Native Libraries Guide

Overview

This guide describes the native hadoop library and includes a small discussion about native shared libraries.

Note: Depending on your environment, the term “native libraries” could refer to all *.so’s you need to compile; and, the term “native compression” could refer to all *.so’s you need to compile that are specifically related to compression. Currently, however, this document only addresses the native hadoop library (libhadoop.so). The document for libhdfs library (libhdfs.so) is here.

Native Hadoop Library

Hadoop has native implementations of certain components for performance reasons and for non-availability of Java implementations. These components are available in a single, dynamically-linked native library called the native hadoop library. On the *nix platforms the library is named libhadoop.so.

Usage

It is fairly easy to use the native hadoop library:

  1. Review the components.
  2. Review the supported platforms.
  3. Either download a hadoop release, which will include a pre-built version of the native hadoop library, or build your own version of the native hadoop library. Whether you download or build, the name for the library is the same: libhadoop.so
  4. Install the compression codec development packages (>zlib-1.2, >gzip-1.2):
    • If you download the library, install one or more development packages - whichever compression codecs you want to use with your deployment.
    • If you build the library, it is mandatory to install both development packages.
  5. Check the runtime log files.

Components

The native hadoop library includes various components:

Supported Platforms

The native hadoop library is supported on *nix platforms only. The library does not to work with Cygwin or the Mac OS X platform.

The native hadoop library is mainly used on the GNU/Linus platform and has been tested on these distributions:

  • RHEL4/Fedora
  • Ubuntu
  • Gentoo

On all the above distributions a 32/64 bit native hadoop library will work with a respective 32/64 bit jvm.

Download

The pre-built 32-bit i386-Linux native hadoop library is available as part of the hadoop distribution and is located in the lib/native directory. You can download the hadoop distribution from Hadoop Common Releases.

Be sure to install the zlib and/or gzip development packages - whichever compression codecs you want to use with your deployment.

Build

The native hadoop library is written in ANSI C and is built using the GNU autotools-chain (autoconf, autoheader, automake, autoscan, libtool). This means it should be straight-forward to build the library on any platform with a standards-compliant C compiler and the GNU autotools-chain (see the supported platforms).

The packages you need to install on the target platform are:

  • C compiler (e.g. GNU C Compiler)
  • GNU Autools Chain: autoconf, automake, libtool
  • zlib-development package (stable version >= 1.2.0)
  • openssl-development package(e.g. libssl-dev)

Once you installed the prerequisite packages use the standard hadoop pom.xml file and pass along the native flag to build the native hadoop library:

   $ mvn package -Pdist,native -DskipTests -Dtar

You should see the newly-built library in:

   $ hadoop-dist/target/hadoop-3.5.0-SNAPSHOT/lib/native

Please note the following:

  • It is mandatory to install both the zlib and gzip development packages on the target platform in order to build the native hadoop library; however, for deployment it is sufficient to install just one package if you wish to use only one codec.
  • It is necessary to have the correct 32/64 libraries for zlib, depending on the 32/64 bit jvm for the target platform, in order to build and deploy the native hadoop library.

Runtime

The bin/hadoop script ensures that the native hadoop library is on the library path via the system property: -Djava.library.path=<path>

During runtime, check the hadoop log files for your MapReduce tasks.

  • If everything is all right, then: DEBUG util.NativeCodeLoader - Trying to load the custom-built native-hadoop library... INFO util.NativeCodeLoader - Loaded the native-hadoop library
  • If something goes wrong, then: WARN util.NativeCodeLoader - Unable to load native-hadoop library for your platform... using builtin-java classes where applicable

Check

NativeLibraryChecker is a tool to check whether native libraries are loaded correctly. You can launch NativeLibraryChecker as follows:

   $ hadoop checknative -a
   14/12/06 01:30:45 WARN bzip2.Bzip2Factory: Failed to load/initialize native-bzip2 library system-native, will use pure-Java version
   14/12/06 01:30:45 INFO zlib.ZlibFactory: Successfully loaded & initialized native-zlib library
   Native library checking:
   hadoop: true /home/ozawa/hadoop/lib/native/libhadoop.so.1.0.0
   zlib:   true /lib/x86_64-linux-gnu/libz.so.1
   zstd: true /usr/lib/libzstd.so.1
   lz4:    true revision:99
   bzip2:  false

Native Shared Libraries

You can load any native shared library using DistributedCache for distributing and symlinking the library files.

This example shows you how to distribute a shared library in Unix-like systems, mylib.so, and load it from a MapReduce task.

  1. First copy the library to the HDFS: bin/hadoop fs -copyFromLocal libmyexample.so.1 /libraries/libmyexample.so.1
  2. The job launching program should contain the following: DistributedCache.createSymlink(conf); DistributedCache.addCacheFile("hdfs://host:port/libraries/libmyexample.so.1#libmyexample.so", conf);
  3. The MapReduce task can contain: System.loadLibrary("myexample");

Note: If you downloaded or built the native hadoop library, you don’t need to use DistibutedCache to make the library available to your MapReduce tasks.