Container exited with a non-zero exit code 1-编程知识

Container exited with a non-zero exit code 1

最近遇到运行yarn pi的时候遇到如下问题。

很明显是container出错了，但是错误没有提示的很清楚。然后去看nodemanager日志也是如此。这时候笔者第一个想到要去看container的执行日志。container具体的日志目录位置是通过YARN的配置文件（如yarn-site.xml）中的yarn.nodemanager.log-dirs属性定义的，默认为$HADOOP_HOME/logs/userlogs。去对应的nodemanager的userlogs下发现没有任何日志，查看nodemanager日志发现如果任务失败了，默认会删除container日志。

笔者觉得可以改一下源码，让他先不删日志。

DefaultContainerExecutor.java

  @Overridepublic void deleteAsUser(DeletionAsUserContext ctx)throws IOException, InterruptedException {Path subDir = ctx.getSubDir();List<Path> baseDirs = ctx.getBasedirs();if (baseDirs == null || baseDirs.size() == 0) {LOG.info("Deleting absolute path : " + subDir);//if (!lfs.delete(subDir, true)) {//Maybe retryLOG.warn("delete returned false for path: [" + subDir + "]");//}return;}for (Path baseDir : baseDirs) {Path del = subDir == null ? baseDir : new Path(baseDir, subDir);LOG.info("Deleting path : " + del);//try {//if (!lfs.delete(del, true)) {//LOG.warn("delete returned false for path: [" + del + "]");//}//} catch (FileNotFoundException e) {//continue;//}}}