python - 在 YARN 上运行 Spark 作业

Question

我正在尝试利用集群上的所有资源来运行 Spark 作业。我在所有节点上都安装了 Cloudera Manager。这是我用来提交作业的命令。

spark-submit --master yarn 
             --deploy-mode cluster 
             file:///[spark python file]
             file://[app argument 1]
             file://[app argument 2]

在执行期间我收到以下错误:

diagnostics: Application application_1450777964379_0027 failed 2 times due to AM Container for appattempt_1450777964379_0027_000002 exited with  exitCode: 1

任何修复方法的想法将不胜感激。

编辑 1 安装 Spark 的机器无法通过 WEB UI 访问我尝试下载源代码并阅读更多关于异常的信息。

------------------------------------------------------------
| Job | Description                                        | 
------------------------------------------------------------
| 0   | saveAsTextFile at NativeMethodAccessorImpl.java:-2 | 
------------------------------------------------------------

Answer 1

取自here ,

If the path starts with file:// or hdfs://, the path becomes ile:// or dfs://. If the path is absolute the first slash is removed.

There is no particular reason for it and needs to be fixed.

尝试使用绝对路径而不是 file://