在 Google Dataproc Spark 中鏈接原生 .so

Question

我有一個本地庫（Freeling），我使用 cmake 編譯並本地生成並通過集群啟動操作安裝（因此，它應該存在於 master 和每個 worker 中）

即便如此，我在調用System.loadLibrary時收到此錯誤

Exception in thread "main" java.lang.UnsatisfiedLinkError: no Jfreeling in java.library.path
    at java.lang.ClassLoader.loadLibrary(ClassLoader.java:1867)
    at java.lang.Runtime.loadLibrary0(Runtime.java:870)
    at java.lang.System.loadLibrary(System.java:1122)

我嘗試使用以下屬性讓程序找到庫（它在靜態代碼塊中調用）

      "properties": {
        "spark.driver.extraClassPath": "/usr/local/share/freeling/APIs/java/Jfreeling.jar:/usr/local/lib/libfreeling.so",
        "spark.executor.extraClassPath": "/usr/local/share/freeling/APIs/java/Jfreeling.jar:/usr/local/lib/libfreeling.so",
        "spark.executor.extraLibraryPath": "/usr/local/lib/libfreeling.so",
        "spark.driver.extraLibraryPath": "/usr/local/lib/libfreeling.so",
        "spark.executorEnv.LD_PRELOAD": "/usr/local/lib/libfreeling.so",
        "spark.yarn.dist.files": "/usr/local/lib/libfreeling.so",
        "spark.yarn.appMasterEnv.LD_PRELOAD": "libfreeling.so",
        "spark.files": "/usr/local/lib/libfreeling.so",
        "spark.executorEnv.LD_LIBRARY_PATH": "libfreeling.so"
      },
      "jarFileUris": [
        "file:///usr/local/share/freeling/APIs/java/Jfreeling.jar",
        "file:///usr/local/lib/libfreeling.so"
      ],

Answer 1

您可以嘗試將您的庫放在/usr/lib/hadoop/lib/native/嗎？ 在/etc/spark/conf/spark-env.sh ，它有

# Spark got rid of SPARK_LIBRARY_PATH in 1.0
# It has properties for extraLibraryPaths, but this is more extensible
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${HADOOP_HOME}/lib/native

Answer 2

您應該將 /usr/local/share/freeling/APIs/java/Jfreeling.jar 添加到您的 CLASSPATH，並將 /usr/local/share/freeling/APIs/java/libJfreeling.so 添加到您的 LD_LIBRARY_PATH。