Support for Management Server Maintenance Mode

agent/src/main/java/com/cloud/agent/Agent.java

@@ -27,6 +27,7 @@
 import java.nio.channels.ClosedChannelException;
 import java.nio.charset.Charset;
 import java.util.ArrayList;
+import java.util.Arrays;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
@@ -40,6 +41,8 @@
 
 import javax.naming.ConfigurationException;
 
+import com.cloud.agent.api.MigrateAgentConnectionAnswer;
+import com.cloud.agent.api.MigrateAgentConnectionCommand;
 import com.cloud.resource.AgentStatusUpdater;
 import com.cloud.resource.ResourceStatusUpdater;
 import com.cloud.agent.api.PingAnswer;
@@ -313,7 +316,6 @@ public void start() {
         }
         _shell.updateConnectedHost();
         scavengeOldAgentObjects();
-
     }
 
     public void stop(final String reason, final String detail) {
@@ -477,13 +479,18 @@ public synchronized void lockStartupTask(final Link link) {
     }
 
     public void sendStartup(final Link link) {
+        sendStartup(link, false);
+    }
+
+    public void sendStartup(final Link link, boolean transfer) {
         final StartupCommand[] startup = _resource.initialize();
         if (startup != null) {
             final String msHostList = _shell.getPersistentProperty(null, "host");
             final Command[] commands = new Command[startup.length];
             for (int i = 0; i < startup.length; i++) {
                 setupStartupCommand(startup[i]);
                 startup[i].setMSHostList(msHostList);
+                startup[i].setConnectionTransferred(transfer);
                 commands[i] = startup[i];
             }
             final Request request = new Request(_id != null ? _id : -1, -1, commands, false, false);
@@ -541,9 +548,14 @@ public Task create(final Task.Type type, final Link link, final byte[] data) {
     }
 
     protected void reconnect(final Link link) {
-        if (!_reconnectAllowed) {
+        reconnect(link, null, null, false);
+    }
+
+    protected void reconnect(final Link link, String preferredHost, List<String> avoidHostList, boolean forTransfer) {
+        if (!(forTransfer || _reconnectAllowed)) {
             return;
         }
+
         synchronized (this) {
             if (_startup != null) {
                 _startup.cancel();
@@ -575,22 +587,29 @@ protected void reconnect(final Link link) {
             _shell.getBackoffAlgorithm().waitBeforeRetry();
         }
 
+        String host = preferredHost;
+        if (StringUtils.isEmpty(host)) {
+            host = _shell.getNextHost();
+        }
+
         do {
-            final String host = _shell.getNextHost();
-            _connection = new NioClient("Agent", host, _shell.getPort(), _shell.getWorkers(), this);
-            logger.info("Reconnecting to host:{}", host);
-            try {
-                _connection.start();
-            } catch (final NioConnectionException e) {
-                logger.info("Attempted to re-connect to the server, but received an unexpected exception, trying again...", e);
-                _connection.stop();
+            if (CollectionUtils.isEmpty(avoidHostList) || !avoidHostList.contains(host)) {
+                _connection = new NioClient("Agent", host, _shell.getPort(), _shell.getWorkers(), this);
+                logger.info("Reconnecting to host:{}", host);
                 try {
-                    _connection.cleanUp();
-                } catch (final IOException ex) {
-                    logger.warn("Fail to clean up old connection. {}", ex);
+                    _connection.start();
+                } catch (final NioConnectionException e) {
+                    logger.info("Attempted to re-connect to the server, but received an unexpected exception, trying again...", e);
+                    _connection.stop();
+                    try {
+                        _connection.cleanUp();
+                    } catch (final IOException ex) {
+                        logger.warn("Fail to clean up old connection. {}", ex);
+                    }
                 }
             }
             _shell.getBackoffAlgorithm().waitBeforeRetry();
+            host = _shell.getNextHost();
         } while (!_connection.isStartup());
         _shell.updateConnectedHost();
         logger.info("Connected to the host: {}", _shell.getConnectedHost());
@@ -703,6 +722,8 @@ protected void processRequest(final Request request, final Link link) {
                         }
                     } else if (cmd instanceof SetupMSListCommand) {
                         answer = setupManagementServerList((SetupMSListCommand) cmd);
+                    } else if (cmd instanceof MigrateAgentConnectionCommand) {
+                        answer = migrateAgentToOtherMS((MigrateAgentConnectionCommand) cmd);
                     } else {
                         if (cmd instanceof ReadyCommand) {
                             processReadyCommand(cmd);
@@ -858,6 +879,53 @@ private Answer setupManagementServerList(final SetupMSListCommand cmd) {
         return new SetupMSListAnswer(true);
     }
 
+    private Answer migrateAgentToOtherMS(final MigrateAgentConnectionCommand cmd) {
+        try {
+            if (CollectionUtils.isNotEmpty(cmd.getMsList())) {
+                processManagementServerList(cmd.getMsList(), cmd.getLbAlgorithm(), cmd.getLbCheckInterval());
+            }
+            migrateAgentConnection(cmd.getAvoidMsList());
+        } catch (Exception e) {
+            String errMsg = "Migrate agent connection failed, due to " + e.getMessage();
+            logger.debug(errMsg, e);
+            return new MigrateAgentConnectionAnswer(errMsg);
+        }
+        return new MigrateAgentConnectionAnswer(true);
+    }
+
+    private void migrateAgentConnection(List<String> avoidMsList) {
+        final String[] msHosts = _shell.getHosts();
+        if (msHosts == null || msHosts.length < 1) {
+            throw new CloudRuntimeException("Management Server hosts empty, not properly configured in agent");
+        }
+
+        List<String> msHostsList = new ArrayList<>(Arrays.asList(msHosts));
+        msHostsList.removeAll(avoidMsList);
+        if (msHostsList.isEmpty() || StringUtils.isEmpty(msHostsList.get(0))) {
+            throw new CloudRuntimeException("No other Management Server hosts to migrate");
+        }
+
+        String preferredHost  = null;
+        for (String msHost : msHostsList) {
+            try (final Socket socket = new Socket()) {
+                socket.connect(new InetSocketAddress(msHost, _shell.getPort()), 5000);
+                preferredHost = msHost;
+                break;
+            } catch (final IOException e) {
+                throw new CloudRuntimeException("Management server host: " + msHost + " is not reachable, to migrate connection");
+            }
+        }
+
+        if (preferredHost == null) {
+            throw new CloudRuntimeException("Management server host(s) are not reachable, to migrate connection");
+        }
+
+        logger.debug("Management server host " + preferredHost + " is found to be reachable, trying to reconnect");
+        _shell.resetHostCounter();
+        _shell.setConnectionTransfer(true);
+        reconnect(_link, preferredHost, avoidMsList, true);
+    }
+
     public void processResponse(final Response response, final Link link) {
         final Answer answer = response.getAnswer();
         logger.debug("Received response: {}", response.toString());
@@ -1153,7 +1221,8 @@ public void doTask(final Task task) throws TaskExecutionException {
             if (task.getType() == Task.Type.CONNECT) {
                 _shell.getBackoffAlgorithm().reset();
                 setLink(task.getLink());
-                sendStartup(task.getLink());
+                sendStartup(task.getLink(), _shell.isConnectionTransfer());
+                _shell.setConnectionTransfer(false);
             } else if (task.getType() == Task.Type.DATA) {
                 Request request;
                 try {
@@ -1178,6 +1247,7 @@ public void doTask(final Task task) throws TaskExecutionException {
                     Thread.sleep(5000);
                 } catch (InterruptedException e) {
                 }
+                _shell.setConnectionTransfer(false);
                 reconnect(task.getLink());
                 return;
             } else if (task.getType() == Task.Type.OTHER) {

agent/src/main/java/com/cloud/agent/AgentShell.java

@@ -77,6 +77,7 @@ public class AgentShell implements IAgentShell, Daemon {
     private String hostToConnect;
     private String connectedHost;
     private Long preferredHostCheckInterval;
+    private boolean connectionTransfer = false;
     protected AgentProperties agentProperties = new AgentProperties();
 
     public AgentShell() {
@@ -215,6 +216,14 @@ public void setPersistentProperty(String prefix, String name, String value) {
             _storage.persist(name, value);
     }
 
+    public boolean isConnectionTransfer() {
+        return connectionTransfer;
+    }
+
+    public void setConnectionTransfer(boolean connectionTransfer) {
+        this.connectionTransfer = connectionTransfer;
+    }
+
     void loadProperties() throws ConfigurationException {
         final File file = PropertiesUtil.findConfigFile("agent.properties");
 

agent/src/main/java/com/cloud/agent/IAgentShell.java

@@ -70,4 +70,8 @@ public interface IAgentShell {
     String getConnectedHost();
 
     void launchNewAgent(ServerResource resource) throws ConfigurationException;
+
+    boolean isConnectionTransfer();
+
+    void setConnectionTransfer(boolean connectionTransfer);
 }

api/src/main/java/com/cloud/host/Host.java

@@ -177,6 +177,8 @@ public static String[] toStrings(Host.Type... types) {
      */
     Long getManagementServerId();
 
+    Long getLastManagementServerId();
+
     /*
      *@return removal date
      */

api/src/main/java/com/cloud/host/Status.java

@@ -127,6 +127,7 @@ public static String[] toStrings(Status... states) {
         s_fsm.addTransition(Status.Connecting, Event.HostDown, Status.Down);
         s_fsm.addTransition(Status.Connecting, Event.Ping, Status.Connecting);
         s_fsm.addTransition(Status.Connecting, Event.ManagementServerDown, Status.Disconnected);
+        s_fsm.addTransition(Status.Connecting, Event.StartAgentRebalance, Status.Rebalancing);
         s_fsm.addTransition(Status.Connecting, Event.AgentDisconnected, Status.Alert);
         s_fsm.addTransition(Status.Up, Event.PingTimeout, Status.Alert);
         s_fsm.addTransition(Status.Up, Event.AgentDisconnected, Status.Alert);

api/src/main/java/com/cloud/resource/ResourceService.java

@@ -23,11 +23,11 @@
 import org.apache.cloudstack.api.command.admin.cluster.UpdateClusterCmd;
 import org.apache.cloudstack.api.command.admin.host.AddHostCmd;
 import org.apache.cloudstack.api.command.admin.host.AddSecondaryStorageCmd;
-import org.apache.cloudstack.api.command.admin.host.CancelMaintenanceCmd;
+import org.apache.cloudstack.api.command.admin.host.CancelHostMaintenanceCmd;
 import org.apache.cloudstack.api.command.admin.host.ReconnectHostCmd;
 import org.apache.cloudstack.api.command.admin.host.UpdateHostCmd;
 import org.apache.cloudstack.api.command.admin.host.UpdateHostPasswordCmd;
-import org.apache.cloudstack.api.command.admin.host.PrepareForMaintenanceCmd;
+import org.apache.cloudstack.api.command.admin.host.PrepareForHostMaintenanceCmd;
 import org.apache.cloudstack.api.command.admin.host.DeclareHostAsDegradedCmd;
 import org.apache.cloudstack.api.command.admin.host.CancelHostAsDegradedCmd;
 
@@ -51,7 +51,7 @@ public interface ResourceService {
 
     Host autoUpdateHostAllocationState(Long hostId, ResourceState.Event resourceEvent) throws NoTransitionException;
 
-    Host cancelMaintenance(CancelMaintenanceCmd cmd);
+    Host cancelMaintenance(CancelHostMaintenanceCmd cmd);
 
     Host reconnectHost(ReconnectHostCmd cmd) throws AgentUnavailableException;
 
@@ -69,7 +69,7 @@ public interface ResourceService {
 
     List<? extends Host> discoverHosts(AddSecondaryStorageCmd cmd) throws IllegalArgumentException, DiscoveryException, InvalidParameterValueException;
 
-    Host maintain(PrepareForMaintenanceCmd cmd);
+    Host maintain(PrepareForHostMaintenanceCmd cmd);
 
     Host declareHostAsDegraded(DeclareHostAsDegradedCmd cmd) throws NoTransitionException;
 

api/src/main/java/com/cloud/server/ManagementServerHostStats.java

@@ -19,6 +19,7 @@
 package com.cloud.server;
 
 import java.util.Date;
+import java.util.List;
 
 /**
  * management server related stats
@@ -70,6 +71,10 @@ public interface ManagementServerHostStats {
 
     String getOsDistribution();
 
+    List<String> getLastAgents();
+
+    List<String> getAgents();
+
     int getAgentCount();
 
     long getHeapMemoryUsed();

api/src/main/java/org/apache/cloudstack/api/ApiConstants.java

@@ -1136,9 +1136,12 @@ public class ApiConstants {
     public static final String LOGOUT = "logout";
     public static final String LIST_IDPS = "listIdps";
 
-    public static final String READY_FOR_SHUTDOWN = "readyforshutdown";
+    public static final String MAINTENANCE_INITIATED = "maintenanceinitiated";
     public static final String SHUTDOWN_TRIGGERED = "shutdowntriggered";
+    public static final String READY_FOR_SHUTDOWN = "readyforshutdown";
     public static final String PENDING_JOBS_COUNT = "pendingjobscount";
+    public static final String AGENTS_COUNT = "agentscount";
+    public static final String AGENTS = "agents";
 
     public static final String PUBLIC_MTU = "publicmtu";
     public static final String PRIVATE_MTU = "privatemtu";

api/src/main/java/org/apache/cloudstack/api/command/admin/host/CancelMaintenanceCmd.java → api/src/main/java/org/apache/cloudstack/api/command/admin/host/CancelHostMaintenanceCmd.java

@@ -33,7 +33,7 @@
 
 @APICommand(name = "cancelHostMaintenance", description = "Cancels host maintenance.", responseObject = HostResponse.class,
         requestHasSensitiveInfo = false, responseHasSensitiveInfo = false)
-public class CancelMaintenanceCmd extends BaseAsyncCmd {
+public class CancelHostMaintenanceCmd extends BaseAsyncCmd {
 
 
     /////////////////////////////////////////////////////

api/src/main/java/org/apache/cloudstack/api/command/admin/host/ListHostsCmd.java

@@ -31,6 +31,7 @@
 import org.apache.cloudstack.api.response.ClusterResponse;
 import org.apache.cloudstack.api.response.HostResponse;
 import org.apache.cloudstack.api.response.ListResponse;
+import org.apache.cloudstack.api.response.ManagementServerResponse;
 import org.apache.cloudstack.api.response.PodResponse;
 import org.apache.cloudstack.api.response.UserVmResponse;
 import org.apache.cloudstack.api.response.ZoneResponse;
@@ -105,6 +106,9 @@ public class ListHostsCmd extends BaseListCmd {
     @Parameter(name = ApiConstants.HYPERVISOR, type = CommandType.STRING, description = "hypervisor type of host: XenServer,KVM,VMware,Hyperv,BareMetal,Simulator")
     private String hypervisor;
 
+    @Parameter(name = ApiConstants.MANAGEMENT_SERVER_ID, type = CommandType.UUID, entityType = ManagementServerResponse.class, description = "the id of the management server", since="4.21.0")
+    private Long managementServerId;
+
     /////////////////////////////////////////////////////
     /////////////////// Accessors ///////////////////////
     /////////////////////////////////////////////////////
@@ -189,6 +193,10 @@ public String getHostOutOfBandManagementPowerState() {
         return outOfBandManagementPowerState;
     }
 
+    public Long getManagementServerId() {
+        return managementServerId;
+    }
+
     /////////////////////////////////////////////////////
     /////////////// API Implementation///////////////////
     /////////////////////////////////////////////////////

api/src/main/java/org/apache/cloudstack/api/command/admin/host/PrepareForMaintenanceCmd.java → api/src/main/java/org/apache/cloudstack/api/command/admin/host/PrepareForHostMaintenanceCmd.java

@@ -33,7 +33,7 @@
 
 @APICommand(name = "prepareHostForMaintenance", description = "Prepares a host for maintenance.", responseObject = HostResponse.class,
         requestHasSensitiveInfo = false, responseHasSensitiveInfo = false)
-public class PrepareForMaintenanceCmd extends BaseAsyncCmd {
+public class PrepareForHostMaintenanceCmd extends BaseAsyncCmd {
 
 
     /////////////////////////////////////////////////////

api/src/main/java/org/apache/cloudstack/api/response/AsyncJobResponse.java

@@ -83,9 +83,13 @@ public class AsyncJobResponse extends BaseResponse {
     @Param(description = "the unique ID of the instance/entity object related to the job")
     private String jobInstanceId;
 
-    @SerializedName("managementserverid")
+    @SerializedName(ApiConstants.MANAGEMENT_SERVER_ID)
     @Param(description = "the msid of the management server on which the job is running", since = "4.19")
-    private Long msid;
+    private String managementServerId;
+
+    @SerializedName(ApiConstants.MANAGEMENT_SERVER_NAME)
+    @Param(description = "the management server name of the host", since = "4.21.0")
+    private String managementServerName;
 
     @SerializedName(ApiConstants.CREATED)
     @Param(description = "  the created date of the job")
@@ -156,7 +160,11 @@ public void setRemoved(final Date removed) {
         this.removed = removed;
     }
 
-    public void setMsid(Long msid) {
-        this.msid = msid;
+    public void setManagementServerId(String managementServerId) {
+        this.managementServerId = managementServerId;
+    }
+
+    public void setManagementServerName(String managementServerName) {
+        this.managementServerName = managementServerName;
     }
 }