Speed up ci test when am crashed (#594)

Signed-off-by: zhangjunfan <[email protected]> Signed-off-by: zhangjunfan <[email protected]>
tony-framework · Sep 2, 2021 · c2b67f7 · c2b67f7
1 parent bde944c
commit c2b67f7
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 2 deletions.
diff --git a/tony-core/src/main/java/com/linkedin/tony/TonyClient.java b/tony-core/src/main/java/com/linkedin/tony/TonyClient.java
@@ -1078,7 +1078,6 @@ public boolean monitorApplication() throws YarnException, IOException, Interrupt
       }
 
       if (YarnApplicationState.FINISHED == appState || YarnApplicationState.FAILED == appState) {
-        updateTaskInfoAndReturn();
         LOG.info("-----  Application finished, status of ALL tasks -----");
         // log detailed task info including URL so that users can check the URL of failed worker
         // quickly without the need to scroll up to the top to find out the URL.
@@ -1089,6 +1088,7 @@ public boolean monitorApplication() throws YarnException, IOException, Interrupt
             tonyConf.get(TonyConfigurationKeys.TONY_PORTAL_URL, TonyConfigurationKeys.DEFAULT_TONY_PORTAL_URL);
         Utils.printTonyPortalUrl(tonyPortalUrl, appId.toString(), LOG);
         result = FinalApplicationStatus.SUCCEEDED == finalApplicationStatus;
+        signalAMToFinish();
         break;
       }
 
@@ -1103,7 +1103,6 @@ public boolean monitorApplication() throws YarnException, IOException, Interrupt
       }
     }
 
-    signalAMToFinish();
     return result;
   }
 

diff --git a/tony-mini/src/main/java/com/linkedin/minitony/cluster/MiniCluster.java b/tony-mini/src/main/java/com/linkedin/minitony/cluster/MiniCluster.java
@@ -56,6 +56,11 @@ public void start() throws Exception {
     yarnClusterConf = yarnCluster.getConfig();
     hdfsClusterConf = dfsCluster.getConfiguration(0);
     yarnClusterConf.setBoolean("ipc.client.fallback-to-simple-auth-allowed", true);
+    yarnClusterConf.setInt("ipc.client.connection.maxidletime", 1000);
+    yarnClusterConf.setInt("ipc.client.connect.max.retries", 3);
+    yarnClusterConf.setInt("ipc.client.connect.retry.interval", 10);
+    yarnClusterConf.setInt("ipc.client.connect.timeout", 1000);
+    yarnClusterConf.setInt("ipc.client.connect.max.retries.on.timeouts", 3);
     hdfsClusterConf.setBoolean("ipc.client.fallback-to-simple-auth-allowed", true);
   }