Hadoop—archive

     HDFS并不擅长存储小文件，因为每个文件最少一个block，每个block的元数据都会在namenode节点占用内存，如果存在这样大量的小文件，它们会吃掉namenode节点的大量内存。Hadoop Archive是一个高效地将小文件放入HDFS块中的文件存档文件格式，它能够将多个小文件打包成一个后缀为.har文件，这样减少namenode内存使用的同时，仍然允许对文件进行透明的访问。

    Hadoop Archive目录包含metadata(in the form of _index and _masterindex)和data (part-*)文件。The _index file contains the name of the files that are part of the archive and the location within the part files.

    Hadoop官网的Hadoop Archives Guide

1. 怎么使用Archive

    用法：hadoop archive -archiveName name -p <parent> [-r <replication factor>] <src>* <dest>

    -archiveName用来指定要创建的archive的文件名，必须以.har结尾，例如：foo.har；

    -p用来指定Archive文件的父路径，指定了之后，后面的src、dest都是相对路径，例如：-p /foo/bar a/b/c e/f/g；

    -r用来指定所需的复制因子，这个参数是可有可无的，如果没有的话，默认选择10作为复制因子。

[hadoop@master ~]$ hadoop fs -ls -R /
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:56 /user
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:59 /user/hadoop
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:59 /user/hadoop/har
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:58 /user/hadoop/test_archive
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 /user/hadoop/test_archive/2016-1
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 /user/hadoop/test_archive/2016-2
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 /user/hadoop/test_archive/2016-3

     现在要将/user/hadoop/test_archive目录下的2016-1、2016-2、2016-3文件归档到/user/hadoop/har目录下,可以使用如下命令：hadoop archive -archiveName combine.har -p /user/hadoop test_archive/2016-1 test_archive/2016-2 test_archive/2016-3 har。当然了，也可以使用通配符来完成：hadoop archive -archiveName combine2.har -p /user/hadoop test_archive/2016-[1-3] har

2. 怎么查看Archive文件

     archive作为文件系统层暴露给外界。所以所有的fs shell命令都能在archive上运行，但是要使用不同的URI。 另外，archive是不可改变的。所以重命名，删除和创建都会返回错误。Hadoop Archive的URI是：har://scheme-hostname:port/archivepath/fileinarchive。

    如果没提供scheme-hostname，它会使用默认的文件系统。这种情况下URI是这种形式:har:///archivepath/fileinarchive

[hadoop@master ~]$ hadoop fs -ls /user/hadoop/har
Found 1 items
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 20:02 /user/hadoop/har/combine.har

[hadoop@master ~]$ hadoop fs -ls /user/hadoop/har/combine.har
Found 4 items
-rw-r--r--   3 hadoop supergroup          0 2016-01-30 20:02 /user/hadoop/har/combine.har/_SUCCESS
-rw-r--r--   5 hadoop supergroup        378 2016-01-30 20:02 /user/hadoop/har/combine.har/_index
-rw-r--r--   5 hadoop supergroup         23 2016-01-30 20:02 /user/hadoop/har/combine.har/_masterindex
-rw-r--r--   3 hadoop supergroup          6 2016-01-30 20:02 /user/hadoop/har/combine.har/part-0

[hadoop@master ~]$ hadoop fs -ls -R har://hdfs-master:9000/user/hadoop/har/combine.har
16/01/30 20:27:45 WARN hdfs.DFSClient: DFSInputStream has been closed already
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:58 har://hdfs-master:9000/user/hadoop/har/combine.har/test_archive
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har://hdfs-master:9000/user/hadoop/har/combine.har/test_archive/2016-1
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har://hdfs-master:9000/user/hadoop/har/combine.har/test_archive/2016-2
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har://hdfs-master:9000/user/hadoop/har/combine.har/test_archive/2016-3

     也可以省略掉hdfs-master:9000

[hadoop@master ~]$ hadoop fs -ls -R har:///user/hadoop/har/combine.har
16/01/30 20:31:32 WARN hdfs.DFSClient: DFSInputStream has been closed already
drwxr-xr-x   - hadoop supergroup          0 2016-01-30 19:58 har:///user/hadoop/har/combine.har/test_archive
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har:///user/hadoop/har/combine.har/test_archive/2016-1
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har:///user/hadoop/har/combine.har/test_archive/2016-2
-rw-r--r--   3 hadoop supergroup          2 2016-01-30 19:58 har:///user/hadoop/har/combine.har/test_archive/2016-3

3. 怎么解压Archive文件

    Since all the fs shell commands in the archives worktransparently, unarchiving is just a matter of copying.

    To unarchive sequentially（串行解压）:

    hdfs dfs -cp har:///user/zoo/foo.har/dir1 hdfs:/user/zoo/newdir

    To unarchive in parallel（并行解压）, use DistCp:

    hadoop distcp har:///user/zoo/foo.har/dir1 hdfs:/user/zoo/newdir

4. Hadoop Archives and MapReduce

    Using Hadoop Archives in MapReduce is as easy as specifying a different input filesystem than the default file system. If you have a hadoop archive stored in HDFS in /user/zoo/foo.har then for using this archive for MapReduce input, all you need to specify the input directory as har:///user/zoo/foo.har. Since Hadoop Archives is exposed as a file system MapReduce will be able to use all the logical input files in Hadoop Archives as input.