本文档介绍文件存储HDFS和对象存储OSS之间的数据迁移操作过程。您可以将文件存储HDFS数据迁移到对象存储OSS，也可以将对象存储OSS的数据迁移到文件存储HDFS。

背景信息

阿里云文件存储HDFS是面向阿里云ECS实例及容器服务等计算资源的文件存储服务。文件存储HDFS允许您就像在Hadoop的分布式文件系统中一样管理和访问数据，让您拥有高性能的热数据访问能力。对象存储OSS是海量、安全、低成本、高可靠的云存储服务，提供标准型、归档型等多种存储类型。您可以通过文件存储HDFS和对象存储OSS之间双向数据迁移，从而为热、温、冷数据合理分层，不但实现对热数据的高性能访问，更能有效控制存储成本。

准备工作

  1. 开通文件存储HDFS服务并创建文件系统实例和挂载点，详情请参见快速入门。
  2. 搭建完成Hadoop集群。建议您使用的Hadoop版本不低于2.7.2，本文档中使用的Hadoop版本为Apache Hadoop 2.7.2。
  3. 在Hadoop集群所有节点上安装JDK。本操作要求JDK版本不低于1.8。
  4. 在Hadoop集群中配置文件存储HDFS实例，详情请参见挂载文件系统。
  5. 在Hadoop集群安装OSS客户端JindoFS SDK。JindoFS SDK详细介绍请参见JindoFS SDK。
    1. 下载jindofs-sdk.jar。
      cp ./jindofs-sdk-*.jar  ${HADOOP_HOME}/share/hadoop/hdfs/lib/jindofs-sdk.jar
    2. 为Hadoop集群所有节点创建JindoFS SDK配置文件。
      1. 添加如下环境变量到/etc/profile文件。
        export B2SDK_CONF_DIR=/etc/jindofs-sdk-conf
      2. 创建OSS存储工具配置文件/etc/jindofs-sdk-conf/bigboot.cfg。
        [bigboot]
logger.dir=/tmp/bigboot-log[bigboot-client]
client.oss.retry=5
client.oss.upload.threads=4
client.oss.upload.queue.size=5
client.oss.upload.max.parallelism=16
client.oss.timeout.millisecond=30000
client.oss.connection.timeout.millisecond=3000
      3. 加载环境变量使之生效。
        source /etc/profile
    3. 在Hadoop集群使用OSS客户端。
      hadoop fs -ls oss://<accessKeyId>:<accessKeySecret>@<bucket-name>.<endpoint>/

将文件存储HDFS数据迁移到对象存储OSS

  1. 准备测试数据。本文文件存储HDFS待迁移测试数据100 GB。
    数据迁移测试数据
  2. 启动Hadoop MapReduce任务（DistCp）将测试数据迁移至对象存储OSS。
    ./hadoop-2.7.2/bin/hadoop distcp  \
dfs://f-xxxxxxxxxxx.cn-shanghai.dfs.aliyuncs.com:10290/dfs2oss/data/data_100g/ \
oss://<accessKeyId>:<accessKeySecret>@<bucket-name>.<endpoint>/data_100g

    参数说明如下表所示。

    参数 说明
    accessKeyId 访问对象存储OSS API的密钥。获取方式请参见创建AccessKey。
    accessKeySecret
    bucket-name.endpoint 对象存储OSS的访问域名，包括存储空间（Bucket）名称和对应的地域域名（Endpoint）地址。
  3. 任务执行完成后，查看迁移结果。

    如果回显包含如下类似信息，说明迁移成功。

    20/07/27 14:10:09 INFO mapreduce.Job: Job job_1595829170826_0002 completed successfully
20/07/27 14:10:09 INFO mapreduce.Job: Counters: 38
        File System Counters
                DFS: Number of bytes read=107374209810
                DFS: Number of bytes written=0
                DFS: Number of read operations=427
                DFS: Number of large read operations=0
                DFS: Number of write operations=42
                FILE: Number of bytes read=0
                FILE: Number of bytes written=2549873
                FILE: Number of read operations=0
                FILE: Number of large read operations=0
                FILE: Number of write operations=0
                OSS: Number of bytes read=0
                OSS: Number of bytes written=215586406400
                OSS: Number of read operations=0
                OSS: Number of large read operations=0
                OSS: Number of write operations=13209500
        Job Counters
                Launched map tasks=21
                Other local map tasks=21
                Total time spent by all maps in occupied slots (ms)=7459391
                Total time spent by all reduces in occupied slots (ms)=0
                Total time spent by all map tasks (ms)=7459391
                Total vcore-milliseconds taken by all map tasks=7459391
                Total megabyte-milliseconds taken by all map tasks=7638416384
        Map-Reduce Framework
                Map input records=101
                Map output records=0
                Input split bytes=2814
                Spilled Records=0
                Failed Shuffles=0
                Merged Map outputs=0
                GC time elapsed (ms)=12483
                CPU time spent (ms)=912380
                Physical memory (bytes) snapshot=15281192960
                Virtual memory (bytes) snapshot=79722381312
                Total committed heap usage (bytes)=8254390272
        File Input Format Counters
                Bytes Read=24596
        File Output Format Counters
                Bytes Written=0
        org.apache.hadoop.tools.mapred.CopyMapper$Counter
                BYTESCOPIED=107374182400
                BYTESEXPECTED=107374182400
                COPY=101
20/07/27 14:10:09 INFO common.AbstractJindoFileSystem: Read total statistics: oss read average -1 us, cache read average -1 us, read oss percent 0%
  4. 验证迁移结果。
    查看迁移到对象存储OSS的测试数据大小。
    [root@master1 local]# ./hadoop-2.7.2/bin/hadoop fs -du -s -h oss://<accessKeyId>:<accessKeySecret>@<bucket-name>.oss-cn-shanghai-internal.aliyuncs.com/data_100g
100 G  oss://<accessKeyId>:<accessKeySecret>@<bucket-name>.<endpoint>/data_100g
    成功迁移到OSS的数据大小验证图

将对象存储OSS数据迁移到文件存储HDFS

  1. 准备测试数据。本文对象存储待迁移测试数据为100 GB。
    OSS数据迁移测试数据
  2. 启动Hadoop MapReduce任务（DistCp）将测试数据迁移至文件存储HDFS。
    ./hadoop-2.7.2/bin/hadoop distcp  \
oss://<accessKeyId>:<accessKeySecret>@<bucket-name>.<endpoint>/data_100g \
dfs://f-xxxxxxxxxxx.cn-shanghai.dfs.aliyuncs.com:10290/oss2dfs/data/data_100g/
  3. 任务执行完成后，查看迁移结果。

    如果回显包含如下类似信息，说明迁移成功。

    20/07/27 14:51:38 INFO mapreduce.Job: Job job_1595829170826_0003 completed successfully
20/07/27 14:51:38 INFO mapreduce.Job: Counters: 38
        File System Counters
                DFS: Number of bytes read=30800
                DFS: Number of bytes written=107374182400
                DFS: Number of read operations=761
                DFS: Number of large read operations=0
                DFS: Number of write operations=241
                FILE: Number of bytes read=0
                FILE: Number of bytes written=2427170
                FILE: Number of read operations=0
                FILE: Number of large read operations=0
                FILE: Number of write operations=0
                OSS: Number of bytes read=0
                OSS: Number of bytes written=0
                OSS: Number of read operations=0
                OSS: Number of large read operations=0
                OSS: Number of write operations=0
        Job Counters
                Launched map tasks=20
                Other local map tasks=20
                Total time spent by all maps in occupied slots (ms)=3384423
                Total time spent by all reduces in occupied slots (ms)=0
                Total time spent by all map tasks (ms)=3384423
                Total vcore-milliseconds taken by all map tasks=3384423
                Total megabyte-milliseconds taken by all map tasks=3465649152
        Map-Reduce Framework
                Map input records=101
                Map output records=0
                Input split bytes=2660
                Spilled Records=0
                Failed Shuffles=0
                Merged Map outputs=0
                GC time elapsed (ms)=22358
                CPU time spent (ms)=1879870
                Physical memory (bytes) snapshot=14572785664
                Virtual memory (bytes) snapshot=82449666048
                Total committed heap usage (bytes)=12528386048
        File Input Format Counters
                Bytes Read=28140
        File Output Format Counters
                Bytes Written=0
        org.apache.hadoop.tools.mapred.CopyMapper$Counter
                BYTESCOPIED=107374182400
                BYTESEXPECTED=107374182400
                COPY=101
20/07/27 14:51:38 INFO common.AbstractJindoFileSystem: Read total statistics: oss read average -1 us, cache read average -1 us, read oss percent 0%
  4. 验证迁移结果。
    查看迁移到文件存储HDFS的测试数据大小。
    [root@master1 local]# ./hadoop-2.7.2/bin/hadoop fs -du -s -h dfs://f-xxxxxxxxxxx.cn-shanghai.dfs.aliyuncs.com:10290/oss2dfs/data/data_100g
100 G  dfs://f-xxxxxxxxxxx.cn-shanghai.dfs.aliyuncs.com:10290/oss2dfs/data/data_100g
    OSS数据迁移成功数据大小

常见问题

对于正在写入的文件，进行迁移时会遗漏最新写入的数据吗？

Hadoop兼容文件系统提供单写者多读者并发语义，针对同一个文件，同一时刻可以有一个写者写入和多个读者读出。以文件存储HDFS到对象存储OSS的数据迁移为例，数据迁移任务打开文件存储HDFS的文件F，根据当前系统状态决定文件F的长度L，将L字节迁移到对象存储OSS。如果在数据迁移过程中，有并发的写者写入，文件F的长度将超过L，但是数据迁移任务无法感知到最新写入的数据。因此，建议当您在做数据迁移时，请避免往迁移的文件中写入数据。