Overview

COSDistCp is a MapReduce-based distributed file copy tool mainly used for data copy between HDFS and COS. It introduces the following features:

• Performs incremental copy as well as real-time verification, and obtain the list of different files based on the length and CRC checksum.
• Filters files in the source directory with regular expression.
• Decompresses files in the source directory and compresses them to the target compression format.
• Aggregates text files based on regular expression.
• Preserves user/user group, extension attributes, and time of the source file and directory.
• Limits the read bandwidth.

Operating Environment

Operating system

Linux

Software requirements

Hadoop 2.6.0 or above; Hadoop-COS 5.8.7 or above.

Download and Installation

Obtaining the COSDistCp JAR package

If your Hadoop version is 2.x, you can download cos-distcp-1.3-2.8.5.jar and verify the integrity of the downloaded JAR package according to the MD5 checksum of the package.

If your Hadoop version is 3.x, you can download cos-distcp-1.3-3.1.0.jar and verify the integrity of the downloaded JAR package according to the MD5 checksum of the package.

Notes

In the Hadoop environment, install Hadoop-COS and then run the COSDistCp tool.

How It Works

COSDistCp uses the MapReduce framework. Mappers group files, while multi-thread reducers perform file copy, compression, verification, attribute preservation, and retry. COSDistCp will overwrite files with the same name in the destination location. If data copy or verification fails, the corresponding file may fail to be copied and information about these files will be written in a temporary directory. If new files are added to your source file system or the file content changes, you can use the --skipMode or --diffMode parameter to compare the length or CRC checksum of the files to implement incremental copy.

Parameters

You can run the hadoop jar cos-distcp-${version}.jar --help (${version} is the version number) command to view the COSDistCp-supported parameters. The following table describes the COSDistCp parameters:

Examples

Viewing the help option

Run the following command with --help to view the parameters supported by COSDistCp:

hadoop jar cos-distcp-${version}.jar --help

In the command above, ${version} is the version ID of the COSDistCp. For example, the name of the COSDistCp JAR package (version 1.0) is cos-distcp-1.0.jar.

Specifying the source and destination locations for the files to copy

Run the following command with the --src and --dest parameters:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse

COSDistCp will retry 5 times for files that failed to be copied. If the copy still fails, these files will be written to the /tmp/${randomUUID}/output/failed/ directory, where ${randomUUID} is a random string.

The following information about a source file might be contained in the output:

1. SRC_MISS: The copy fails because the source file contained in the manifest is not found.
2. COPY_FAILED: The copy fails due to other reasons.

You can run the following command to obtain the list of different files except for those recorded as SRC_MISS:

hadoop fs -getmerge /tmp/${randomUUID}/output/failed/ failed-manifest
grep -v '"comment":"SRC_MISS"' failed-manifest |gzip > failed-manifest.gz

Run the following command to recopy files that failed to be copied:

hadoop  jar cos-distcp-${version}.jar --reducerNumber=20 --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse/ --previousManifest=file:///usr/local/service/hadoop/failed-manifest.gz --copyFromManifest

Run the following command to obtain the log of the MapReduce job. In this way, you can find out the cause of the copy failure. Note that application_1610615435237_0021 is the application ID.

yarn logs -applicationId application_1610615435237_0021 > application_1610615435237_0021.log

Querying Counters

When the copy operation ends, statistics on the copy will be output. The counters are as follows:

Copy operation:

• BYTES_EXPECTED: total size (in bytes) expected to copy according to the source directory
• FILES_EXPECTED: number of files to copy according to the source directory
• BYTES_SKIPPED: total size (in bytes) of files that can be skipped (same length or checksum value)
• FILES_SKIPPED: number of source files that can be skipped (same length or checksum value)
• FILES_COPIED: number of source files that is successfully copied
• FILES_FAILED: number of source files that failed to be copied

CosDistCp Counters
        BYTES_EXPECTED=10198247
        BYTES_SKIPPED=10196880
        FILES_COPIED=1
        FILES_EXPECTED=7
        FILES_FAILED=1
        FILES_SKIPPED=5

Filtering source files with a regular expression

Run the following command with the --srcPattern parameter. In this example, only files whose extension is ".log" in the /data/warehouse/logs directory are copied.

hadoop jar cos-distcp-${version}.jar  --src /data/warehouse/logs --dest cosn://examplebucket-1250000000/data/warehouse --srcPattern='.*/logs/.*\.log'

Specifying the number of reducers and the number of threads for each reducer process

Run the following command with the --reducerNumber and --workersNumber parameters. COSDistCp adopts a multi-process, multi-thread framework for the copy operation. You can:

• Use --reducerNumber to specify the number of reducer processes.
• Use --workerNumber to specify the number of threads for each reducer process.

hadoop jar cos-distcp-${version}.jar --src /data/warehouse/ --dest cosn://examplebucket-1250000000/data/warehouse --reducerNumber=10 --workerNumber=5

Deleting the source files

Run the command with the --deleteOnSuccess parameter. The following example deletes the corresponding source files in the /data/warehouse directory immediately after they are copied from HDFS to COS:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse --deleteOnSuccess

Note：

If --deleteOnSuccess is specified, each source file is deleted immediately after the file is copied, but not after all source files are copied.

Restricting the read bandwidth for a single file

Run the command with the --bandWidth parameter (in MB). The following command example restricts the read bandwidth of each copied file to 10 MB/s:

hadoop jar cos-distcp-${version}.jar  --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse --bandWidth=10

Specifying the checksum type of Hadoop-COS

Run the following command with the --cosChecksumType parameter. Valid values are CRC32C (default) and CRC64.

hadoop jar cos-distcp-${version}.jar  --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse --cosChecksumType=CRC32C

Skipping files with the same length

Run the following command with the --skipMode parameter. The following command example skips files with the same length:

hadoop jar cos-distcp-${version}.jar  --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse  --skipMode=length

--skipMode is used to verify whether the source and destination files are the same before the copy. If they are the same, the file will be skipped. Valid values are none (no verification), length, checksum, and length-checksum (length + CRC checksum).

If the checksum algorithms of the source and destination file systems are different, the source file will be read for calculating a new checksum. If your source is HDFS, you can identify whether the HDFS source supports the COMPOSITE-CRC32C algorithm as follows:

hadoop fs  -Ddfs.checksum.combine.mode=COMPOSITE_CRC -checksum /data/test.txt
/data/test.txt  COMPOSITE-CRC32C        6a732798

Verifying whether the source and destination files have the same CRC checksum

Run the following command with the --checkMode parameter. The following command example verifies whether the checksums of the source and destination files are the same when the copy is completed:

When you are copying files from a non-COS file system to COS, if the CRC algorithms of the source and Hadoop-COS are different, the CRC checksum will be calculated during the copy operation. When the copy operation is completed, the CRC checksum of the destination file will be obtained and compared with the calculated CRC checksum of the source file.

hadoop jar cos-distcp-${version}.jar   --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse --checkMode=checksum

Specifying the output compression codec

Run the command with the --outputCodec parameter, which allows you to compress HDFS data to COS in real time to reduce storage costs. Valid values are keep, none, gzip, lzop, and snappy. If set to none, the files will be copied uncompressed. If set to keep, the files will be copied with no change in their compression. The following is an example:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse/logs --dest cosn://examplebucket-1250000000/data/warehouse/logs-gzip --outputCodec=gzip

Note：

If not set to keep, the files will be decompressed and converted to the target compression format. Due to the difference in compression parameters, the content of the destination files might be different from that of the source files, but the files will be the same after decompression.

Copying multiple directories

You can create a local file (for example, srcPrefixes.txt) and add multiple directories to copy to the file. After this, you can run the cat command to view the directories as follows:

cat srcPrefixes.txt 
/data/warehouse/20181121/
/data/warehouse/20181122/

Then, you can use --srcPrefixesFile to specify this file. The command is as follows:

hadoop jar  cos-distcp-${version}.jar --src /data/warehouse  --srcPrefixesFile file:///usr/local/service/hadoop/srcPrefixes.txt --dest  cosn://examplebucket-1250000000/data/warehouse/ --reducerNumber=20

Generating the target manifest and specifying the previous manifest

Run the command with the --outputManifest and --previousManifest parameters.

• --outputManifest generates a local manifest.gz (Gzip compressed) file. When the copy operation is successful, the file is moved to the directory specified in --dest.
• --previousManifest specifies the destination files that are copied during the previous copy operation (--outputManifest). COSDistCp will skip files of the same size.

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest  cosn://examplebucket-1250000000/data/warehouse/ --outputManifest=manifest.gz --previousManifest= cosn://examplebucket-1250000000/data/warehouse/manifest-2020-01-10.gz

Note：

The command above performs incremental copy only. Only files with size changes can be copied. If the file content is changed, you can refer to the example of --diffMode and determine the changed manifest files based on the CRC checksum.

Obtaining the list of different files and performing incremental copy based on the CRC checksum

Run the command with the --diffMode and --diffOutput parameters:

• --diffMode can be set to length or length-checksum.
  • --diffMode=length obtains the list of different files based on whether the file sizes are the same.
  • --diffMode=length-checksum obtains the list of different files based on whether the file size and CRC checksum are the same.
• --diffOutput specifies the output directory for the diff operation.

The following example verifies whether the source and destination files are the same based on the file size and CRC checksum:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse/ --diffMode=length-checksum --diffOutput=/tmp/diff-output

Note：

If the destination file system is COS, and the CRC algorithms of the source and destination file systems are different, COSDistCp will pull the source files and calculate the new CRC checksum for CRC checksum comparison.

After the command above is executed, a list of different files will be generated in the /tmp/diff-output directory of HDFS. The following information about a source file might be contained in the output:

1. DEST_MISS: The destination file does not exist.
2. SRC_MISS: The source file contained in the source file manifest is not found during the verification.
3. LENGTH_DIFF: Sizes of the source and destination files are different.
4. CHECKSUM_DIFF: CRC checksums of the source and destination files are different.
5. DIFF_FAILED: The diff operation fails due to insufficient permissions or other reasons.

You can run the following command to obtain the list of different files except for those contained SRC_MISS:

hadoop fs -getmerge /tmp/diff-output diff-manifest
grep -v '"comment":"SRC_MISS"' diff-manifest |gzip > diff-manifest.gz

Run the following command to implement incremental copy based on the list of different files:

hadoop  jar cos-distcp-${version}.jar --reducerNumber=20 --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse/ --previousManifest=file:///usr/local/service/hadoop/diff-manifest.gz --copyFromManifest

Specifying the storage class for COS objects

Run the following command with the --storageClass parameter:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse/ --outputManifest=manifest-2020-01-10.gz --storageClass=STANDARD_IA

Copying metadata of the source file

Run the following command with the --preserveStatus parameter. The following command example copies the user, group, permission, and timestamps (modification time and access time) metadata of the source file/directory to the destination file/directory:

hadoop jar cos-distcp-${version}.jar --src /data/warehouse --dest cosn://examplebucket-1250000000/data/warehouse/ --preserveStatus=ugpt

FAQs

How can I run COSDistCp if Hadoop-COS is not configured in the environment?

You can download a specific version of the COSDistCp JAR package according to the Hadoop version and specify the Hadoop-COS-related parameters to perform the copy operation.

hadoop jar cos-distcp-1.3-2.8.5.jar \
-Dfs.cosn.credentials.provider=org.apache.hadoop.fs.auth.SimpleCredentialProvider \
-Dfs.cosn.userinfo.secretId=COS_SECRETID \
-Dfs.cosn.userinfo.secretKey=COS_SECRETKEY \
-Dfs.cosn.bucket.region=ap-guangzhou \
-Dfs.cosn.impl=org.apache.hadoop.fs.CosFileSystem \
-Dfs.AbstractFileSystem.cosn.impl=org.apache.hadoop.fs.CosN \
--src /data/warehouse \
--dest cosn://examplebucket-1250000000/warehouse

What do I do if the result shows that some files failed to be copied?

COSDistCp will retry 5 times for IOException occurred during the copy process. If the copy still fails, information about the failed files will be written to the /tmp/${randomUUID}/output/failed/ directory, where ${randomUUID} is a random string. Common reasons for the copy failure are as follows:

1. The source file contained in the copy manifest is not found during the copy (recorded as SRC_MISS).
2. The job initiator does not have permission to read the source file or write the destination file, or the copy failed due to other reasons (recorded as COPY_FAILED).

If the log message indicates that the source file does not exist, and the source file is ignorable, you can run the following command to obtain the list of different files except for those recorded as SRC_MISS:

hadoop fs -getmerge /tmp/${randomUUID}/output/failed/ failed-manifest
grep -v '"comment":"SRC_MISS"' failed-manifest |gzip > failed-manifest.gz

Except for those recorded as SRC_MISS, if there are other failed files, you can locate the failure reasons by referring to the error log messages in the /tmp/${randomUUID}/output/logs/ directory and pulling the application logs. The following command example pulls the logs of the yarn application:

yarn logs -applicationId application_1610615435237_0021 > application_1610615435237_0021.log

In the command above, application_1610615435237_0021 is the application ID.