• reduce task數(shù)量的決定機制

1、業(yè)務邏輯需要

2、數(shù)據(jù)量大小

設置方法：

job.setNumReduceTasks(5)

• map task數(shù)量的決定機制

由于map task之間沒有協(xié)作關系，每一個map task都是各自為政，在map task的處理中沒法做“全局”性的聚合操作，所以map task的數(shù)量完全取決于所處理的數(shù)據(jù)量的大小

決定機制：

對待處理數(shù)據(jù)進行“切片”

每一個切片分配一個map task來處理

Mapreduce框架中默認的切片機制：

TextInputFormat.getSplits()繼承自FileInputFormat.getSplits()

1：定義一個切片大小：可以通過參數(shù)來調節(jié)，默認情況下等于“hdfs中設置的blocksize”,通常是128M

2：獲取輸入數(shù)據(jù)目錄下所有待處理文件List

3：遍歷文件List，逐個逐個文件進行切片

?????? for(file:List)

?????? ?? 對file從0偏移量開始切，每到128M就構成一個切片，比如a.txt（200M），就會被切成兩個切片：?? a.txt: 0-128M,? a.txt :128M-256M

??????? 再比如b.txt（80M），就會切成一個切片， b.txt :0-80M

• 如果要處理的數(shù)據(jù)是大量的小文件，使用上述這種默認切片機制，就會導致大量的切片，從而maptask進程數(shù)特別多，但是每一個切片又非常小，每個maptask的處理數(shù)據(jù)量就很小，從而，整體的效率會很低。

通用解決方案：就是將多個小文件劃分成一個切片；實現(xiàn)辦法就是自定義一個Inputformat子類重寫里面的getSplits方法；

Mapreduce框架中自帶了一個用于此場景的Inputformat實現(xiàn)類：CombineFileInputformat

數(shù)據(jù)切片與map任務數(shù)的機制

示例觀察（多文件，大文件）

源碼跟蹤

TextInputFormat源碼閱讀

isSplitable() 判斷要處理的數(shù)據(jù)是否可以做切片

getSplit()? 規(guī)劃切片信息(實現(xiàn)在FileInputFormat類中)

----TextInputformat切片邏輯： 對每一個文件單獨切片；切片大小默認等于blocksize

但是有兩個參數(shù)可以調整：

如果是大量小文件，這種切片邏輯會有重大弊端：切片數(shù)量太多,maptask太多

createRecordReader()? 構造一個記錄讀取器

?

具體讀取數(shù)據(jù)的邏輯是實現(xiàn)在LineRecordReader中 （按行讀取數(shù)據(jù)，行起始偏移量作為key，行的內容作為value），比較特別的地方是：

LineRecordReader在讀取一個具體的切片時，總是忽略掉第一行（針對的是：非第一切片），總是跨split多讀一行(針對的是：非最末切片)

• InputFormat的繼承體系

InputFormat子類介紹：

（1）TextInputFormat(默認的輸入格式類)詳解

-- 源碼結構 getsplits()? reader

-- 為何不會出現(xiàn)一行被割斷處理的原理

• 在LineRecordReader中，對split的第一行忽略

public void initialize(InputSplit genericSplit,TaskAttemptContext context) throws IOException {FileSplit split = (FileSplit) genericSplit;Configuration job = context.getConfiguration();… ………..// open the file and seek to the start of the splitfinal FileSystem fs = file.getFileSystem(job);fileIn = fs.open(file);CompressionCodec codec = new CompressionCodecFactory(job).getCodec(file);if (null!=codec) {… … … … //我們總是將第一條記錄拋棄（文件第一個split除外） //因為我們總是在nextKeyValue ()方法中跨split多讀了一行（文件最后一個split除外）if (start != 0) {start += in.readLine(new Text(), 0, maxBytesToConsume(start));}this.pos = start;}

• 在LineRecordReader中，nextKeyValue ()方法總是跨split多讀一行

public boolean nextKeyValue() throws IOException {if (key == null) {key = new LongWritable();}key.set(pos);if (value == null) {value = new Text();}int newSize = 0;// 使用<=來多讀取一行while (getFilePosition() <= end || in.needAdditionalRecordAfterSplit()) {newSize = in.readLine(value, maxLineLength,Math.max(maxBytesToConsume(pos), maxLineLength));pos += newSize;if (newSize < maxLineLength) {break;…. …. }

CombineTextInputFormat?

它的切片邏輯跟TextInputformat完全不同：

CombineTextInputFormat可以將多個小文件劃為一個切片

這種機制在處理海量小文件的場景下能提高效率

(小文件處理的機制，最優(yōu)的是將小文件先合并再處理)

思路

CombineFileInputFormat涉及到三個重要的屬性：

mapred.max.split.size：同一節(jié)點或同一機架的數(shù)據(jù)塊形成切片時，切片大小的最大值；

mapred.min.split.size.per.node：同一節(jié)點的數(shù)據(jù)塊形成切片時，切片大小的最小值；

mapred.min.split.size.per.rack：同一機架的數(shù)據(jù)塊形成切片時，切片大小的最小值。

切片形成過程：

（1）逐個節(jié)點（數(shù)據(jù)塊）形成切片；

a.遍歷并累加這個節(jié)點上的數(shù)據(jù)塊，如果累加數(shù)據(jù)塊大小大于或等于mapred.max.split.size，則將這些數(shù)據(jù)塊形成一個切片，繼承該過程，直到剩余數(shù)據(jù)塊累加大小小于mapred.max.split.size，則進行下一步；

b.如果剩余數(shù)據(jù)塊累加大小大于或等于mapred.min.split.size.per.node，則將這些剩余數(shù)據(jù)塊形成一個切片，如果剩余數(shù)據(jù)塊累加大小小于mapred.min.split.size.per.node，則這些數(shù)據(jù)塊留待后續(xù)處理。

（2）逐個機架（數(shù)據(jù)塊）形成切片；

a.遍歷并累加這個機架上的數(shù)據(jù)塊（這些數(shù)據(jù)塊即為上一步遺留下來的數(shù)據(jù)塊），如果累加數(shù)據(jù)塊大小大于或等于mapred.max.split.size，則將這些數(shù)據(jù)塊形成一個切片，繼承該過程，直到剩余數(shù)據(jù)塊累加大小小于mapred.max.split.size，則進行下一步；

b.如果剩余數(shù)據(jù)塊累加大小大于或等于mapred.min.split.size.per.rack，則將這些剩余數(shù)據(jù)塊形成一個切片，如果剩余數(shù)據(jù)塊累加大小小于mapred.min.split.size.per.rack，則這些數(shù)據(jù)塊留待后續(xù)處理。

（3）遍歷并累加剩余數(shù)據(jù)塊，如果數(shù)據(jù)塊大小大于或等于mapred.max.split.size，則將這些數(shù)據(jù)塊形成一個切片，繼承該過程，直到剩余數(shù)據(jù)塊累加大小小于mapred.max.split.size，則進行下一步；

（4）剩余數(shù)據(jù)塊形成一個切片。

核心實現(xiàn)

// mapping from a rack name to the list of blocks it has HashMap<String,List<OneBlockInfo>> rackToBlocks = new HashMap<String,List<OneBlockInfo>>(); // mapping from a block to the nodes on which it has replicas HashMap<OneBlockInfo,String[]> blockToNodes = new HashMap<OneBlockInfo,String[]>(); // mapping from a node to the list of blocks that it contains HashMap<String,List<OneBlockInfo>> nodeToBlocks = new HashMap<String,List<OneBlockInfo>>();

開始形成切片之前，需要初始化三個重要的映射關系：

rackToBlocks：機架和數(shù)據(jù)塊的對應關系，即某一個機架上有哪些數(shù)據(jù)塊；

blockToNodes：數(shù)據(jù)塊與節(jié)點的對應關系，即一塊數(shù)據(jù)塊的“拷貝”位于哪些節(jié)點；

nodeToBlocks：節(jié)點和數(shù)據(jù)塊的對應關系，即某一個節(jié)點上有哪些數(shù)據(jù)塊；

初始化過程如下代碼所示，其中每一個Path代表的文件被形成一個OneFileInfo對象，映射關系也在形成OneFileInfo的過程中被維護。

// populate all the blocks for all fileslong totLength = 0; for (int i = 0; i < paths.length; i++) {files[i] = new OneFileInfo(paths[i], job, rackToBlocks, blockToNodes, nodeToBlocks, rackToNodes);totLength += files[i].getLength(); }

逐個節(jié)點（數(shù)據(jù)塊）形成切片，代碼如下：

// 保存當前切片所包含的數(shù)據(jù)塊ArrayList<OneBlockInfo> validBlocks = new ArrayList<OneBlockInfo>();// 保存當前切片中的數(shù)據(jù)塊屬于哪些節(jié)點ArrayList<String> nodes = new ArrayList<String>();// 保存當前切片的大小long curSplitSize = 0;// process all nodes and create splits that arelocalto a node. // 依次處理每個節(jié)點上的數(shù)據(jù)塊for (Iterator<Map.Entry<String, List<OneBlockInfo>>> iter = nodeToBlocks.entrySet().iterator(); iter.hasNext();) {Map.Entry<String, List<OneBlockInfo>> one = iter.next();nodes.add(one.getKey());List<OneBlockInfo> blocksInNode = one.getValue();// for each block, copy it into validBlocks. Delete it from blockToNodes so that the same block does not appear in// two different splits.// 依次處理每個數(shù)據(jù)塊，注意blockToNodes變量的作用，它保證了同一數(shù)據(jù)塊不會出現(xiàn)在兩個切片中for (OneBlockInfo oneblock : blocksInNode) {if (blockToNodes.containsKey(oneblock)) {validBlocks.add(oneblock);blockToNodes.remove(oneblock);curSplitSize += oneblock.length;// if the accumulated split size exceeds the maximum, then create this split.// 如果數(shù)據(jù)塊累積大小大于或等于maxSize，則形成一個切片if (maxSize != 0 && curSplitSize >= maxSize) {//create an input split andadd it to the splits array addCreatedSplit(job, splits, nodes, validBlocks);curSplitSize = 0;validBlocks.clear();}}}// if there were any blocks left over and their combined size is// larger than minSplitNode, then combine them into one split.// Otherwise add them back to the unprocessed pool. It is likely // that they will be combined with other blocks from the same rack later on.// 如果剩余數(shù)據(jù)塊大小大于或等于minSizeNode，則將這些數(shù)據(jù)塊構成一個切片；// 如果剩余數(shù)據(jù)塊大小小于minSizeNode，則將這些數(shù)據(jù)塊歸還給blockToNodes，交由后期“同一機架”過程處理if (minSizeNode != 0 && curSplitSize >= minSizeNode) {//create an input split andadd it to the splits array addCreatedSplit(job, splits, nodes, validBlocks);} else {for (OneBlockInfo oneblock : validBlocks) {blockToNodes.put(oneblock, oneblock.hosts);}}validBlocks.clear();nodes.clear();curSplitSize = 0;}

（2）逐個機架（數(shù)據(jù)塊）形成切片，代碼如下：

// if blocks in a rack are below the specified minimum size, then keep them// in 'overflow'. After the processing of all racks is complete, these overflow// blocks will be combined into splits.// overflowBlocks用于保存“同一機架”過程處理之后剩余的數(shù)據(jù)塊ArrayList<OneBlockInfo> overflowBlocks = new ArrayList<OneBlockInfo>();ArrayList<String> racks = new ArrayList<String>();// Process all racks over and over again until there is no more work to do.while (blockToNodes.size() > 0) {//Create one split for this rack before moving over to the next rack. // Come back to this rack after creating a single split for each of the // remaining racks.// Process one rack location at a time, Combine all possible blocks that// reside on this rack as one split. (constrained by minimum and maximum// split size).// iterate over all racks // 依次處理每個機架for (Iterator<Map.Entry<String, List<OneBlockInfo>>> iter = rackToBlocks.entrySet().iterator(); iter.hasNext();) {Map.Entry<String, List<OneBlockInfo>> one = iter.next();racks.add(one.getKey());List<OneBlockInfo> blocks = one.getValue();// for each block, copy it into validBlocks. Delete it from// blockToNodes so that the same block does not appear in// two different splits.boolean createdSplit = false;// 依次處理該機架的每個數(shù)據(jù)塊for (OneBlockInfo oneblock : blocks) {if (blockToNodes.containsKey(oneblock)) {validBlocks.add(oneblock);blockToNodes.remove(oneblock);curSplitSize += oneblock.length;// if the accumulated split size exceeds the maximum, then create this split.// 如果數(shù)據(jù)塊累積大小大于或等于maxSize，則形成一個切片if (maxSize != 0 && curSplitSize >= maxSize) {//create an input split andadd it to the splits array addCreatedSplit(job, splits, getHosts(racks), validBlocks);createdSplit = true;break;}}}// if we created a split, then just go to the next rackif (createdSplit) {curSplitSize = 0;validBlocks.clear();racks.clear();continue;}if (!validBlocks.isEmpty()) {// 如果剩余數(shù)據(jù)塊大小大于或等于minSizeRack，則將這些數(shù)據(jù)塊構成一個切片if (minSizeRack != 0 && curSplitSize >= minSizeRack) {// if there is a mimimum size specified, then create a single split// otherwise, store these blocks into overflow data structure addCreatedSplit(job, splits, getHosts(racks), validBlocks);} else {// There were a few blocks in this rack that remained to be processed.// Keep them in 'overflow' block list. These will be combined later.// 如果剩余數(shù)據(jù)塊大小小于minSizeRack，則將這些數(shù)據(jù)塊加入overflowBlocks overflowBlocks.addAll(validBlocks);}}curSplitSize = 0;validBlocks.clear();racks.clear();}}

（3）遍歷并累加剩余數(shù)據(jù)塊，代碼如下：

// Process all overflow blocksfor (OneBlockInfo oneblock : overflowBlocks) {validBlocks.add(oneblock); curSplitSize += oneblock.length;// This might cause an exiting rack location to be re-added, // but it should be ok.for (int i = 0; i < oneblock.racks.length; i++) {racks.add(oneblock.racks[i]); } // if the accumulated split size exceeds the maximum, then //create this split. // 如果剩余數(shù)據(jù)塊大小大于或等于maxSize，則將這些數(shù)據(jù)塊構成一個切片if (maxSize != 0 && curSplitSize >= maxSize) { //create an input split andadd it to the splits array addCreatedSplit(job, splits, getHosts(racks), validBlocks); curSplitSize = 0;validBlocks.clear();racks.clear();}}

（4）剩余數(shù)據(jù)塊形成一個切片,代碼如下：

// Process any remaining blocks, if any.if (!validBlocks.isEmpty()) {addCreatedSplit(job, splits, getHosts(racks), validBlocks);}

總結

CombineFileInputFormat形成切片過程中考慮數(shù)據(jù)本地性（同一節(jié)點、同一機架），首先處理同一節(jié)點的數(shù)據(jù)塊，然后處理同一機架的數(shù)據(jù)塊，最后處理剩余的數(shù)據(jù)塊，可見本地性是逐步減弱的。另外CombineFileInputFormat是抽象的，具體使用時需要自己實現(xiàn)getRecordReader方法。

（3）SequenceFileInputFormat/SequenceFileOutputFormat

sequenceFile是hadoop中非常重要的一種數(shù)據(jù)格式

sequenceFile文件內部的數(shù)據(jù)組織形式是：K-V對

讀入/寫出為hadoop序列文件

?

?

?

?

?

?

?

?

?

?

?

?

?

?

?

?

總結

以上是生活随笔為你收集整理的Hadoop大数据--Mapreduce程序运行并发度的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。