CLOUDSTACK-10326: Fix for infinite loop on PrepareForMaintenance

Author: nvazquez

engine/schema/src/com/cloud/vm/dao/VMInstanceDao.java

@@ -151,8 +151,5 @@ public interface VMInstanceDao extends GenericDao<VMInstanceVO, Long>, StateDao<
 
     boolean isPowerStateUpToDate(long instanceId);
 
-    /**
-     * List running VMs which host and lastHost id are the same
-     */
-    List<VMInstanceVO> listRunningVmsByHostAndLastHostSameId(long hostId);
+    List<VMInstanceVO> listNonMigratingVmsByHostEqualsLastHost(long hostId);
 }

engine/schema/src/com/cloud/vm/dao/VMInstanceDaoImpl.java

@@ -93,6 +93,7 @@ public class VMInstanceDaoImpl extends GenericDaoBase<VMInstanceVO, Long> implem
     protected GenericSearchBuilder<VMInstanceVO, String> DistinctHostNameSearch;
     protected SearchBuilder<VMInstanceVO> HostAndStateSearch;
     protected SearchBuilder<VMInstanceVO> StartingWithNoHostSearch;
+    protected SearchBuilder<VMInstanceVO> NotMigratingSearch;
 
     @Inject
     ResourceTagDao _tagsDao;
@@ -280,6 +281,11 @@ protected void init() {
         DistinctHostNameSearch.join("nicSearch", nicSearch, DistinctHostNameSearch.entity().getId(), nicSearch.entity().getInstanceId(), JoinBuilder.JoinType.INNER);
         DistinctHostNameSearch.done();
 
+        NotMigratingSearch = createSearchBuilder();
+        NotMigratingSearch.and("host", NotMigratingSearch.entity().getHostId(), Op.EQ);
+        NotMigratingSearch.and("lastHost", NotMigratingSearch.entity().getLastHostId(), Op.EQ);
+        NotMigratingSearch.and("state", NotMigratingSearch.entity().getState(), Op.NEQ);
+        NotMigratingSearch.done();
     }
 
     @Override
@@ -305,11 +311,11 @@ public List<VMInstanceVO> listByHostId(long hostid) {
     }
 
     @Override
-    public List<VMInstanceVO> listRunningVmsByHostAndLastHostSameId(long hostId) {
-        SearchCriteria<VMInstanceVO> sc = AllFieldsSearch.create();
+    public List<VMInstanceVO> listNonMigratingVmsByHostEqualsLastHost(long hostId) {
+        SearchCriteria<VMInstanceVO> sc = NotMigratingSearch.create();
         sc.setParameters("host", hostId);
         sc.setParameters("lastHost", hostId);
-        sc.setParameters("state", State.Running);
+        sc.setParameters("state", State.Migrating);
         return listBy(sc);
     }
 

server/src/com/cloud/resource/ResourceManagerImpl.java

@@ -1296,11 +1296,17 @@ public boolean checkAndMaintain(final long hostId) {
             if (host.getType() != Host.Type.Storage) {
                 final List<VMInstanceVO> vos = _vmDao.listByHostId(hostId);
                 final List<VMInstanceVO> vosMigrating = _vmDao.listVmsMigratingFromHost(hostId);
-                final List<VMInstanceVO> failedMigratedVms = _vmDao.listRunningVmsByHostAndLastHostSameId(hostId);
-                if (vos.isEmpty() && vosMigrating.isEmpty() && failedMigratedVms.isEmpty()) {
-                    resourceStateTransitTo(host, ResourceState.Event.InternalEnterMaintenance, _nodeId);
-                    hostInMaintenance = true;
-                    ActionEventUtils.onCompletedActionEvent(CallContext.current().getCallingUserId(), CallContext.current().getCallingAccountId(), EventVO.LEVEL_INFO, EventTypes.EVENT_MAINTENANCE_PREPARE, "completed maintenance for host " + hostId, 0);
+                final List<VMInstanceVO> failedMigratedVms = _vmDao.listNonMigratingVmsByHostEqualsLastHost(hostId);
+                if (vos.isEmpty() && vosMigrating.isEmpty()) {
+                    if (!failedMigratedVms.isEmpty()) {
+                        s_logger.debug("Unable to migrate " + failedMigratedVms.size() + " VM(s) from host " + host.getUuid());
+                        resourceStateTransitTo(host, ResourceState.Event.UnableToMigrate, _nodeId);
+                    } else {
+                        s_logger.debug("Host " + host.getUuid() + " entering in Maintenance");
+                        resourceStateTransitTo(host, ResourceState.Event.InternalEnterMaintenance, _nodeId);
+                        hostInMaintenance = true;
+                        ActionEventUtils.onCompletedActionEvent(CallContext.current().getCallingUserId(), CallContext.current().getCallingAccountId(), EventVO.LEVEL_INFO, EventTypes.EVENT_MAINTENANCE_PREPARE, "completed maintenance for host " + hostId, 0);
+                    }
                 }
             }
         } catch (final NoTransitionException e) {