapFile;
private Map<Object, Integer> app2Partition;
/**
* Map<Object, Integer> app2Partition中key值的类型：默认值为0，0表示Integer，非零表示String
*/
private int type;
/**
* 默认节点在map中的key
*/
private static final String DEFAULT_NODE = "DEFAULT_NODE";
/**
* 默认节点:小于0表示不设置默认节点，大于等于0表示设置默认节点
*
* 默认节点的作用：枚举分片时，如果碰到不识别的枚举值，就让它路由到默认节点
* 如果不配置默认节点（defaultNode值小于0表示不配置默认节点），碰到
* 不识别的枚举值就会报错，
* like this：can't find datanode for sharding column:column_name val:ffffffff
*/
private int defaultNode = -1;
@Override
public void init() {
initialize();
}
public void setMapFile(String mapFile) {
this.mapFile = mapFile;
}
public void setType(int type) {
this.type = type;
}
public void setDefaultNode(int defaultNode) {
this.defaultNode = defaultNode;
}
@Override
public Integer calculate(String columnValue) {
try {
Object value = columnValue;
if (type == 0) {
value = Integer.valueOf(columnValue);
}
Integer rst = null;
Integer pid = app2Partition.get(value);
if (pid != null) {
rst = pid;
} else {
rst = app2Partition.get(DEFAULT_NODE);
}
return rst;
} catch (NumberFormatException e){
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please check if the format satisfied.").toString(),e);
}
}
@Override
public int getPartitionNum() {
Set<Integer> set = new HashSet<Integer>(app2Partition.values());
int count = set.size();
return count;
}
private void initialize() {
BufferedReader in = null;
try {
// FileInputStream fin = new FileInputStream(new File(fileMapPath));
InputStream fin = this.getClass().getClassLoader()
.getResourceAsStream(mapFile);
if (fin == null) {
throw new RuntimeException("can't find class resource file "
+ mapFile);
}
in = new BufferedReader(new InputStreamReader(fin));
app2Partition = new HashMap<Object, Integer>();
for (String line = null; (line = in.readLine()) != null;) {
line = line.trim();
if (line.startsWith("#") || line.startsWith("//")) {
continue;
}
int ind = line.indexOf('=');
if (ind < 0) {
continue;
}
try {
String key = line.substring(0, ind).trim();
int pid = Integer.parseInt(line.substring(ind + 1).trim());
if(type == 0) {
app2Partition.put(Integer.parseInt(key), pid);
} else {
app2Partition.put(key, pid);
}
} catch (Exception e) {
}
}
//设置默认节点
if(defaultNode >= 0) {
app2Partition.put(DEFAULT_NODE, defaultNode);
}
} catch (Exception e) {
if (e instanceof RuntimeException) {
throw (RuntimeException) e;
} else {
throw new RuntimeException(e);
}
} finally {
try {
in.close();
} catch (Exception e2) {
}
}
}
}
View Code
3. Hash固定分片（固定分片Hash算法）
固定分片Hash算法，最多有1024个分片
在rule.xml里面的配置：
<tableRule name="rule1">
<!--
rule标签:策略定义标签
-->
<rule>
<!--
columns标签:对应的分片字段
-->
<columns>id</columns>
<!--
algorithm标签:tableRule分片策略对应的function名称
-->
<algorithm>func1</algorithm>
</rule>
</tableRule>
<function name="func1" class="io.mycat.route.function.PartitionByLong">
<property name="partitionCount">1,1,2,3,1</property><!-- 分片数 -->
<property name="partitionLength">128,128,128,128,128</property><!-- 分片长度 -->
</function>
说明：
1) partitionCount.length必须等于partitionLength.length
2) sum((partitionCount[i]*partitionLength[j])) === 1024——>partitionCount[0]*partitionLength[0]+partitionCount[1]*partitionLength[1] === 1024
即：1*128+1*128+2*128+3*128+1*128 === 1024
eg：
8个分片表
第1个分片表的下标为0: 0-127
第2个分片表的下标为1: 127-255
……………….
第8个分片表的下标为7: 896-1024
如何确定落在哪个分片上呢？分片id的值与1024取余可确定在哪个分片上：
如id%1024 = 128 则落在第2个分片上
对应代码：
package io.mycat.route.function;
import io.mycat.config.model.rule.RuleAlgorithm;
import io.mycat.route.util.PartitionUtil;
public final class PartitionByLong extends AbstractPartitionAlgorithm implements RuleAlgorithm {
protected int[] count;
protected int[] length;
protected PartitionUtil partitionUtil;
private static int[] toIntArray(String string) {
String[] strs = io.mycat.util.SplitUtil.split(string, ',', true);
int[] ints = new int[strs.length];
for (int i = 0; i < strs.length; ++i) {
ints[i] = Integer.parseInt(strs[i]);
}
return ints;
}
public void setPartitionCount(String partitionCount) {
this.count = toIntArray(partitionCount);
}
public void setPartitionLength(String partitionLength) {
this.length = toIntArray(partitionLength);
}
@Override
public void init() {
partitionUtil = new PartitionUtil(count, length);
}
@Override
public Integer calculate(String columnValue) {
// columnValue = NumberParseUtil.eliminateQoute(columnValue);
try {
long key = Long.parseLong(columnValue);
return partitionUtil.partition(key);
} catch (NumberFormatException e){
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please eliminate any quote and non number within it.").toString(),e);
}
}
@Override
public Integer[] calculateRange(String beginValue, String endValue) {
return AbstractPartitionAlgorithm.calculateSequenceRange(this, beginValue, endValue);
}
// @Override
// public int getPartitionCount() {
// int nPartition = 0;
// for(int i = 0; i < count.length; i++) {
// nPartition += count[i];
// }
// return nPartition;
// }
}
View Code
4. 求模分片
分片字段id%分片数=分片下标
在rule.xml里面的配置：
<tableRule name="mod-long">
<rule>
<columns>id</columns><!--分片字段 -->
<algorithm>mod-long</algorithm>
</rule>
</tableRule>
<function name="mod-long" class="io.mycat.route.function.PartitionByMod">
<!-- how many data nodes -->
<property name="count">3</property><!--分片数 -->
</function>
对应代码：
package io.mycat.route.function;
import java.math.BigInteger;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import io.mycat.config.model.rule.RuleAlgorithm;
/**
* number column partion by Mod operator
* if count is 10 then 0 to 0,21 to 1 (21 % 10 =1)
* @author wuzhih
*
*/
public class PartitionByMod extends AbstractPartitionAlgorithm implements RuleAlgorithm {
private int count;
@Override
public void init() {
}
public void setCount(int count) {
this.count = count;
}
@Override
public Integer calculate(String columnValue) {
// columnValue = NumberParseUtil.eliminateQoute(columnValue);
try {
BigInteger bigNum = new BigInteger(columnValue).abs();
return (bigNum.mod(BigInteger.valueOf(count))).intValue();
} catch (NumberFormatException e){
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please eliminate any quote and non number within it.").toString(),e);
}
}
@Override
public int getPartitionNum() {
int nPartition = this.count;
return nPartition;
}
private static void hashTest() {
PartitionByMod hash=new PartitionByMod();
hash.setCount(11);
hash.init();
int[] bucket=new int[hash.count];
Map<Integer,List<Integer>> hashed=new HashMap<>();
int total=1000_0000;//数据量
int c=0;
for(int i=100_0000;i<total+100_0000;i++){//假设分片键从100万开始
c++;
int h=hash.calculate(Integer.toString(i));
bucket[h]++;
List<Integer> list=hashed.get(h);
if(list==null){
list=new ArrayList<>();
hashed.put(h, list);
}
list.add(i);
}
System.out.println(c+" "+total);
double d=0;
c=0;
int idx=0;
System.out.println("index bucket ratio");
for(int i:bucket){
d+=i/(double)total;
c+=i;
System.out.println(idx+++" "+i+" "+(i/(double)total));
}
System.out.println(d+" "+c);
System.out.println("****************************************************");
rehashTest(hashed.get(0));
}
private static void rehashTest(List<Integer> partition) {
PartitionByMod hash=new PartitionByMod();
hash.count=110;//分片数
hash.init();
int[] bucket=new int[hash.count];
int total=partition.size();//数据量
int c=0;
for(int i:partition){//假设分片键从100万开始
c++;
int h=hash.calculate(Integer.toString(i));
bucket[h]++;
}
System.out.println(c+" "+total);
c=0;
int idx=0;
System.out.println("index bucket ratio");
for(int i:bucket){
c+=i;
System.out.println(idx+++" "+i+" "+(i/(double)total));
}
}
public static void main(String[] args) {
// hashTest();
PartitionByMod partitionByMod = new PartitionByMod();
partitionByMod.count=8;
partitionByMod.calculate("\"6\"");
partitionByMod.calculate("\'6\'");
}
}
View Code
5. 自然月分片
按照自然月的方式进行分片
在rule.xml里面的配置：
<tableRule name="sharding-by-month">
<rule>
<columns>create_time</columns>
<algorithm>partbymonth</algorithm>
</rule>
</tableRule>
<function name="partbymonth"
class="io.mycat.route.function.PartitionByMonth">
<property name="dateFormat">yyyy-MM-dd</property>
<property name="sBeginDate">2015-01-01</property>
</function>
说明：
如果月份超过了分片数，则通过设置sEndDated的值来解决
如有3个分片，分别插入2015-01-12，2015-02-12、2015-03-12、2016-11-12，月份超过了分片数，此时设置sEndDated= 2015-04-12表示4个月放一个分片，如下可知2016-11-12在分片2上
| 分片0 |
分片1 |
分片2 |
|
1月 |
5月 |
9月 |
|
2月 |
6月 |
10月 |
|
3月 |
7月 |
11月 |
|
4月 |
8月 |
12月 |
对应代码：
package io.mycat.route.function;
import java.text.ParseException;
import java.text.SimpleDateFormat;
import java.util.ArrayList;
import java.util.Calendar;
import java.util.Collections;
import java.util.List;
import io.mycat.config.model.rule.RuleAlgorithm;
import org.apache.log4j.Logger;
/**
* 例子 按月份列分区 ，每个自然月一个分片，格式 between操作解析的范例
*
* @author wzh
*
*/
public class PartitionByMonth extends AbstractPartitionAlgorithm implements
RuleAlgorithm {
private static final Logger LOGGER = Logger.getLogger(PartitionByDate.class);
private String sBeginDate;
private String dateFormat;
private String sEndDate;
private Calendar beginDate;
private Calendar endDate;
private int nPartition;
private ThreadLocal<SimpleDateFormat> formatter;
@Override
public void init() {
try {
beginDate = Calendar.getInstance();
beginDate.setTime(new SimpleDateFormat(dateFormat)
.parse(sBeginDate));
formatter = new ThreadLocal<SimpleDateFormat>() {
@Override
protected SimpleDateFormat initialValue() {
return new SimpleDateFormat(dateFormat);
}
};
if(sEndDate!=null&&!sEndDate.equals("")) {
endDate = Calendar.getInstance();
endDate.setTime(new SimpleDateFormat(dateFormat).parse(sEndDate));
nPartition = ((endDate.get(Calendar.YEAR) - beginDate.get(Calendar.YEAR)) * 12
+ endDate.get(Calendar.MONTH) - beginDate.get(Calendar.MONTH)) + 1;
if (nPartition <= 0) {
throw new java.lang.IllegalArgumentException("Incorrect time range for month partitioning!");
}
} else {
nPartition = -1;
}
} catch (ParseException e) {
throw new java.lang.IllegalArgumentException(e);
}
}
/**
* For circulatory partition, calculated value of target partition needs to be
* rotated to fit the partition range
*/
private int reCalculatePartition(int targetPartition) {
/**
* If target date is previous of start time of partition setting, shift
* the delta range between target and start date to be positive value
*/
if (targetPartition < 0) {
targetPartition = nPartition - (-targetPartition) % nPartition;
}
if (targetPartition >= nPartition) {
targetPartition = targetPartition % nPartition;
}
return targetPartition;
}
@Override
public Integer calculate(String columnValue) {
try {
int targetPartition;
Calendar curTime = Calendar.getInstance();
curTime.setTime(formatter.get().parse(columnValue));
targetPartition = ((curTime.get(Calendar.YEAR) - beginDate.get(Calendar.YEAR))
* 12 + curTime.get(Calendar.MONTH)
- beginDate.get(Calendar.MONTH));
/**
* For circulatory partition, calculated value of target partition needs to be
* rotated to fit the partition range
*/
if (nPartition > 0) {
targetPartition = reCalculatePartition(targetPartition);
}
return targetPartition;
} catch (ParseException e) {
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please check if the format satisfied.").toString(),e);
}
}
@Override
public Integer[] calculateRange(String beginValue, String endValue) {
try {
int startPartition, endPartition;
Calendar partitionTime = Calendar.getInstance();
SimpleDateFormat format = new SimpleDateFormat(dateFormat);
partitionTime.setTime(format.parse(beginValue));
startPartition = ((partitionTime.get(Calendar.YEAR) - beginDate.get(Calendar.YEAR))
* 12 + partitionTime.get(Calendar.MONTH)
- beginDate.get(Calendar.MONTH));
partitionTime.setTime(format.parse(endValue));
endPartition = ((partitionTime.get(Calendar.YEAR) - beginDate.get(Calendar.YEAR))
* 12 + partitionTime.get(Calendar.MONTH)
- beginDate.get(Calendar.MONTH));
List<Integer> list = new ArrayList<>();
while (startPartition <= endPartition) {
Integer nodeValue = reCalculatePartition(startPartition);
if (Collections.frequency(list, nodeValue) < 1)
list.add(nodeValue);
startPartition++;
}
int size = list.size();
return (list.toArray(new Integer[size]));
} catch (ParseException e) {
LOGGER.error("error",e);
return new Integer[0];
}
}
@Override
public int getPartitionNum() {
int nPartition = this.nPartition;
return nPartition;
}
public void setsBeginDate(String sBeginDate) {
this.sBeginDate = sBeginDate;
}
public void setDateFormat(String dateFormat) {
this.dateFormat = dateFormat;
}
public void setsEndDate(String sEndDate) {
this.sEndDate = sEndDate;
}
}
View Code
6. 匹配求模分片
根据prefixLength截取n个字符并charAt(i)每个字符的值进行累加得到一个整数，然后和分区长度patternValue进行求模，得出的值就是分区编号。
在rule.xml里面的配置：
<tableRule name="partitionbyprefixpattern">
<rule>
<columns>id</columns>
<algorithm>partitionbyprefixpattern</algorithm>
</rule>
</tableRule>
<function name="partitionbyprefixpattern" class="io.mycat.route.function.PartitionByPrefixPattern">
<property name="patternValue">3</property> <!-- 分区长度/分区数量 -->
<property name="prefixLength">6</property> <!-- 截取多少字符串 -->
</function>
说明：
有下面这种类型的数据
年月+大区+流水编号
201801 01 10001
就可以采用匹配求模分片，把分片字段columns取前6个字符串201801并charAt(i)每个字符的值进行累加得到一个整数，然后和分区长度3进行求模，得出的值就是分区编号
对应代码：
package io.mycat.route.function;
import java.io.BufferedReader;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.util.HashSet;
import java.util.LinkedList;
import java.util.Set;
import io.mycat.config.model.rule.RuleAlgorithm;
import io.mycat.route.function.AutoPartitionByLong.LongRange;
/**
* partition by Prefix length ,can be used in String partition
*
* @author hexiaobin
*/
public class PartitionByPrefixPattern extends AbstractPartitionAlgorithm implements RuleAlgorithm {
private static final int PARTITION_LENGTH = 1024;
private int patternValue = PARTITION_LENGTH;// 分区长度，取模数值(默认为1024)
private int prefixLength;// 字符前几位进行ASCII码取和
private String mapFile;
private LongRange[] longRongs;
@Override
public void init() {
initialize();
}
public void setMapFile(String mapFile) {
this.mapFile = mapFile;
}
public void setPatternValue(int patternValue) {
this.patternValue = patternValue;
}
public void setPrefixLength(int prefixLength) {
this.prefixLength = prefixLength;
}
@Override
public Integer calculate(String columnValue) {
try {
int Length = Integer.valueOf(prefixLength);
Length = columnValue.length() < Length ? columnValue.length() : Length;
int sum = 0;
for (int i = 0; i < Length; i++) {
sum = sum + columnValue.charAt(i);
}
Integer rst = null;
for (LongRange longRang : this.longRongs) {
long hash = sum % patternValue;
if (hash <= longRang.valueEnd && hash >= longRang.valueStart) {
return longRang.nodeIndx;
}
}
return rst;
} catch (NumberFormatException e){
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please eliminate any quote and non number within it.").toString(),e);
}
}
@Override
public int getPartitionNum() {
// int nPartition = this.longRongs.length;
/*
* fix #1284 这里的统计应该统计Range的nodeIndex的distinct总数
*/
Set<Integer> distNodeIdxSet = new HashSet<Integer>();
for(LongRange range : longRongs) {
distNodeIdxSet.add(range.nodeIndx);
}
int nPartition = distNodeIdxSet.size();
return nPartition;
}
private void initialize() {
BufferedReader in = null;
try {
// FileInputStream fin = new FileInputStream(new File(fileMapPath));
InputStream fin = this.getClass().getClassLoader()
.getResourceAsStream(mapFile);
if (fin == null) {
throw new RuntimeException("can't find class resource file "
+ mapFile);
}
in = new BufferedReader(new InputStreamReader(fin));
LinkedList<LongRange> longRangeList = new LinkedList<LongRange>();
for (String line = null; (line = in.readLine()) != null;) {
line = line.trim();
if (line.startsWith("#") || line.startsWith("//")) {
continue;
}
int ind = line.indexOf('=');
if (ind < 0) {
System.out.println(" warn: bad line int " + mapFile + " :"
+ line);
continue;
}
String pairs[] = line.substring(0, ind).trim().split("-");
long longStart = NumberParseUtil.parseLong(pairs[0].trim());
long longEnd = NumberParseUtil.parseLong(pairs[1].trim());
int nodeId = Integer.parseInt(line.substring(ind + 1)
.trim());
longRangeList
.add(new LongRange(nodeId, longStart, longEnd));
}
longRongs = longRangeList.toArray(new LongRange[longRangeList
.size()]);
} catch (Exception e) {
if (e instanceof RuntimeException) {
throw (RuntimeException) e;
} else {
throw new RuntimeException(e);
}
} finally {
try {
in.close();
} catch (Exception e2) {
}
}
}
static class LongRange {
public final int nodeIndx;
public final long valueStart;
public final long valueEnd;
public LongRange(int nodeIndx, long valueStart, long valueEnd) {
super();
this.nodeIndx = nodeIndx;
this.valueStart = valueStart;
this.valueEnd = valueEnd;
}
}
}
View Code
7. 冷热数据分片
根据日期查询日志数据冷热数据分布 ，最近 n 个月的到实时交易库查询，超过 n 个月的按照 m 天分片。
在rule.xml里面的配置：
<tableRule name="sharding-by-date">
<rule>
<columns>create_time</columns>
<algorithm>sharding-by-hotdate</algorithm>
</rule>
</tableRule>
<function name="sharding-by-hotdate" class="org.opencloudb.route.function.PartitionByHotDate">
<property name="dateFormat">yyyy-MM-dd</property> <!-- 定义日期格式 -->
<property name="sLastDay">30</property> <!-- 热库存储多少天数据 -->
<property name="sPartionDay">30</property> <!-- 超过热库期限的数据按照多少天来分片 -->
</function>
对应代码：
package io.mycat.route.function;
import java.text.ParseException;
import java.text.SimpleDateFormat;
import java.util.Calendar;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import io.mycat.config.model.rule.RuleAlgorithm;
/**
* 根据日期查询日志数据 冷热数据分布 ，最近n个月的到实时交易库查询，超过n个月的按照m天分片
*
* @author sw
*
* <tableRule name="sharding-by-date">
<rule>
<columns>create_time</columns>
<algorithm>sharding-by-hotdate</algorithm>
</rule>
</tableRule>
<function name="sharding-by-hotdate" class="org.opencloudb.route.function.PartitionByHotDate">
<property name="dateFormat">yyyy-MM-dd</property>
<property name="sLastDay">10</property>
<property name="sPartionDay">30</property>
</function>
*/
public class PartitionByHotDate extends AbstractPartitionAlgorithm implements RuleAlgorithm {
private static final Logger LOGGER = LoggerFactory.getLogger(PartitionByHotDate.class);
private String dateFormat;
private String sLastDay;
private String sPartionDay;
private long sLastTime;
private long partionTime;
private ThreadLocal<SimpleDateFormat> formatter;
private long beginDate;
private static final long oneDay = 86400000;
@Override
public void init() {
try {
formatter = new ThreadLocal<SimpleDateFormat>() {
@Override
protected SimpleDateFormat initialValue() {
return new SimpleDateFormat(dateFormat);
}
};
sLastTime = Integer.valueOf(sLastDay);
partionTime = Integer.parseInt(sPartionDay) * oneDay;
} catch (Exception e) {
throw new java.lang.IllegalArgumentException(e);
}
}
@Override
public Integer calculate(String columnValue) {
Integer targetPartition = -1;
try {
long targetTime = formatter.get().parse(columnValue).getTime();
Calendar now = Calendar.getInstance();
long nowTime = now.getTimeInMillis();
beginDate = nowTime - sLastTime * oneDay;
long diffDays = (nowTime - targetTime) / (1000 * 60 * 60 * 24) + 1;
if(diffDays-sLastTime <= 0 || diffDays<0 ){
targetPartition = 0;
}else{
targetPartition = (int) ((beginDate - targetTime) / partionTime) + 1;
}
LOGGER.debug("PartitionByHotDate calculate for " + columnValue + " return " + targetPartition);
return targetPartition;
} catch (ParseException e) {
throw new IllegalArgumentException(new StringBuilder().append("columnValue:").append(columnValue).append(" Please check if the format satisfied.").toString(),e);
}
}
@Override
public Integer[] calculateRange(String beginValue, String endValue) {
Integer[] targetPartition = null;
try {
long startTime = formatter.get().parse(beginValue).getTime();
long endTime = formatter.get().parse(endValue).getTime();
Calendar now = Calendar.getInstance();
long nowTime = now.getTimeInMillis();
long limitDate = nowTime - sLastTime * oneDay;
long diffDays = (nowTime - startTime) / (1000 * 60 * 60 * 24) + 1;
if(diffDays-sLastTime <= 0 || diffDays<0 ){
Integer [] re = new Integer[1];
re[0] = 0;
targetPartition = re ;
}else{
Integer [] re = null;
Integer begin = 0, end = 0;
end = this.calculate(beginValue);
boolean hasLimit = false;
if(endTime-limitDate > 0){
endTime = limitDate;
hasLimit = true;
}
begin = this.calculate(formatter.get().format(endTime));
if(begin == null || end == null){
return re;
}
if (end >= begin) {
int len = end-begin+1;
if(hasLimit){
re = new Integer[len+1];
re[0] = 0;
for(int i =0;i<len;i++){
re[i+1]=begin+i;
}
}else{
re = new Integer[len];
for(int i=0;i<len;i++){
re[i]=begin+i;
}
}
return re;
}else{
return re;
}
}
} catch (ParseException e) {
throw new IllegalArgumentException(new StringBuilder().append("endValue:").append(endValue).append(" Please check if the format satisfied.").toString(),e);
}
return targetPartition;
}
public void setsPartionDay(String sPartionDay) {
this.sPartionDay = sPartionDay;
}
public void setDateFormat(String dateFormat) {
this.dateFormat = dateFormat;
}
public String getsLastDay() {
return sLastDay;
}
public void setsLastDay(String sLastDay) {
this.sLastDay = sLastDay;
}
}
View Code
8. 一致性哈希分片
1）首先求出mysql服务器（节点）的哈希值，并将其配置到0～2^32的圆（continuum）上。
2）为每台mysql服务器物理节点虚拟出多个虚拟节点，并计算hash值映射到相同的圆上。
3）然后从数据映射到的mysql服务器虚拟节点的位置开始顺时针查找，将数据保存到找到的第一个mysql服务器上。如果超过232仍然找不到服务器，就会保存到第一台mysql服务器上。
特点：解决数据均匀分布
在rule.xml里面的配置：
<tableRule name="sharding-by-murmur">
<rule>
<columns>id</columns>
<algorithm>murmur</algorithm>
</rule>
</tableRule>
<function name="murmur"
class="io.mycat.route.function.PartitionByMurmurHash">
<property name="seed">0</property><!-- 默认是0 -->
<property name="count">2</property><!-- 要分片的数据库节点数量，必须指定，否则没法分片 -->
<property name="virtualBucketTimes">160</property><!-- 一个实际的数据库节点被映射为这么多虚拟节点，默认是160倍，也就是虚拟节点数是物理节点数的160倍 -->
<!-- <property name="weightMapFile">weightMapFile</property> 节点的权重，没有指定权重的节点默认是1。以properties文件的格式填写，以从0开始到count-1的整数值也就是节点索引为key，以节点权重值为值。所有权重值必须是正整数，否则以1代替 -->
<!-- <property name="bucketMapPath">/etc/mycat/bucketMapPath</property>
用于测试时观察各物理节点与虚拟节点的分布情况，如果指定了这个属性，会把虚拟节点的murmur hash值与物理节点的映射按行输出到这个文件，没有默认值，如果不指定，就不会输出任何东西 -->
</function>
对应代码：
package io.mycat.route.function;
import java.io.BufferedReader;
import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileNotFoundException;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.io.OutputStream;
import java.nio.charset.Charset;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.SortedMap;
import java.util.TreeMap;
import com.google.common.hash.HashFunction;
import com.google.common.hash.Hashing;
import io.mycat.config.model.rule.RuleAlgorithm;
import io.mycat.util.exception.MurmurHashException;
/**
* consistancy hash, murmur hash
* implemented by Guava
* @author wuzhih
*
*/
public class PartitionByMurmurHash extends AbstractPartitionAlgorithm implements RuleAlgorithm {
private static final int DEFAULT_VIRTUAL_BUCKET_TIMES=160;
private static final int DEFAULT_WEIGHT=1;
private static final Charset DEFAULT_CHARSET=Charset.forName("UTF-8");
private int seed;
private int count;
private int virtualBucketTimes=DEFAULT_VIRTUAL_BUCKET_TIMES;
private Map<Integer,Integer> weightMap=new HashMap<>();
// private String bucketMapPath;
private HashFunction hash;
private SortedMap<Integer,Integer> bucketMap;
@Override
public void init() {
try{
bucketMap=new TreeMap<>();
// boolean serializableBucketMap=bucketMapPath!=null && bucketMapPath.length()>0;
// if(serializableBucketMap){
// File bucketMapFile=new File(bucketMapPath);
// if(bucketMapFile.exists() && bucketMapFile.length()>0){
// loadBucketMapFile();
// return;
// }
// }
generateBucketMap();
// if(serializableBucketMap){
// storeBucketMap();
// }
}catch(Exception e){
throw new MurmurHashException(e);
}
}
private void generateBucketMap(){
hash=Hashing.murmur3_32(seed);//计算一致性哈希的对象
for(int i=0;i<count;i++){//构造一致性哈希环，用TreeMap表示
StringBuilder hashName=new StringBuilder("SHARD-").append(i);
for(int n=0,shard=virtualBucketTimes*getWeight(i);n<shard;n++){
bucketMap.put(hash.hashUnencodedChars(hashName.append("-NODE-").append(n)).asInt(),i);
}
}
weightMap=null;
}
// private void storeBucketMap() throws IOException{
// try(OutputStream store=new FileOutputStream(bucketMapPath)){
// Properties props=new Properties();
// for(Map.Entry entry:bucketMap.entrySet()){
// props.setProperty(entry.getKey().toString(), entry.getValue().toString());
// }
// props.store(store,null);
// }
// }
// private void loadBucketMapFile() throws FileNotFoundException, IOException{
// try(InputStream in=new FileInputStream(bucketMapPath)){
// Properties props=new Properties();
// props.load(in);
// for(Map.Entry entry:props.entrySet()){
// bucketMap.put(Integer.parseInt(entry.getKey().toString()), Integer.parseInt(entry.getValue().toString()));
// }
// }
// }
/**
* 得到桶的权重，桶就是实际存储数据的DB实例
* 从0开始的桶编号为key，权重为值，权重默认为1。
* 键值必须都是整数
* @param bucket
* @return
*/
private int getWeight(int bucket){
Integer w=weightMap.get(bucket);
if(w==null){
w=DEFAULT_WEIGHT;
}
return w;
}
/**
* 创建murmur_hash对象的种子，默认0
* @param seed
*/
public void setSeed(int seed){
this.seed=seed;
}
/**
* 节点的数量
* @param count
*/
public void setCount(int count) {
this.count = count;
}
/**
* 虚拟节点倍数，virtualBucketTimes*count就是虚拟结点数量
* @param virtualBucketTimes
*/
public void setVirtualBucketTimes(int virtualBucketTimes){
this.virtualBucketTimes=virtualBucketTimes;
}
/**
* 节点的权重，没有指定权重的节点默认是1。以properties文件的格式填写，以从0开始到count-1的整数值也就是节点索引为key，以节点权重值为值。
* 所有权重值必须是正整数，否则以1代替
* @param weightMapPath
* @throws IOException
* @throws
*/
public void setWeightMapFile(String weightMapPath) throws IOException{
Properties props=new Properties();
try(BufferedReader reader=new BufferedReader(new InputStreamReader(this.getClass().getClassLoader().getResourceAsStream(weightMapPath), DEFAULT_CHARSET))){
props.load(reader);
for(Map.Entry entry:props.entrySet()){
int weight=Integer.parseInt(entry.getValue().toString());
weightMap.put(Integer.parseInt(entry.getKey().toString()), weight>0?weight:1);
}
}
}
// /**
// * 保存一致性hash的虚拟节点文件路径。
// * 如果这个文件不存在或是空文件就按照指定的count, weightMapFile等构造新的MurmurHash数据结构并保存到这个路径的文件里。
// * 如果这个文件已存在且不是空文件就加载这个文件里的内容作为MurmurHash数据结构，此时其它参数都忽略。
// * 除第一次以外在之后增加节点时可以直接修改这个文件，不过不推荐这么做。如果节点数量变化了，推荐删除这个文件。
// * 可以不指定这个路径，不指定路径时不会保存murmur hash
// * @param bucketMapPath
// */
// public void setBucketMapPath(String bucketMapPath){
// this.bucketMapPath=bucketMapPath;
// }
@Override
public Integer calculate(String columnValue) {
SortedMap<Integer, Integer> tail = bucketMap.tailMap(hash.hashUnencodedChars(columnValue).asInt());
if (tail.isEmpty()) {
return bucketMap.get(bucketMap.firstKey());
}
return tail.get(tail.firstKey());
}
@Override
public int getPartitionNum() {
int nPartition = this.count;
return nPartition;
}
private static void hashTest() throws IOException{
PartitionByMurmurHash hash=new PartitionByMurmurHash();
hash.count=10;//分片数
hash.init();
int[] bucket=new int[hash.count];
Map<Integer,List<Integer>> hashed=new HashMap<>();
int total=1000_0000;//数据量
int c=0;
for(int i=100_0000;i<total+100_0000;i++){//假设分片键从100万开始
c++;
int h=hash.calculate(Integer.toString(i));
bucket[h]++;
List<Integer> list=hashed.get(h);
if(list==null){
list=new ArrayList<>();
hashed.put(h, list);
}
list.add(i);
}
System.out.println(c+" "+total);
double d=0;
c=0;
int idx=0;
System.out.println("index bucket ratio");
for(int i:bucket){
d+=i/(double)total;
c+=i;
System.out.println(idx+++" "+i+" "+(i/(double)total));
}
System.out.println(d+" "+c);
Properties props=new Properties();
for(Map.Entry entry:hash.bucketMap.entrySet()){
props.setProperty(entry.getKey().toString(), entry.getValue().toString());
}
ByteArrayOutputStream out=new ByteArrayOutputStream();
props.store(out, null);
props.clear();
props.load(new ByteArrayInputStream(out.toByteArray()));
System.out.println(props);
System.out.println("****************************************************");
// rehashTest(hashed.get(0));
}
private static void rehashTest(List<Integer> partition){
PartitionByMurmurHash hash=new PartitionByMurmurHash();
hash.count=12;//分片数
hash.init();
int[] bucket=new int[hash.count];
int total=partition.size();//数据量
int c=0;
for(int i:partition){//假设分片键从100万开始
c++;
int h=hash.calculate(Integer.toString(i));
bucket[h]++;
}
System.out.println(c+" "+total);
c=0;
int idx=0;
System.out.println("index bucket ratio");
for(int i:bucket){
c+=i;
System.out.println(idx+++" "+i+" "+(i/(double)total));
}
}
public static void main(String[] args) throws IOException {
hashTest();
}
}
View Code
来源：https://www.cnblogs.com/leeSmall/category/1103208.html
