Hive mapreduce的map与reduce个数由什么决定？
文章目录
- 1.MapTask的数量决定
- 2.如何来调整MapTask的数量
-
- 2.1 增加map的数量 : 调小maxsize (要小于blockSize才有效,比如100byte)
- 2.2 减少map的数量 : 调大minSize （要大于blockSize才有效，比如250M）
- 2.3生产中一般不调整，但是要知道原理。
- 3. ReduceTask的数量决定
-
- 3.1 在执行hive shell的时候可以看到下列日志
- 3.2 官网对这三个参数的解释
- 3.3 通过源码分析 hive是如何通过方式一 动态计算reduce的个数的
- 4.如何调整reduceTask的数量
-
- 4.1.hive.exec.reducers.bytes.per.reducer与hive.exec.reducers.max
- 4.2.mapreduce.job.reduces
- 4.3 设置reduce数量无效的情况
-
- 4.3.1 order by
- 4.3.2 笛卡尔积
- 4.3.3 map端输出的数据量很小
1.MapTask的数量决定
(1)文件的个数
(2)文件大小
(3)blocksize
简单来说：输入的目录中文件的数量决定多少个map会被运行起来，应用针对每一个分片运行一个map，一般而言，对于每一个输入的文件会有一个map split（我们常常所说的小文件问题，可能引起Hadoop集群资源雪崩，就是这个原因。如在Flink实时写入Hive表场景，如果不进行小文件合并，极可能导致Hadoop集群资源问题）。如果输入文件太大，超过了hdfs块的大小（128M）那么对于同一个输入文件我们会有多余2个的map运行起来
mapTask的数量由文件决定的数学含义：
MapReduce的每一个map处理数据是不能跨越文件的。也就是说minMapNum>=inputFileNum,所以，最终的map个数应该为：
mapNum=max(computeMapNum,inputFileNum)
下面讲述文件大小和blockSize是如果影响mapTask的数量的。
2.如何来调整MapTask的数量
在不改变blockSize的情况下，变更map的情况
计算splitSize的公式：
//下面的调整都是按照此公式进行调整（具体解释看我的另外一篇博客 https://blog.csdn.net/lihuazaizheli/article/details/107370695）
splitSize=Min(maxSize,Max(minSize,blockSize))
当 minSize《 blockSize 《 maxSize 时，splitSize=blockSize
当 blockSize《 minSize《 maxSize 时，splitSize=minSize ，可应用于减少map数量，调大minSize（至少大于blockSize）
当 minSize《 maxSize 《blockSize 时，splitSize=maxSize ， 可应用于增加map数量，调小maxSize(至少小于blockSize)
如果开启了本地测试模式，需要set hive.exec.mode.local.auto=fase关闭才能够看出具的map与reduce个数的日志
重要结论：
(1)【重要】如果不进行任何设置，默认的map个数是和blcok_size相关的。
default_num = 输入文件的整体大小 / block_size;
(2)可以通过参数mapred.map.tasks来设置程序员期望的map个数，但是这个个数只有在大于default_num的时候，才会生效。
(3)一般是结合hive使用，在hive中的更加通用的配置(下文2.1中介绍的是fileinputformat的输入形式配置)
a.minSize相关的参数：
hive (default)> set mapred.min.split.size;
mapred.min.split.size=1
hive (default)> set mapred.min.split.size.per.node;
mapred.min.split.size.per.node=1
hive (default)> set mapred.min.split.size.per.rack;
mapred.min.split.size.per.rack=1
b.maxSize相关的参数
hive (default)> set mapred.max.split.size;
mapred.max.split.size=256000000
c.本地聚合参数
hive (default)> set hive.input.format;
hive.input.format=org.apache.hadoop.hive.ql.io.CombineHiveInputFormat
#官网解释
hive.input.format
Default Value: org.apache.hadoop.hive.ql.io.CombineHiveInputFormat
Added In: Hive 0.5.0
The default input format. Set this to HiveInputFormat if you encounter problems with CombineHiveInputFormat.
Also see:
Configuration Properties#hive.tez.input.format
2.1 增加map的数量 : 调小maxsize (要小于blockSize才有效,比如100byte)
注意，下面的例子是fileinputformat格式的数据
方式一：
（1）maxsize的默认值为256M
hive (default)> set mapreduce.input.fileinputformat.split.maxsize;
mapreduce.input.fileinputformat.split.maxsize=256000000
mapreduce.input.fileinputformat.split.maxsize
（2) 将次值调小为100字节，此时splitSize=100;
set mapreduce.input.fileinputformat.split.maxsize=100;
select count(1) from empt;
可以看到日志
number of mappers: 2; number of reducers: 1
方式二：
设置mapred.map.tasks 为一个较大的值(大于default_num=输入文件整体大小/splitSize)。
2.2 减少map的数量 : 调大minSize （要大于blockSize才有效，比如250M）
(1)minsize的默认值是1
hive (default)> set mapreduce.input.fileinputformat.split.minsize;
mapreduce.input.fileinputformat.split.minsize=1
(2)调整minsize为256M,此时splitSize=256M
set mapreduce.input.fileinputformat.split.minsize=256000000;
select count(1) from empt;
2.3生产中一般不调整，但是要知道原理。
一个文件有2列数据，有128M, 默认blockSize为128M的情况下，只有1个map,这个时候可以将map数量调大，以更快的处理数据。
3. ReduceTask的数量决定
reduce个数决定文件的输出个数
(1)决定方式一 后面看源码解析
hive.exec.reducers.bytes.per.reducer 参数1(default:256M)
hive.exec.reducers.max 参数2 (default:1009)
min(参数2, 总数据量/参数1)
一般参数2的值不变动，在普通集群规模下，hive根据数据量自动决定reduce的个数为： 输入总数据量/hive.exec.reducers.bytes.per.reducer
(2)决定方式二
set mapreduce.job.reduces=1;
3.1 在执行hive shell的时候可以看到下列日志
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
set hive.exec.reducers.bytes.per.reducer=
In order to limit the maximum number of reducers:
set hive.exec.reducers.max=
In order to set a constant number of reducers:
set mapreduce.job.reduces=
这三个参数决定了reduce的个数,下面详细解释
3.2 官网对这三个参数的解释
(1)hive.exec.reducers.bytes.per.reducer
Default Value: 1,000,000,000 prior to Hive 0.14.0; 256 MB (256,000,000) in Hive 0.14.0 and later
Added In: Hive 0.2.0; default changed in 0.14.0 with HIVE-7158 (and HIVE-7917)
Size per reducer. The default in Hive 0.14.0 and earlier is 1 GB, that is, if the input size is 10 GB then 10 reducers will be used.
# 重点看这句话
In Hive 0.14.0 and later the default is 256 MB, that is, if the input size is 1 GB then 4 reducers will be used.
# chd5.16.2 hive环境中是128M
hive (default)> set hive.exec.reducers.bytes.per.reducer;
hive.exec.reducers.bytes.per.reducer=134217728
(2)hive.exec.reducers.max
Default Value: 999 prior to Hive 0.14.0; 1009 in Hive 0.14.0 and later
Added In: Hive 0.2.0; default changed in 0.14.0 with HIVE-7158 (and HIVE-7917)
Maximum number of reducers that will be used. If the one specified in the configuration property Configuration Properties#mapred.reduce.tasks is negative, Hive will use this as the maximum number of reducers when automatically determining the number of reducers
# chd5.16.2 hive环境
hive (default)> set hive.exec.reducers.max;
hive.exec.reducers.max=500
(3)mapreduce.job.reduces
Default Value: -1 (disabled)
Added in: Hive 1.1.0 with HIVE-7567
Sets the number of reduce tasks for each Spark shuffle stage (e.g. the number of partitions when performing a Spark shuffle). This is set to -1 by default (disabled); instead the number of reduce tasks is dynamically calculated based on Hive data statistics. Setting this to a constant value sets the same number of partitions for all Spark shuffle stages.
# 重点读这句话， 默认-1是禁用，并且是根据hive的数据量动态计算reduce的个数
the number of reduce tasks is dynamically calculated based on Hive data statistics.
# chd5.16.2 hive环境
hive (default)> set mapreduce.job.reduces;
mapreduce.job.reduces=-1
3.3 通过源码分析 hive是如何通过方式一 动态计算reduce的个数的
在org.apache.hadoop.hive.ql.exec.mr包下的 MapRedTask类中
//方法类调用逻辑
MapRedTask
| ----setNumberOfReducers
| ---- estimateNumberOfReducers
|---- estimateReducers
(1)核心方法setNumberOfReducers
/**
* Set the number of reducers for the mapred work.
*/
private void setNumberOfReducers() throws IOException {
ReduceWork rWork = work.getReduceWork();
// this is a temporary hack to fix things that are not fixed in the compiler
// 获取通过外部传参设置reduce数量的值 rWork.getNumReduceTasks()
Integer numReducersFromWork = rWork == null ? 0 : rWork.getNumReduceTasks();
if (rWork == null) {
console
.printInfo("Number of reduce tasks is set to 0 since there's no reduce operator");
} else {
if (numReducersFromWork >= 0) {
//如果手动设置了reduce的数量 大于等于0 ，则进来，控制台打印日志
console.printInfo("Number of reduce tasks determined at compile time: "
+ rWork.getNumReduceTasks());
} else if (job.getNumReduceTasks() > 0) {
//如果手动设置了reduce的数量，获取配置中的值，并传入到work中
int reducers = job.getNumReduceTasks();
rWork.setNumReduceTasks(reducers);
console
.printInfo("Number of reduce tasks not specified. Defaulting to jobconf value of: "
+ reducers);
} else {
//如果没有手动设置reduce的数量，进入方法
if (inputSummary == null) {
inputSummary = Utilities.getInputSummary(driverContext.getCtx(), work.getMapWork(), null);
}
//【重中之中】estimateNumberOfReducers
int reducers = Utilities.estimateNumberOfReducers(conf, inputSummary, work.getMapWork(),
work.isFinalMapRed());
rWork.setNumReduceTasks(reducers);
console
.printInfo("Number of reduce tasks not specified. Estimated from input data size: "
+ reducers);
}
//hive shell中所看到的控制台打印日志就在这里
console
.printInfo("In order to change the average load for a reducer (in bytes):");
console.printInfo(" set " + HiveConf.ConfVars.BYTESPERREDUCER.varname
+ "=");
console.printInfo("In order to limit the maximum number of reducers:");
console.printInfo(" set " + HiveConf.ConfVars.MAXREDUCERS.varname
+ "=");
console.printInfo("In order to set a constant number of reducers:");
console.printInfo(" set " + HiveConf.ConfVars.HADOOPNUMREDUCERS
+ "=");
}
}
(2)如果没有手动设置reduce的个数，hive是如何动态计算reduce个数的？
int reducers = Utilities.estimateNumberOfReducers(conf, inputSummary, work.getMapWork(),
work.isFinalMapRed());
/**
* Estimate the number of reducers needed for this job, based on job input,
* and configuration parameters.
*
* The output of this method should only be used if the output of this
* MapRedTask is not being used to populate a bucketed table and the user
* has not specified the number of reducers to use.
*
* @return the number of reducers.
*/
public static int estimateNumberOfReducers(HiveConf conf, ContentSummary inputSummary,
MapWork work, boolean finalMapRed) throws IOException {
// bytesPerReducer默认值为256M BYTESPERREDUCER("hive.exec.reducers.bytes.per.reducer", 256000000L)
long bytesPerReducer = conf.getLongVar(HiveConf.ConfVars.BYTESPERREDUCER);
//maxReducers的默认值1009 MAXREDUCERS("hive.exec.reducers.max", 1009)
int maxReducers = conf.getIntVar(HiveConf.ConfVars.MAXREDUCERS);
//对totalInputFileSize的计算
double samplePercentage = getHighestSamplePercentage(work);
long totalInputFileSize = getTotalInputFileSize(inputSummary, work, samplePercentage);
// if all inputs are sampled, we should shrink the size of reducers accordingly.
if (totalInputFileSize != inputSummary.getLength()) {
LOG.info("BytesPerReducer=" + bytesPerReducer + " maxReducers="
+ maxReducers + " estimated totalInputFileSize=" + totalInputFileSize);
} else {
LOG.info("BytesPerReducer=" + bytesPerReducer + " maxReducers="
+ maxReducers + " totalInputFileSize=" + totalInputFileSize);
}
// If this map reduce job writes final data to a table and bucketing is being inferred,
// and the user has configured Hive to do this, make sure the number of reducers is a
// power of two
boolean powersOfTwo = conf.getBoolVar(HiveConf.ConfVars.HIVE_INFER_BUCKET_SORT_NUM_BUCKETS_POWER_TWO) &&
finalMapRed && !work.getBucketedColsByDirectory().isEmpty();
//【真正计算reduce个数的方法】看源码的技巧return的方法是重要核心方法
return estimateReducers(totalInputFileSize, bytesPerReducer, maxReducers, powersOfTwo);
}
(3) 计算reduce个数的方法 estimateReducers
public static int estimateReducers(long totalInputFileSize, long bytesPerReducer,
int maxReducers, boolean powersOfTwo) {
// 假设totalInputFileSize 1000M
// bytes=Math.max(1024M,256M)=1000M
double bytes = Math.max(totalInputFileSize, bytesPerReducer);
//reducers=(int)Math.ceil(1000M/256M)=4 此公式说明如果totalInputFileSize 小于256M ，则reducers=1 ;繁殖 则通过
int reducers = (int) Math.ceil(bytes / bytesPerReducer);
//Math.max(1, 4)=4 ，reducers的结果还是4
reducers = Math.max(1, reducers);
//Math.min(1009,4)=4; reducers的结果还是4
reducers = Math.min(maxReducers, reducers);
int reducersLog = (int)(Math.log(reducers) / Math.log(2)) + 1;
int reducersPowerTwo = (int)Math.pow(2, reducersLog);
if (powersOfTwo) {
// If the original number of reducers was a power of two, use that
if (reducersPowerTwo / 2 == reducers) {
// nothing to do
} else if (reducersPowerTwo > maxReducers) {
// If the next power of two greater than the original number of reducers is greater
// than the max number of reducers, use the preceding power of two, which is strictly
// less than the original number of reducers and hence the max
reducers = reducersPowerTwo / 2;
} else {
// Otherwise use the smallest power of two greater than the original number of reducers
reducers = reducersPowerTwo;
}
}
return reducers;
}
4.如何调整reduceTask的数量
调整hive的reduce个数的两种方法：
4.1.hive.exec.reducers.bytes.per.reducer与hive.exec.reducers.max
a.解释：
在生产中，一般不调整这两个参数，这两个参数是 如果我们不指定hive的reduce个数，hive程序通过上面两个参数进行动态计算 决定reduce的个数。
b.生产一般不调整
4.2.mapreduce.job.reduces
a.解释： 一般在生产中对reduce的个数也不做太多调整，但是有时候reduce的个数太多，hdfs上的小文件太多。 此时就可以通过 调小mapreduce.job.reduces的个数，来减少hdfs上输出文件的个数。
b.生产手动调整reduce个数，使用此参数
c.案例 可以看到number of reducers的个数为5
hive (default)>set mapreduce.job.reduces=5;
hive (default)> select * from empt sort by length(ename);
Query ID = root_20200725161515_730c6c65-9945-4cec-bbaa-e284bcdbb3ce
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks not specified. Defaulting to jobconf value of: 5
In order to change the average load for a reducer (in bytes):
set hive.exec.reducers.bytes.per.reducer=
In order to limit the maximum number of reducers:
set hive.exec.reducers.max=
In order to set a constant number of reducers:
set mapreduce.job.reduces=
Starting Job = job_1594965583131_0089, Tracking URL = http://hadoop:7776/proxy/application_1594965583131_0089/
Kill Command = /hadoop/hadoop/bin/hadoop job -kill job_1594965583131_0089
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 5
4.3 设置reduce数量无效的情况
4.3.1 order by
sql中使用了order by ，由于order by是全局排序，只能在一个reduce中完成，无论怎么调整reduce的数量都是无效的。
hive (default)>set mapreduce.job.reduces=5;
hive (default)> select * from empt order by length(ename);
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1
4.3.2 笛卡尔积
可以看到此案例中的笛卡尔积不走reduce，无论设置多少个都无效。其实本质原因是 数据量比较小，hive的0.10以上版本都是将Mapjoin开启的。
hive (default)>set mapreduce.job.reduces=5;
hive (default)> select a.*,b.* from empt a join dept b;
Warning: Map Join MAPJOIN[7][bigTable=a] in task 'Stage-3:MAPRED' is a cross product
Hadoop job information for Stage-3: number of mappers: 1; number of reducers: 0
关闭掉hive-site.xml中的mapJoin参数
hive.auto.convert.join
false
开启MapJoin功能
再次执行,可以看到日志，笛卡尔积其实也是全局聚合，只能够一个reduce来处理。即使设置了5个reduce也没有效果。
hive (default)> set mapreduce.job.reduces=5;
hive (default)> select a.*,b.* from empt a join dept b;
Warning: Map Join MAPJOIN[16][bigTable=?] in task 'Stage- 3:MAPRED' is a cross product
Warning: Shuffle Join JOIN[4][tables = [a, b]] in Stage 'Stage-1:MAPRED' is a cross product
Hadoop job information for Stage-1: number of mappers: 2; number of reducers: 1
4.3.3 map端输出的数据量很小
计算reduce个数的方法 estimateReducers中有这三行代码：
int reducers = (int) Math.ceil(bytes / bytesPerReducer);
reducers = Math.max(1, reducers);
reducers = Math.min(maxReducers, reducers);
如果map端输出的数据量(假如只有1M)小于hive.exec.reducers.bytes.per.reducer(default:256M)参数值，maxReducers默认为1009，可以计算
int reducers = (int) Math.ceil(1 / 256M)=1;
reducers = Math.max(1, 1)=1;
reducers = Math.min(1009, 1)=1;
所以此时即使你set mapreduce.job.reduces=10是没用的，最后还是只有1个。