mirror of https://github.com/apache/cloudstack.git
Merge remote-tracking branch 'origin/4.11'
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This commit is contained in:
commit
92cc4514ea
|
|
@ -38,12 +38,20 @@ import com.cloud.host.Status;
|
|||
import com.cloud.hypervisor.Hypervisor.HypervisorType;
|
||||
import com.cloud.resource.ResourceState.Event;
|
||||
import com.cloud.utils.fsm.NoTransitionException;
|
||||
import org.apache.cloudstack.framework.config.ConfigKey;
|
||||
import org.apache.cloudstack.framework.config.Configurable;
|
||||
|
||||
/**
|
||||
* ResourceManager manages how physical resources are organized within the
|
||||
* CloudStack. It also manages the life cycle of the physical resources.
|
||||
*/
|
||||
public interface ResourceManager extends ResourceService {
|
||||
public interface ResourceManager extends ResourceService, Configurable {
|
||||
|
||||
ConfigKey<Integer> HostMaintenanceRetries = new ConfigKey<>("Advanced", Integer.class,
|
||||
"host.maintenance.retries","20",
|
||||
"Number of retries when preparing a host into Maintenance Mode is faulty before failing",
|
||||
true, ConfigKey.Scope.Cluster);
|
||||
|
||||
/**
|
||||
* Register a listener for different types of resource life cycle events.
|
||||
* There can only be one type of listener per type of host.
|
||||
|
|
|
|||
|
|
@ -26,11 +26,17 @@ import java.util.Iterator;
|
|||
import java.util.List;
|
||||
import java.util.Map;
|
||||
import java.util.Random;
|
||||
import java.util.concurrent.ConcurrentHashMap;
|
||||
|
||||
import javax.inject.Inject;
|
||||
import javax.naming.ConfigurationException;
|
||||
|
||||
import com.cloud.vm.dao.UserVmDetailsDao;
|
||||
import org.apache.cloudstack.framework.config.ConfigKey;
|
||||
import org.apache.commons.lang.ObjectUtils;
|
||||
import org.apache.log4j.Logger;
|
||||
import org.springframework.stereotype.Component;
|
||||
|
||||
import org.apache.cloudstack.api.ApiConstants;
|
||||
import org.apache.cloudstack.api.command.admin.cluster.AddClusterCmd;
|
||||
import org.apache.cloudstack.api.command.admin.cluster.DeleteClusterCmd;
|
||||
|
|
@ -47,9 +53,6 @@ import org.apache.cloudstack.storage.datastore.db.PrimaryDataStoreDao;
|
|||
import org.apache.cloudstack.storage.datastore.db.StoragePoolVO;
|
||||
import org.apache.cloudstack.utils.identity.ManagementServerNode;
|
||||
import org.apache.commons.collections.CollectionUtils;
|
||||
import org.apache.commons.lang.ObjectUtils;
|
||||
import org.apache.log4j.Logger;
|
||||
import org.springframework.stereotype.Component;
|
||||
|
||||
import com.cloud.agent.AgentManager;
|
||||
import com.cloud.agent.api.Answer;
|
||||
|
|
@ -270,6 +273,8 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
|
||||
private SearchBuilder<HostGpuGroupsVO> _gpuAvailability;
|
||||
|
||||
private Map<Long,Integer> retryHostMaintenance = new ConcurrentHashMap<>();
|
||||
|
||||
private void insertListener(final Integer event, final ResourceListener listener) {
|
||||
List<ResourceListener> lst = _lifeCycleListeners.get(event);
|
||||
if (lst == null) {
|
||||
|
|
@ -1222,6 +1227,7 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
|
||||
ActionEventUtils.onStartedActionEvent(CallContext.current().getCallingUserId(), CallContext.current().getCallingAccountId(), EventTypes.EVENT_MAINTENANCE_PREPARE, "starting maintenance for host " + hostId, true, 0);
|
||||
_agentMgr.pullAgentToMaintenance(hostId);
|
||||
setHostMaintenanceRetries(host);
|
||||
|
||||
/* TODO: move below to listener */
|
||||
if (host.getType() == Host.Type.Routing) {
|
||||
|
|
@ -1249,6 +1255,16 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
return true;
|
||||
}
|
||||
|
||||
/**
|
||||
* Set retries for transiting the host into Maintenance
|
||||
*/
|
||||
protected void setHostMaintenanceRetries(HostVO host) {
|
||||
Integer retries = HostMaintenanceRetries.valueIn(host.getClusterId());
|
||||
retryHostMaintenance.put(host.getId(), retries);
|
||||
s_logger.debug(String.format("Setting the host %s (%s) retries for Maintenance mode: %s",
|
||||
host.getId(), host.getName(), retries));
|
||||
}
|
||||
|
||||
@Override
|
||||
public boolean maintain(final long hostId) throws AgentUnavailableException {
|
||||
final Boolean result = propagateResourceEvent(hostId, ResourceState.Event.AdminAskMaintenace);
|
||||
|
|
@ -1348,7 +1364,23 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
return CollectionUtils.isEmpty(failedMigrations) ?
|
||||
setHostIntoMaintenance(host) :
|
||||
setHostIntoErrorInMaintenance(host, failedMigrations);
|
||||
} else if (retryHostMaintenance.containsKey(host.getId())) {
|
||||
Integer retriesLeft = retryHostMaintenance.get(host.getId());
|
||||
if (retriesLeft != null) {
|
||||
if (retriesLeft <= 0) {
|
||||
retryHostMaintenance.remove(host.getId());
|
||||
s_logger.debug(String.format("No retries left while preparing KVM host %s (%s) for Maintenance, " +
|
||||
"please investigate this connection.",
|
||||
host.getId(), host.getName()));
|
||||
return setHostIntoErrorInMaintenance(host, failedMigrations);
|
||||
}
|
||||
retriesLeft--;
|
||||
retryHostMaintenance.put(host.getId(), retriesLeft);
|
||||
s_logger.debug(String.format("Retries left preparing KVM host %s (%s) for Maintenance: %s",
|
||||
host.getId(), host.getName(), retriesLeft));
|
||||
}
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
|
|
@ -2314,6 +2346,7 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
try {
|
||||
resourceStateTransitTo(host, ResourceState.Event.AdminCancelMaintenance, _nodeId);
|
||||
_agentMgr.pullAgentOutMaintenance(hostId);
|
||||
retryHostMaintenance.remove(hostId);
|
||||
|
||||
// for kvm, need to log into kvm host, restart cloudstack-agent
|
||||
if ((host.getHypervisorType() == HypervisorType.KVM && !vms_migrating) || host.getHypervisorType() == HypervisorType.LXC) {
|
||||
|
|
@ -2908,4 +2941,14 @@ public class ResourceManagerImpl extends ManagerBase implements ResourceManager,
|
|||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
@Override
|
||||
public String getConfigComponentName() {
|
||||
return ResourceManagerImpl.class.getSimpleName();
|
||||
}
|
||||
|
||||
@Override
|
||||
public ConfigKey<?>[] getConfigKeys() {
|
||||
return new ConfigKey<?>[] {HostMaintenanceRetries};
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -56,6 +56,7 @@ import com.cloud.org.Cluster;
|
|||
import com.cloud.resource.ResourceState.Event;
|
||||
import com.cloud.utils.component.ManagerBase;
|
||||
import com.cloud.utils.fsm.NoTransitionException;
|
||||
import org.apache.cloudstack.framework.config.ConfigKey;
|
||||
|
||||
public class MockResourceManagerImpl extends ManagerBase implements ResourceManager {
|
||||
|
||||
|
|
@ -625,4 +626,14 @@ public class MockResourceManagerImpl extends ManagerBase implements ResourceMana
|
|||
// TODO Auto-generated method stub
|
||||
return false;
|
||||
}
|
||||
|
||||
@Override
|
||||
public String getConfigComponentName() {
|
||||
return null;
|
||||
}
|
||||
|
||||
@Override
|
||||
public ConfigKey<?>[] getConfigKeys() {
|
||||
return new ConfigKey[0];
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -118,6 +118,7 @@ public class ResourceManagerImplTest {
|
|||
when(host.getId()).thenReturn(hostId);
|
||||
when(host.getResourceState()).thenReturn(ResourceState.Enabled);
|
||||
when(host.getHypervisorType()).thenReturn(Hypervisor.HypervisorType.VMware);
|
||||
when(host.getClusterId()).thenReturn(1L);
|
||||
when(hostDao.findById(hostId)).thenReturn(host);
|
||||
when(vm1.getId()).thenReturn(vm1Id);
|
||||
when(vm2.getId()).thenReturn(vm2Id);
|
||||
|
|
@ -188,4 +189,21 @@ public class ResourceManagerImplTest {
|
|||
verify(userVmDetailsDao).addDetail(eq(vm2Id), eq("kvm.vnc.port"), eq(String.valueOf(vm2VncPort)), anyBoolean());
|
||||
verify(agentManager).pullAgentToMaintenance(hostId);
|
||||
}
|
||||
|
||||
@Test
|
||||
public void testCheckAndMaintainErrorInMaintenanceRetries() throws NoTransitionException {
|
||||
resourceManager.setHostMaintenanceRetries(host);
|
||||
|
||||
List<VMInstanceVO> failedMigrations = Arrays.asList(vm1, vm2);
|
||||
when(vmInstanceDao.listByHostId(host.getId())).thenReturn(failedMigrations);
|
||||
when(vmInstanceDao.listNonMigratingVmsByHostEqualsLastHost(host.getId())).thenReturn(failedMigrations);
|
||||
|
||||
Integer retries = ResourceManager.HostMaintenanceRetries.valueIn(host.getClusterId());
|
||||
for (int i = 0; i <= retries; i++) {
|
||||
resourceManager.checkAndMaintain(host.getId());
|
||||
}
|
||||
|
||||
verify(resourceManager, times(retries + 1)).isHostInMaintenance(host, failedMigrations, new ArrayList<>(), failedMigrations);
|
||||
verify(resourceManager).setHostIntoErrorInMaintenance(host, failedMigrations);
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in New Issue