NSX: Handle CheckHealthCommand to avoid host disconnection and errors…

… on APIs * NSX: Handle CheckHealthCommand to avoid host disconnection and errors on APIs * Remove unused string
apache · nvazquez · Sep 6, 2024 · Mar 11, 2024 · Mar 28, 2024 · Apr 10, 2024
commit e79d77199a4399aecf466c018349dc621ad17c42
@@ -18,6 +18,8 @@
 
 import com.cloud.agent.IAgentControl;
 import com.cloud.agent.api.Answer;
+import com.cloud.agent.api.CheckHealthAnswer;
+import com.cloud.agent.api.CheckHealthCommand;
 import com.cloud.agent.api.Command;
 import com.cloud.agent.api.PingCommand;
 import com.cloud.agent.api.ReadyAnswer;
@@ -102,6 +104,8 @@ public PingCommand getCurrentStatus(long id) {
     public Answer executeRequest(Command cmd) {
         if (cmd instanceof ReadyCommand) {
             return executeRequest((ReadyCommand) cmd);
+        } else if (cmd instanceof CheckHealthCommand) {
+            return executeRequest((CheckHealthCommand) cmd);
         } else if (cmd instanceof DeleteNsxTier1GatewayCommand) {
             return executeRequest((DeleteNsxTier1GatewayCommand) cmd);
         } else if (cmd instanceof DeleteNsxSegmentCommand) {
@@ -293,6 +297,10 @@ private Answer executeRequest(ReadyCommand cmd) {
         return new ReadyAnswer(cmd);
     }
 
+    private Answer executeRequest(CheckHealthCommand cmd) {
+        return new CheckHealthAnswer(cmd, nsxApiClient.isNsxControllerActive());
+    }
+
     private Answer executeRequest(CreateNsxTier1GatewayCommand cmd) {
         String tier1GatewayName = NsxControllerUtils.getTier1GatewayName(cmd.getDomainId(), cmd.getAccountId(), cmd.getZoneId(), cmd.getNetworkResourceId(), cmd.isResourceVpc());
         boolean sourceNatEnabled = cmd.isSourceNatEnabled();

@@ -18,6 +18,9 @@
 
 import com.cloud.network.Network;
 import com.cloud.utils.exception.CloudRuntimeException;
+import com.vmware.nsx.cluster.Status;
+import com.vmware.nsx.model.ClusterStatus;
+import com.vmware.nsx.model.ControllerClusterStatus;
 import com.vmware.nsx.model.TransportZone;
 import com.vmware.nsx.model.TransportZoneListResult;
 import com.vmware.nsx_policy.infra.DhcpRelayConfigs;
@@ -113,6 +116,7 @@ public class NsxApiClient {
     protected Logger logger = LogManager.getLogger(getClass());
 
     // Constants
+    private static final String CLUSTER_STATUS_STABLE = "STABLE";
     private static final String TIER_1_RESOURCE_TYPE = "Tier1";
     private static final String TIER_1_LOCALE_SERVICE_ID = "default";
     private static final String SEGMENT_RESOURCE_TYPE = "Segment";
@@ -200,6 +204,26 @@ public NsxApiClient(String hostname, String port, String username, char[] passwo
         nsxService = apiClient::createStub;
     }
 
+    public boolean isNsxControllerActive() {
+        try {
+            Status statusService = (Status) nsxService.apply(Status.class);
+            ClusterStatus clusterStatus = statusService.get();
+            if (clusterStatus == null) {
+                LOGGER.error("Cannot get NSX Cluster Status");
+                return false;
+            }
+            ControllerClusterStatus status = clusterStatus.getControlClusterStatus();
+            if (status == null) {
+                LOGGER.error("Cannot get NSX Controller Cluster Status");
+                return false;
+            }
+            return CLUSTER_STATUS_STABLE.equalsIgnoreCase(status.getStatus());
+        } catch (Error error) {
+            LOGGER.error(String.format("Error checking NSX Controller Health: %s", error.getMessage()));
+            return false;
+        }
+    }
+
     public void createTier1NatRule(String tier1GatewayName, String natId, String natRuleId,
                                    String action, String translatedIp) {
         NatRules natRulesService = (NatRules) nsxService.apply(NatRules.class);

@@ -17,6 +17,9 @@
 package org.apache.cloudstack.service;
 
 import com.cloud.network.Network;
+import com.vmware.nsx.cluster.Status;
+import com.vmware.nsx.model.ClusterStatus;
+import com.vmware.nsx.model.ControllerClusterStatus;
 import com.vmware.nsx_policy.infra.domains.Groups;
 import com.vmware.nsx_policy.model.Group;
 import com.vmware.nsx_policy.model.PathExpression;
@@ -93,4 +96,16 @@ public void testGetGroupsForTrafficEgress() {
         Assert.assertEquals(List.of(String.format("%s/%s", NsxApiClient.GROUPS_PATH_PREFIX, segmentName)), sourceGroups);
         Assert.assertEquals(List.of("ANY"), destinationGroups);
     }
+
+    @Test
+    public void testIsNsxControllerActive() {
+        Status statusService = Mockito.mock(Status.class);
+        Mockito.when(nsxService.apply(Status.class)).thenReturn(statusService);
+        ClusterStatus clusterStatus = Mockito.mock(ClusterStatus.class);
+        ControllerClusterStatus status = Mockito.mock(ControllerClusterStatus.class);
+        Mockito.when(status.getStatus()).thenReturn("stable");
+        Mockito.when(statusService.get()).thenReturn(clusterStatus);
+        Mockito.when(clusterStatus.getControlClusterStatus()).thenReturn(status);
+        Assert.assertTrue(client.isNsxControllerActive());
+    }
 }