bug 11910: full sync will only stop unknown VMs on agent

api/src/com/cloud/agent/api/ClusterSyncAnswer.java

@@ -25,8 +25,9 @@ import com.cloud.vm.VirtualMachine.State;
 public class ClusterSyncAnswer extends Answer {
     private long _clusterId;
     private HashMap<String, Pair<String, State>> _newStates;
+    private HashMap<String, Pair<String, State>> _allStates;
     private int _type = -1; // 0 for full, 1 for delta
-    private boolean isExecuted=false;  // this is to avoidf double execution first time, due to framework ???
+    private boolean isExecuted=false;  // this is to avoid double execution first time, due to framework ???
     
     
     public static final int FULL_SYNC=0;
@@ -39,10 +40,19 @@ public class ClusterSyncAnswer extends Answer {
         _type = -1;
     }
     
-    public ClusterSyncAnswer(long clusterId, HashMap<String, Pair<String, State>> newStates, int type){
+
+    public ClusterSyncAnswer(long clusterId, HashMap<String, Pair<String, State>> newStates){
         _clusterId = clusterId;
         _newStates = newStates;
-        _type = type;
+        _type = DELTA_SYNC;
+        result = true;
+    }
+    
+    public ClusterSyncAnswer(long clusterId, HashMap<String, Pair<String, State>> newStates, HashMap<String, Pair<String, State>> allStates){
+        _clusterId = clusterId;
+        _newStates = newStates;
+        _allStates = allStates;
+        _type = FULL_SYNC;
         result = true;
     }
 
@@ -53,6 +63,10 @@ public class ClusterSyncAnswer extends Answer {
     public HashMap<String, Pair<String, State>> getNewStates() {
         return _newStates;
     }   
+
+    public HashMap<String, Pair<String, State>> getAllStates() {
+        return _allStates;
+    }   
     
     public boolean isFull(){
         return _type==0;

core/src/com/cloud/hypervisor/xen/resource/CitrixResourceBase.java

@@ -6578,23 +6578,17 @@ public abstract class CitrixResourceBase implements ServerResource, HypervisorRe
             s_logger.warn("Check for master failed, failing the Cluster sync command");
             return new ClusterSyncAnswer(cmd.getClusterId());
         } 
-        HashMap<String, Pair<String, State>> newStates;
+        HashMap<String, Pair<String, State>> newStates = deltaClusterSync(conn);
-        int sync_type=-1;
         if (cmd.isRightStep()){
             // do full sync
-            newStates=fullClusterSync(conn);
+        	 HashMap<String, Pair<String, State>> allStates=fullClusterSync(conn);
-            sync_type = ClusterSyncAnswer.FULL_SYNC;
+             cmd.incrStep();
+             return new ClusterSyncAnswer(cmd.getClusterId(), newStates, allStates);
         }
         else {
-            // do delta sync
+            cmd.incrStep();
-            newStates = deltaClusterSync(conn);
+            return new ClusterSyncAnswer(cmd.getClusterId(), newStates);
-            if (newStates == null) {
-                s_logger.warn("Unable to get current status from sync");
-            }
-            sync_type = ClusterSyncAnswer.DELTA_SYNC;
         }
-        cmd.incrStep();
-        return new ClusterSyncAnswer(cmd.getClusterId(), newStates, sync_type);
     }
 
 

server/src/com/cloud/configuration/Config.java

@@ -159,7 +159,7 @@ public enum Config {
 	PingInterval("Advanced", AgentManager.class, Integer.class, "ping.interval", "60", "Ping interval in seconds", null),
 	PingTimeout("Advanced", AgentManager.class, Float.class, "ping.timeout", "2.5", "Multiplier to ping.interval before announcing an agent has timed out", null),
 	ClusterDeltaSyncInterval("Advanced", AgentManager.class, Integer.class, "sync.interval", "60", "Cluster Delta sync interval in seconds", null),
-	ClusterFullSyncSkipSteps("Advanced", AgentManager.class, Integer.class, "skip.steps", "30", "Cluster full sync skip steps count", null),
+	ClusterFullSyncSkipSteps("Advanced", AgentManager.class, Integer.class, "skip.steps", "60", "Cluster full sync skip steps count", null),
 	Port("Advanced", AgentManager.class, Integer.class, "port", "8250", "Port to listen on for agent connection.", null),
 	RouterCpuMHz("Advanced", NetworkManager.class, Integer.class, "router.cpu.mhz", String.valueOf(VirtualNetworkApplianceManager.DEFAULT_ROUTER_CPU_MHZ), "Default CPU speed (MHz) for router VM.", null),
 	RestartRetryInterval("Advanced", HighAvailabilityManager.class, Integer.class, "restart.retry.interval", "600", "Time (in seconds) between retries to restart a vm", null),

server/src/com/cloud/vm/VirtualMachineManagerImpl.java

@@ -1703,7 +1703,6 @@ public class VirtualMachineManagerImpl implements VirtualMachineManager, Listene
     }
 
 
-
     public Commands fullSync(final long clusterId, Map<String, Pair<String, State>> newStates) {
         Commands commands = new Commands(OnError.Continue);
         Map<Long, AgentVmInfo> infos = convertToInfos(newStates);
@@ -1711,30 +1710,6 @@ public class VirtualMachineManagerImpl implements VirtualMachineManager, Listene
         final List<VMInstanceVO> vms = _vmDao.listByClusterId(clusterId);
         for (VMInstanceVO vm : vms) {
             AgentVmInfo info = infos.remove(vm.getId());
-            VMInstanceVO castedVm = null;
-            if (info == null) {
-                // the vm is not there on cluster, check the vm status in DB
-                if (vm.getState() == State.Starting && (DateUtil.currentGMTTime().getTime() - vm.getUpdateTime().getTime()) < 10*60*1000){
-                    continue; // ignoring this VM as it is still settling
-                }
-                info = new AgentVmInfo(vm.getInstanceName(), getVmGuru(vm), vm, State.Stopped);
-                castedVm = info.guru.findById(vm.getId());
-                hId = vm.getHostId() == null ? vm.getLastHostId() : vm.getHostId();
-            } else {
-                castedVm = info.vm;
-                String hostGuid = info.getHostUuid();
-                Host host = _resourceMgr.findHostByGuid(hostGuid);
-                if (host == null) {
-                    infos.put(vm.getId(), info);
-                    continue;
-                }
-                hId = host.getId();
-            }
-            HypervisorGuru hvGuru = _hvGuruMgr.getGuru(castedVm.getHypervisorType());
-            Command command = compareState(hId, castedVm, info, false, hvGuru.trackVmHostChange());
-            if (command != null) {
-                commands.addCommand(command);
-            }
         }
         for (final AgentVmInfo left : infos.values()) {
             s_logger.warn("Stopping a VM that we have no record of: " + left.name);
@@ -2072,7 +2047,8 @@ public class VirtualMachineManagerImpl implements VirtualMachineManager, Listene
                 ClusterSyncAnswer hs = (ClusterSyncAnswer) answer;
                 if (hs.execute()){
                     if (hs.isFull()) {
-                        fullSync(hs.getClusterId(), hs.getNewStates());
+                        deltaSync(hs.getNewStates());
+                        fullSync(hs.getClusterId(), hs.getAllStates());
                     } else if (hs.isDelta()){
                         deltaSync(hs.getNewStates());
                     }
@@ -2149,7 +2125,7 @@ public class VirtualMachineManagerImpl implements VirtualMachineManager, Listene
 
         Long clusterId = agent.getClusterId();
         long agentId = agent.getId();
-        if (agent.getHypervisorType() == HypervisorType.XenServer) { // only fro Xen
+        if (agent.getHypervisorType() == HypervisorType.XenServer) { // only for Xen
             ClusterSyncCommand syncCmd = new ClusterSyncCommand(Integer.parseInt(Config.ClusterDeltaSyncInterval.getDefaultValue()),
                     Integer.parseInt(Config.ClusterFullSyncSkipSteps.getDefaultValue()), clusterId);
             try {