private List<Cluster> doPrivilegedLookup(String partitionName, String vmTypeName)
throws NotEnoughResourcesException {
if (Partition.DEFAULT_NAME.equals(partitionName)) {
Iterable<Cluster> authorizedClusters =
Iterables.filter(
Clusters.getInstance().listValues(),
RestrictedTypes.filterPrivilegedWithoutOwner());
Multimap<VmTypeAvailability, Cluster> sorted = TreeMultimap.create();
for (Cluster c : authorizedClusters) {
sorted.put(c.getNodeState().getAvailability(vmTypeName), c);
}
if (sorted.isEmpty()) {
throw new NotEnoughResourcesException(
"Not enough resources: no availability zone is available in which you have permissions to run instances.");
} else {
return Lists.newArrayList(sorted.values());
}
} else {
ServiceConfiguration ccConfig =
Topology.lookup(ClusterController.class, Partitions.lookupByName(partitionName));
Cluster cluster = Clusters.lookup(ccConfig);
if (cluster == null) {
throw new NotEnoughResourcesException("Can't find cluster " + partitionName);
}
if (!RestrictedTypes.filterPrivilegedWithoutOwner().apply(cluster)) {
throw new NotEnoughResourcesException("Not authorized to use cluster " + partitionName);
}
return Lists.newArrayList(cluster);
}
}
private int checkAvailability(String vmTypeName, List<Cluster> authorizedClusters)
throws NotEnoughResourcesException {
int available = 0;
for (Cluster authorizedCluster : authorizedClusters) {
VmTypeAvailability vmAvailability =
authorizedCluster.getNodeState().getAvailability(vmTypeName);
available += vmAvailability.getAvailable();
LOG.info(
"Availability: "
+ authorizedCluster.getName()
+ " -> "
+ vmAvailability.getAvailable());
}
return available;
}
public static void handleOrphan(Cluster cluster, ClusterAddressInfo address) {
Integer orphanCount = 1;
orphanCount = orphans.putIfAbsent(address, orphanCount);
orphanCount = (orphanCount == null) ? 1 : orphanCount;
orphans.put(address, orphanCount + 1);
EventRecord.caller(
ClusterState.class,
EventType.ADDRESS_STATE,
"Updated orphaned public ip address: "
+ LogUtil.dumpObject(address)
+ " count="
+ orphanCount)
.debug();
if (orphanCount > AddressingConfiguration.getInstance().getMaxKillOrphans()) {
EventRecord.caller(
ClusterState.class,
EventType.ADDRESS_STATE,
"Unassigning orphaned public ip address: "
+ LogUtil.dumpObject(address)
+ " count="
+ orphanCount)
.warn();
try {
final Address addr = Addresses.getInstance().lookup(address.getAddress());
if (addr.isPending()) {
try {
addr.clearPending();
} catch (Exception ex) {
}
}
try {
if (addr.isAssigned() && "0.0.0.0".equals(address.getInstanceIp())) {
addr.unassign().clearPending();
if (addr.isSystemOwned()) {
addr.release();
}
} else if (addr.isAssigned() && !"0.0.0.0".equals(address.getInstanceIp())) {
AsyncRequests.newRequest(new UnassignAddressCallback(address))
.sendSync(cluster.getConfiguration());
if (addr.isSystemOwned()) {
addr.release();
}
} else if (!addr.isAssigned() && addr.isAllocated() && addr.isSystemOwned()) {
addr.release();
}
} catch (ExecutionException ex) {
if (!addr.isAssigned() && addr.isAllocated() && addr.isSystemOwned()) {
addr.release();
}
}
} catch (InterruptedException ex) {
Exceptions.maybeInterrupted(ex);
} catch (NoSuchElementException ex) {
} finally {
orphans.remove(address);
}
}
}
private int checkZoneAvailability(
String vmTypeName, Partition partition, List<Cluster> authorizedClusters)
throws NotEnoughResourcesException {
int available = 0;
for (Cluster authorizedCluster : authorizedClusters) {
if (!authorizedCluster.getConfiguration().lookupPartition().equals(partition)) continue;
VmTypeAvailability vmAvailability =
authorizedCluster.getNodeState().getAvailability(vmTypeName);
available += vmAvailability.getAvailable();
LOG.info(
"Availability: "
+ authorizedCluster.getName()
+ " -> "
+ vmAvailability.getAvailable());
}
return available;
}
@Override
public void allocate(Allocation allocInfo) throws Exception {
Partition reqPartition = allocInfo.getPartition();
String zoneName = reqPartition.getName();
String vmTypeName = allocInfo.getVmType().getName();
/* Validate min and max amount */
final int minAmount = allocInfo.getMinCount();
final int maxAmount = allocInfo.getMaxCount();
if (minAmount > maxAmount)
throw new RuntimeException(
"Maximum instance count must not be smaller than minimum instance count");
/* Retrieve our context and list of clusters associated with this zone */
List<Cluster> authorizedClusters = this.doPrivilegedLookup(zoneName, vmTypeName);
int remaining = maxAmount;
int allocated = 0;
int available;
LOG.info(
"Found authorized clusters: "
+ Iterables.transform(authorizedClusters, HasName.GET_NAME));
/* Do we have any VM available throughout our clusters? */
if ((available = checkAvailability(vmTypeName, authorizedClusters)) < minAmount) {
throw new NotEnoughResourcesException(
"Not enough resources ("
+ available
+ " in "
+ zoneName
+ " < "
+ minAmount
+ "): vm instances.");
} else {
for (Cluster cluster : authorizedClusters) {
if (remaining <= 0) {
break;
} else {
ResourceState state = cluster.getNodeState();
Partition partition = cluster.getConfiguration().lookupPartition();
/* Has a partition been set if the AZ was not specified? */
if (allocInfo.getPartition().equals(Partition.DEFAULT)) {
/*
* Ok, do we have enough slots in this partition to support our request? We should have at least
* the minimum. The list is sorted in order of resource availability from the cluster with the most
* available to the cluster with the least amount available. This is why we don't check against the
* maxAmount value since its a best effort at this point. If we select the partition here and we
* can't fit maxAmount, based on the sorting order, the next partition will not fit maxAmount anyway.
*/
int zoneAvailable = checkZoneAvailability(vmTypeName, partition, authorizedClusters);
if (zoneAvailable < minAmount) continue;
/* Lets use this partition */
allocInfo.setPartition(partition);
} else if (!allocInfo.getPartition().equals(partition)) {
/* We should only pick clusters that are part of the selected AZ */
continue;
}
if (allocInfo.getBootSet().getMachine() instanceof BlockStorageImageInfo) {
try {
Topology.lookup(Storage.class, partition);
} catch (Exception ex) {
allocInfo.abort();
allocInfo.setPartition(reqPartition);
throw new NotEnoughResourcesException(
"Not enough resources: Cannot run EBS instances in partition w/o a storage controller: "
+ ex.getMessage(),
ex);
}
}
try {
int tryAmount =
(remaining > state.getAvailability(vmTypeName).getAvailable())
? state.getAvailability(vmTypeName).getAvailable()
: remaining;
List<ResourceToken> tokens =
this.requestResourceToken(allocInfo, tryAmount, maxAmount);
remaining -= tokens.size();
allocated += tokens.size();
} catch (Exception t) {
LOG.error(t);
Logs.extreme().error(t, t);
allocInfo.abort();
allocInfo.setPartition(reqPartition);
/* if we still have some allocation remaining AND no more resources are available */
if (((available = checkZoneAvailability(vmTypeName, partition, authorizedClusters))
< remaining)
&& (remaining > 0)) {
throw new NotEnoughResourcesException(
"Not enough resources ("
+ available
+ " in "
+ zoneName
+ " < "
+ minAmount
+ "): vm instances.",
t);
} else {
throw new NotEnoughResourcesException(t.getMessage(), t);
}
}
}
}
/* Were we able to meet our minimum requirements? */
if ((allocated < minAmount) && (remaining > 0)) {
allocInfo.abort();
allocInfo.setPartition(reqPartition);
if (reqPartition.equals(Partition.DEFAULT)) {
throw new NotEnoughResourcesException(
"Not enough resources available in all zone for " + minAmount + "): vm instances.");
} else {
available = checkZoneAvailability(vmTypeName, reqPartition, authorizedClusters);
throw new NotEnoughResourcesException(
"Not enough resources ("
+ available
+ " in "
+ zoneName
+ " < "
+ minAmount
+ "): vm instances.");
}
}
}
}
private List<ResourceToken> requestResourceToken(
final Allocation allocInfo, final int tryAmount, final int maxAmount) throws Exception {
ServiceConfiguration config =
Topology.lookup(ClusterController.class, allocInfo.getPartition());
Cluster cluster = Clusters.lookup(config);
/**
* TODO:GRZE: this is the call path which needs to trigger gating.
* It shouldn't be handled directly here, but instead be handled in {@link ResourceState#requestResourceAllocation().
*
*/
if (cluster.getGateLock().readLock().tryLock(60, TimeUnit.SECONDS)) {
try {
final ResourceState state = cluster.getNodeState();
/**
* NOTE: If the defined instance type has an ordering conflict w/ some other type then it
* isn't safe to service TWO requests which use differing types during the same resource
* refresh duty cycle. This determines whether or not an asynchronous allocation is safe
* to do for the request instance type or whether a synchronous resource availability
* refresh is needed.
*/
boolean unorderedType = VmTypes.isUnorderedType(allocInfo.getVmType());
boolean forceResourceRefresh = state.hasUnorderedTokens() || unorderedType;
/**
* GRZE: if the vm type is not "nicely" ordered then we force a refresh of the actual
* cluster state. Note: we already hold the cluster gating lock here so this update will
* be mutual exclusive wrt both resource allocations and cluster state updates.
*/
if (forceResourceRefresh) {
cluster.refreshResources();
}
final List<ResourceToken> tokens =
state.requestResourceAllocation(allocInfo, tryAmount, maxAmount);
final Iterator<ResourceToken> tokenIterator = tokens.iterator();
try {
final Supplier<ResourceToken> allocator =
new Supplier<ResourceToken>() {
@Override
public ResourceToken get() {
final ResourceToken ret = tokenIterator.next();
allocInfo.getAllocationTokens().add(ret);
return ret;
}
};
RestrictedTypes.allocateUnitlessResources(tokens.size(), allocator);
} finally {
// release any tokens that were not allocated
Iterators.all(
tokenIterator,
new Predicate<ResourceToken>() {
@Override
public boolean apply(final ResourceToken resourceToken) {
state.releaseToken(resourceToken);
return true;
}
});
}
return allocInfo.getAllocationTokens();
} finally {
cluster.getGateLock().readLock().unlock();
}
} else {
throw new ServiceStateException(
"Failed to allocate resources in the zone "
+ cluster.getPartition()
+ ", it is currently locked for maintenance.");
}
}
protected static Address lookupOrCreate(
final Cluster cluster, final ClusterAddressInfo addrInfo) {
Address addr = null;
VmInstance vm = null;
try {
addr = Addresses.getInstance().lookupDisabled(addrInfo.getAddress());
LOG.trace("Found address in the inactive set cache: " + addr);
} catch (final NoSuchElementException e1) {
try {
addr = Addresses.getInstance().lookup(addrInfo.getAddress());
LOG.trace("Found address in the active set cache: " + addr);
} catch (final NoSuchElementException e) {
}
}
if (addrInfo.hasMapping()) {
vm =
Helper.maybeFindVm(
addr != null ? addr.getInstanceId() : null,
addrInfo.getAddress(),
addrInfo.getInstanceIp());
if ((addr != null) && (vm != null)) {
Helper.ensureAllocated(addr, vm);
clearOrphan(addrInfo);
} else if (addr != null && !addr.isPending() && vm != null && VmStateSet.DONE.apply(vm)) {
handleOrphan(cluster, addrInfo);
} else if ((addr != null && addr.isAssigned() && !addr.isPending()) && (vm == null)) {
handleOrphan(cluster, addrInfo);
} else if ((addr == null) && (vm != null)) {
addr =
new Address(
Principals.systemFullName(),
addrInfo.getAddress(),
vm.getInstanceUuid(),
vm.getInstanceId(),
vm.getPrivateAddress());
clearOrphan(addrInfo);
} else if ((addr == null) && (vm == null)) {
addr = new Address(addrInfo.getAddress(), cluster.getPartition());
handleOrphan(cluster, addrInfo);
}
} else {
if ((addr != null) && addr.isAssigned() && !addr.isPending()) {
handleOrphan(cluster, addrInfo);
} else if ((addr != null)
&& !addr.isAssigned()
&& !addr.isPending()
&& addr.isSystemOwned()) {
try {
addr.release();
} catch (final Exception ex) {
LOG.error(ex);
}
} else if ((addr != null) && Address.Transition.system.equals(addr.getTransition())) {
handleOrphan(cluster, addrInfo);
} else if (addr == null) {
addr = new Address(addrInfo.getAddress(), cluster.getPartition());
Helper.clearVmState(addrInfo);
}
}
return addr;
}
public DetachVolumeResponseType detach(DetachVolumeType request) throws EucalyptusCloudException {
DetachVolumeResponseType reply = (DetachVolumeResponseType) request.getReply();
Context ctx = Contexts.lookup();
Volume vol;
try {
vol = Volumes.lookup(ctx.getUserFullName().asAccountFullName(), request.getVolumeId());
} catch (Exception ex1) {
throw new EucalyptusCloudException("Volume does not exist: " + request.getVolumeId());
}
if (!RestrictedTypes.filterPrivileged().apply(vol)) {
throw new EucalyptusCloudException(
"Not authorized to detach volume "
+ request.getVolumeId()
+ " by "
+ ctx.getUser().getName());
}
VmInstance vm = null;
AttachedVolume volume = null;
try {
VmVolumeAttachment vmVolAttach = VmInstances.lookupVolumeAttachment(request.getVolumeId());
volume = VmVolumeAttachment.asAttachedVolume(vmVolAttach.getVmInstance()).apply(vmVolAttach);
vm = vmVolAttach.getVmInstance();
} catch (NoSuchElementException ex) {
/** no such attachment * */
}
if (volume == null) {
throw new EucalyptusCloudException("Volume is not attached: " + request.getVolumeId());
}
if (!RestrictedTypes.filterPrivileged().apply(vm)) {
throw new EucalyptusCloudException(
"Not authorized to detach volume from instance "
+ request.getInstanceId()
+ " by "
+ ctx.getUser().getName());
}
if (!vm.getInstanceId().equals(request.getInstanceId())
&& request.getInstanceId() != null
&& !request.getInstanceId().equals("")) {
throw new EucalyptusCloudException(
"Volume is not attached to instance: " + request.getInstanceId());
}
if (request.getDevice() != null
&& !request.getDevice().equals("")
&& !volume.getDevice().equals(request.getDevice())) {
throw new EucalyptusCloudException(
"Volume is not attached to device: " + request.getDevice());
}
Cluster cluster = null;
ServiceConfiguration ccConfig = null;
try {
ccConfig = Topology.lookup(ClusterController.class, vm.lookupPartition());
cluster = Clusters.lookup(ccConfig);
} catch (NoSuchElementException e) {
LOG.debug(e, e);
throw new EucalyptusCloudException(
"Cluster does not exist in partition: " + vm.getPartition());
}
ServiceConfiguration scVm;
try {
scVm = Topology.lookup(Storage.class, vm.lookupPartition());
} catch (Exception ex) {
LOG.error(ex, ex);
throw new EucalyptusCloudException(
"Failed to lookup SC for partition: " + vm.getPartition(), ex);
}
request.setVolumeId(volume.getVolumeId());
request.setRemoteDevice(volume.getRemoteDevice());
request.setDevice(volume.getDevice().replaceAll("unknown,requested:", ""));
request.setInstanceId(vm.getInstanceId());
VolumeDetachCallback ncDetach = new VolumeDetachCallback(request);
try {
AsyncRequests.sendSync(scVm, new DetachStorageVolumeType(volume.getVolumeId()));
} catch (Exception e) {
LOG.debug(e);
Logs.extreme().debug(e, e);
// GRZE: attach is idempotent, failure here is ok. throw new EucalyptusCloudException(
// e.getMessage( ) );
}
AsyncRequests.newRequest(ncDetach).dispatch(cluster.getConfiguration());
EventRecord.here(VolumeManager.class, EventClass.VOLUME, EventType.VOLUME_DETACH)
.withDetails(vm.getOwner().toString(), volume.getVolumeId(), "instance", vm.getInstanceId())
.withDetails("cluster", ccConfig.getFullName().toString())
.info();
volume.setStatus("detaching");
reply.setDetachedVolume(volume);
return reply;
}
@Override
public void fireEvent(final ClockTick event) {
if (Bootstrap.isFinished() && Hosts.isCoordinator()) {
final List<ResourceAvailabilityEvent> resourceAvailabilityEvents = Lists.newArrayList();
final Map<ResourceAvailabilityEvent.ResourceType, AvailabilityAccumulator> availabilities =
Maps.newEnumMap(ResourceAvailabilityEvent.ResourceType.class);
final Iterable<VmType> vmTypes = Lists.newArrayList(VmTypes.list());
for (final Cluster cluster : Clusters.getInstance().listValues()) {
availabilities.put(Core, new AvailabilityAccumulator(VmType.SizeProperties.Cpu));
availabilities.put(Disk, new AvailabilityAccumulator(VmType.SizeProperties.Disk));
availabilities.put(Memory, new AvailabilityAccumulator(VmType.SizeProperties.Memory));
for (final VmType vmType : vmTypes) {
final ResourceState.VmTypeAvailability va =
cluster.getNodeState().getAvailability(vmType.getName());
resourceAvailabilityEvents.add(
new ResourceAvailabilityEvent(
Instance,
new ResourceAvailabilityEvent.Availability(
va.getMax(),
va.getAvailable(),
Lists.<ResourceAvailabilityEvent.Tag>newArrayList(
new ResourceAvailabilityEvent.Dimension(
"availabilityZone", cluster.getPartition()),
new ResourceAvailabilityEvent.Dimension("cluster", cluster.getName()),
new ResourceAvailabilityEvent.Type("vm-type", vmType.getName())))));
for (final AvailabilityAccumulator availability : availabilities.values()) {
availability.total =
Math.max(
availability.total, va.getMax() * availability.valueExtractor.apply(vmType));
availability.available =
Math.max(
availability.available,
va.getAvailable() * availability.valueExtractor.apply(vmType));
}
}
for (final AvailabilityAccumulator availability : availabilities.values()) {
availability.rollUp(
Lists.<ResourceAvailabilityEvent.Tag>newArrayList(
new ResourceAvailabilityEvent.Dimension(
"availabilityZone", cluster.getPartition()),
new ResourceAvailabilityEvent.Dimension("cluster", cluster.getName())));
}
}
for (final Map.Entry<ResourceAvailabilityEvent.ResourceType, AvailabilityAccumulator> entry :
availabilities.entrySet()) {
resourceAvailabilityEvents.add(
new ResourceAvailabilityEvent(entry.getKey(), entry.getValue().availabilities));
}
for (final ResourceAvailabilityEvent resourceAvailabilityEvent : resourceAvailabilityEvents)
try {
ListenerRegistry.getInstance().fireEvent(resourceAvailabilityEvent);
} catch (Exception ex) {
logger.error(ex, ex);
}
}
}