今天尝试在Hadoop 2.x(YARN)上安装和配置LZO，遇到了很多坑，网上的资料都是基于Hadoop 1.x的，基本没有对于Hadoop 2.x上应用LZO，我在这边记录整个安装配置过程
1. 安装LZO
下载lzo 2.06版本，编译64位版本，同步到集群中
wget http://www.oberhumer.com/opensource/lzo/download/lzo-2.06.tar.gz export CFLAGS=-m64 ./configure -enable-shared -prefix=/usr/local/hadoop/lzo/ make && make test && make install
同步 /usr/local/hadoop/lzo/到整个集群上
2. 安装Hadoop-LZO
注意，Hadoop 1.x的时候我们是直接按照cloudera的文档clone https://github.com/kevinweil/hadoop-lzo.git上编译的，它是fork自https://github.com/twitter/hadoop-lzo。
但是kevinweil这个版本已经很久没有更新了，而且它是基于Hadoop 1.x去编译的，不能用于Hadoop 2.x。而twitter/hadoop-lzo三个月将Ant的编译方式切换为Maven，默认的dependency中Hadoop jar包就是2.x的，所以要clone twitter的hadoop-lzo，用Maven编译jar包和native library。
编译前先想pom中的hadoop-common和hadoop-mapreduce-client-core版本号改成2.1.0-beta
git clone https://github.com/twitter/hadoop-lzo.git export CFLAGS=-m64 export CXXFLAGS=-m64 export C_INCLUDE_PATH=/usr/local/hadoop/lzo/include export LIBRARY_PATH=/usr/local/hadoop/lzo/lib mvn clean package -Dmaven.test.skip=true tar -cBf - -C target/native/Linux-amd64-64/lib . | tar -xBvf - -C /usr/local/hadoop/hadoop-2.1.0-beta/lib/native/ cp target/hadoop-lzo-0.4.18-SNAPSHOT.jar /usr/local/hadoop/hadoop-2.1.0-beta/share/hadoop/common/
-rw-r--r-- 1 hadoop hadoop 104206 Sep 2 10:44 libgplcompression.a -rw-rw-r-- 1 hadoop hadoop 1121 Sep 2 10:44 libgplcompression.la lrwxrwxrwx 1 hadoop hadoop 26 Sep 2 10:47 libgplcompression.so -> libgplcompression.so.0.0.0 lrwxrwxrwx 1 hadoop hadoop 26 Sep 2 10:47 libgplcompression.so.0 -> libgplcompression.so.0.0.0 -rwxrwxr-x 1 hadoop hadoop 67833 Sep 2 10:44 libgplcompression.so.0.0.0 -rw-rw-r-- 1 hadoop hadoop 835968 Aug 29 17:12 libhadoop.a -rw-rw-r-- 1 hadoop hadoop 1482132 Aug 29 17:12 libhadooppipes.a lrwxrwxrwx 1 hadoop hadoop 18 Aug 29 17:12 libhadoop.so -> libhadoop.so.1.0.0 -rwxrwxr-x 1 hadoop hadoop 465801 Aug 29 17:12 libhadoop.so.1.0.0 -rw-rw-r-- 1 hadoop hadoop 580384 Aug 29 17:12 libhadooputils.a -rw-rw-r-- 1 hadoop hadoop 273642 Aug 29 17:12 libhdfs.a lrwxrwxrwx 1 hadoop hadoop 16 Aug 29 17:12 libhdfs.so -> libhdfs.so.0.0.0 -rwxrwxr-x 1 hadoop hadoop 181171 Aug 29 17:12 libhdfs.so.0.0.0将 hadoop-lzo-0.4.18-SNAPSHOT.jar和/usr/local/hadoop/hadoop-2.1.0-beta/lib/native/ 同步到整个集群中
3. 设置环境变量
在hadoop-env.sh中加入
export LD_LIBRARY_PATH=/usr/local/hadoop/lzo/lib
core-site加入
<property> <name>io.compression.codecs</name> <value>org.apache.hadoop.io.compress.GzipCodec,org.apache.hadoop.io.compress.DefaultCodec,com.hadoop.compression.lzo.LzoCodec,com.hadoop.compression.lzo.LzopCodec,org.apache.hadoop.io.compress.BZip2Codec</value> </property> <property> <name>io.compression.codec.lzo.class</name> <value>com.hadoop.compression.lzo.LzoCodec</value> </property>
mapred-site.xml加入
<property> <name>mapred.compress.map.output</name> <value>true</value> </property> <property> <name>mapred.map.output.compression.codec</name> <value>com.hadoop.compression.lzo.LzoCodec</value> </property> <property> <name>mapred.child.env</name> <value>LD_LIBRARY_PATH=/usr/local/hadoop/lzo/lib</value> </property>
其中mapred-site中设置mapred.child.env的 LD_LIBRARY_PATH很重要，因为hadoop-lzo通过JNI调用(java.library.path) libgplcompression.so，然后libgplcompression.so 再通过dlopen这个系统调用（其实是查找系统环境变量LD_LIBRARY_PATH ）来加载liblzo2.so 。container在启动的时候，需要设置LD_LIBRARY_PATH 环境变量，来让LzoCodec加载 native-lzo library，如果不设置的话，会在container的syslog中报下面的错误
2013-09-02 11:20:12,004 INFO [main] com.hadoop.compression.lzo.GPLNativeCodeLoader: Loaded native gpl library 2013-09-02 11:20:12,006 WARN [main] com.hadoop.compression.lzo.LzoCompressor: java.lang.UnsatisfiedLinkError: Cannot load liblzo2.so.2 (liblzo2.so.2: cannot open shared object file: No such file or directory)! 2013-09-02 11:20:12,006 ERROR [main] com.hadoop.compression.lzo.LzoCodec: Failed to load/initialize native-lzo library同步hadoop-env.sh, core-site.xml, mapred-site.xml到集群
LzoCodec加载gplcompression和lzo native library
static {
if (GPLNativeCodeLoader.isNativeCodeLoaded()) {
nativeLzoLoaded = LzoCompressor.isNativeLzoLoaded() &&
LzoDecompressor.isNativeLzoLoaded();
if (nativeLzoLoaded) {
LOG.info("Successfully loaded & initialized native-lzo library [hadoop-lzo rev " + getRevisionHash() + "]");
} else {
LOG.error("Failed to load/initialize native-lzo library");
}
} else {
LOG.error("Cannot load native-lzo without native-hadoop");
}
}
LzoCompressor和LzoDecompressor会调用本地方法initIDs
在impl/lzo/LzoCompressor.c中加载liblzo2.so
Java_com_hadoop_compression_lzo_LzoCompressor_initIDs(
JNIEnv *env, jclass class
) {
// Load liblzo2.so
liblzo2 = dlopen(HADOOP_LZO_LIBRARY, RTLD_LAZY | RTLD_GLOBAL);
if (!liblzo2) {
char* msg = (char*)malloc(1000);
snprintf(msg, 1000, "%s (%s)!", "Cannot load " HADOOP_LZO_LIBRARY, dlerror());
THROW(env, "java/lang/UnsatisfiedLinkError", msg);
return;
}
LzoCompressor_clazz = (*env)->GetStaticFieldID(env, class, "clazz",
"Ljava/lang/Class;");
LzoCompressor_finish = (*env)->GetFieldID(env, class, "finish", "Z");
LzoCompressor_finished = (*env)->GetFieldID(env, class, "finished", "Z");
LzoCompressor_uncompressedDirectBuf = (*env)->GetFieldID(env, class,
"uncompressedDirectBuf",
"Ljava/nio/ByteBuffer;");
LzoCompressor_uncompressedDirectBufLen = (*env)->GetFieldID(env, class,
"uncompressedDirectBufLen", "I");
LzoCompressor_compressedDirectBuf = (*env)->GetFieldID(env, class,
"compressedDirectBuf",
"Ljava/nio/ByteBuffer;");
LzoCompressor_directBufferSize = (*env)->GetFieldID(env, class,
"directBufferSize", "I");
LzoCompressor_lzoCompressor = (*env)->GetFieldID(env, class,
"lzoCompressor", "J");
LzoCompressor_lzoCompressionLevel = (*env)->GetFieldID(env, class,
"lzoCompressionLevel", "I");
LzoCompressor_workingMemoryBufLen = (*env)->GetFieldID(env, class,
"workingMemoryBufLen", "I");
LzoCompressor_workingMemoryBuf = (*env)->GetFieldID(env, class,
"workingMemoryBuf",
"Ljava/nio/ByteBuffer;");
// record lzo library version
void* lzo_version_ptr = NULL;
LOAD_DYNAMIC_SYMBOL(lzo_version_ptr, env, liblzo2, "lzo_version");
liblzo2_version = (NULL == lzo_version_ptr) ? 0
: (jint) ((unsigned (__LZO_CDECL *)())lzo_version_ptr)();
}
创建container上下文信息的时候，会读取mapred.child.env作为子进程环境变量的一部分
4. 测试mapreduce读lzo
hive新建一张表lzo_test
CREATE TABLE lzo_test( col String ) STORED AS INPUTFORMAT "com.hadoop.mapred.DeprecatedLzoTextInputFormat" OUTPUTFORMAT "org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat";下载lzop工具，load一个lzo文件进lzo_test表中，执行“select * from lzo_test"和"select count(1) from lzo_test"正确
同时用户可以通过单机作业或者分布式程序生成lzo.index文件
hadoop jar /usr/local/hadoop/hadoop-2.1.0-beta/share/hadoop/common/hadoop-lzo-0.4.18-SNAPSHOT.jar com.hadoop.compression.lzo.DistributedLzoIndexer /user/hive/warehouse/lzo_test/ hadoop jar /usr/local/hadoop/hadoop-2.1.0-beta/share/hadoop/common/hadoop-lzo-0.4.18-SNAPSHOT.jar com.hadoop.compression.lzo.LzoIndexer /user/hive/warehouse/lzo_test/