YARN-6533. Race condition in writing service record to registry in yarn native services. Contributed by Billie Rinaldi

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-applications/hadoop-yarn-slider/hadoop-yarn-slider-core/src/main/java/org/apache/slider/providers/ProviderUtils.java

@@ -503,7 +503,8 @@ public class ProviderUtils implements RoleKeys, SliderKeys {
    */
   public void updateServiceRecord(StateAccessForProviders amState,
       YarnRegistryViewForProviders yarnRegistry,
-      String containerId, String roleName, List<String> ip, String hostname) {
+      String containerId, String roleName, List<String> ip, String hostname)
+      throws IOException {
     try {
       RoleInstance role = null;
       if(ip != null && !ip.isEmpty()){
@@ -535,9 +536,6 @@ public class ProviderUtils implements RoleKeys, SliderKeys {
     } catch (NoSuchNodeException e) {
       // ignore - there is nothing to do if we don't find a container
       log.warn("Owned container {} not found - {}", containerId, e);
-    } catch (IOException e) {
-      log.warn("Error updating container {} service record in registry",
-          containerId, e);
     }
   }
 }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-applications/hadoop-yarn-slider/hadoop-yarn-slider-core/src/main/java/org/apache/slider/providers/docker/DockerProviderService.java

@@ -146,8 +146,15 @@ public class DockerProviderService extends AbstractService
       return false;
     }
 
-    providerUtils.updateServiceRecord(amState, yarnRegistry,
+    try {
-        containerId.toString(), instance.role, status.getIPs(), status.getHost());
+      providerUtils.updateServiceRecord(amState, yarnRegistry,
+          containerId.toString(), instance.role, status.getIPs(), status.getHost());
+    } catch (IOException e) {
+      // could not write service record to ZK, log and retry
+      log.warn("Error updating container {} service record in registry, " +
+          "retrying", containerId, e);
+      return true;
+    }
     // TODO publish ip and host
     org.apache.slider.api.resource.Container container =
         instance.providerRole.component.getContainer(containerId.toString());

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-applications/hadoop-yarn-slider/hadoop-yarn-slider-core/src/main/java/org/apache/slider/server/appmaster/SliderAppMaster.java

@@ -1182,24 +1182,8 @@ public class SliderAppMaster extends AbstractSliderLaunchedService
       return false;
     }
     // this is where component registrations  go
-    String cid = RegistryPathUtils.encodeYarnID(id.toString());
-    ServiceRecord record = new ServiceRecord();
-    record.set(YarnRegistryAttributes.YARN_ID, cid);
-
-    record.description = roleInstance.getCompInstanceName();
     log.info("Registering component " + roleInstance.getCompInstanceName()
         + ", containerId = " + id);
-    record.set(YarnRegistryAttributes.YARN_PERSISTENCE,
-        PersistencePolicies.CONTAINER);
-    setUserProvidedServiceRecordAttributes(
-        instance.providerRole.component.getConfiguration(), record);
-    try {
-      yarnRegistryOperations.putComponent(cid, record);
-    } catch (IOException e) {
-      log.warn("Failed to register container {}/{}: {}",
-          id, roleInstance.role, e, e);
-      return false;
-    }
     org.apache.slider.api.resource.Container container =
         new org.apache.slider.api.resource.Container();
     container.setId(id.toString());