diff --git a/yarn/src/org/apache/hadoop/yarn/rabit/ApplicationMaster.java b/yarn/src/org/apache/hadoop/yarn/rabit/ApplicationMaster.java index f51502018..46787ba09 100644 --- a/yarn/src/org/apache/hadoop/yarn/rabit/ApplicationMaster.java +++ b/yarn/src/org/apache/hadoop/yarn/rabit/ApplicationMaster.java @@ -216,12 +216,14 @@ public class ApplicationMaster { assert (killedTasks.size() + finishedTasks.size() == numTasks); success = finishedTasks.size() == numTasks; LOG.info("Application completed. Stopping running containers"); - if (success) { - nmClient.stop(); - } diagnostics = "Diagnostics." + ", num_tasks" + this.numTasks + ", finished=" + this.finishedTasks.size() + ", failed=" + this.killedTasks.size() + "\n" + this.abortDiagnosis; + if (success) { + nmClient.stop(); + } else { + throw new Exception("Application not successful" + diagnostics); + } LOG.info(diagnostics); } catch (Exception e) { diagnostics = e.toString(); diff --git a/yarn/src/org/apache/hadoop/yarn/rabit/Client.java b/yarn/src/org/apache/hadoop/yarn/rabit/Client.java index 57d3e9889..63f92d8be 100644 --- a/yarn/src/org/apache/hadoop/yarn/rabit/Client.java +++ b/yarn/src/org/apache/hadoop/yarn/rabit/Client.java @@ -198,7 +198,7 @@ public class Client { amContainer.setLocalResources(this.setupCacheFiles(appId)); amContainer.setEnvironment(this.getEnvironment()); String cmd = "$JAVA_HOME/bin/java" - + " -Xmx256M" + + " -Xmx900M" + " org.apache.hadoop.yarn.rabit.ApplicationMaster" + this.cacheFileArg + ' ' + this.appArgs + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stdout" @@ -208,7 +208,7 @@ public class Client { // Set up resource type requirements for ApplicationMaster Resource capability = Records.newRecord(Resource.class); - capability.setMemory(256); + capability.setMemory(1024); capability.setVirtualCores(1); LOG.info("jobname=" + this.jobName + ",username=" + this.userName);