blob: 289778e2229b74a9b77a4aef12ac45bafa791973 [file] [log] [blame]
/*
* Copyright 2018-present Open Networking Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.onosproject.segmentrouting.mcast;
import com.google.common.cache.Cache;
import com.google.common.cache.CacheBuilder;
import com.google.common.cache.RemovalCause;
import com.google.common.cache.RemovalNotification;
import com.google.common.collect.HashMultimap;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Multimap;
import com.google.common.collect.Sets;
import org.onlab.packet.IpAddress;
import org.onlab.packet.VlanId;
import org.onlab.util.KryoNamespace;
import org.onosproject.cluster.NodeId;
import org.onosproject.core.ApplicationId;
import org.onosproject.core.CoreService;
import org.onosproject.mcast.api.McastEvent;
import org.onosproject.mcast.api.McastRoute;
import org.onosproject.mcast.api.McastRouteData;
import org.onosproject.mcast.api.McastRouteUpdate;
import org.onosproject.net.HostId;
import org.onosproject.net.ConnectPoint;
import org.onosproject.net.DeviceId;
import org.onosproject.net.Link;
import org.onosproject.net.Path;
import org.onosproject.net.PortNumber;
import org.onosproject.net.flowobjective.DefaultObjectiveContext;
import org.onosproject.net.flowobjective.ForwardingObjective;
import org.onosproject.net.flowobjective.NextObjective;
import org.onosproject.net.flowobjective.ObjectiveContext;
import org.onosproject.net.topology.LinkWeigher;
import org.onosproject.net.topology.Topology;
import org.onosproject.net.topology.TopologyService;
import org.onosproject.segmentrouting.SRLinkWeigher;
import org.onosproject.segmentrouting.SegmentRoutingManager;
import org.onosproject.segmentrouting.storekey.McastStoreKey;
import org.onosproject.store.serializers.KryoNamespaces;
import org.onosproject.store.service.ConsistentMap;
import org.onosproject.store.service.Serializer;
import org.onosproject.store.service.Versioned;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.time.Instant;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Optional;
import java.util.Set;
import java.util.concurrent.ScheduledExecutorService;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;
import java.util.stream.Collectors;
import static java.util.concurrent.Executors.newScheduledThreadPool;
import static org.onlab.util.Tools.groupedThreads;
import static org.onosproject.mcast.api.McastEvent.Type.ROUTE_ADDED;
import static org.onosproject.mcast.api.McastEvent.Type.ROUTE_REMOVED;
import static org.onosproject.mcast.api.McastEvent.Type.SOURCES_ADDED;
import static org.onosproject.mcast.api.McastEvent.Type.SOURCES_REMOVED;
import static org.onosproject.mcast.api.McastEvent.Type.SINKS_ADDED;
import static org.onosproject.mcast.api.McastEvent.Type.SINKS_REMOVED;
import static org.onosproject.segmentrouting.mcast.McastRole.EGRESS;
import static org.onosproject.segmentrouting.mcast.McastRole.INGRESS;
import static org.onosproject.segmentrouting.mcast.McastRole.TRANSIT;
/**
* Handles Multicast related events.
*/
public class McastHandler {
// Logger instance
private static final Logger log = LoggerFactory.getLogger(McastHandler.class);
// Reference to srManager and most used internal objects
private final SegmentRoutingManager srManager;
private final TopologyService topologyService;
private final McastUtils mcastUtils;
// Internal store of the Mcast nextobjectives
private final ConsistentMap<McastStoreKey, NextObjective> mcastNextObjStore;
// Internal store of the Mcast roles
private final ConsistentMap<McastStoreKey, McastRole> mcastRoleStore;
// Wait time for the cache
private static final int WAIT_TIME_MS = 1000;
/**
* The mcastEventCache is implemented to avoid race condition by giving more time to the
* underlying subsystems to process previous calls.
*/
private Cache<McastCacheKey, McastEvent> mcastEventCache = CacheBuilder.newBuilder()
.expireAfterWrite(WAIT_TIME_MS, TimeUnit.MILLISECONDS)
.removalListener((RemovalNotification<McastCacheKey, McastEvent> notification) -> {
// Get group ip, sink and related event
IpAddress mcastIp = notification.getKey().mcastIp();
HostId sink = notification.getKey().sinkHost();
McastEvent mcastEvent = notification.getValue();
RemovalCause cause = notification.getCause();
log.debug("mcastEventCache removal event. group={}, sink={}, mcastEvent={}, cause={}",
mcastIp, sink, mcastEvent, cause);
// If it expires or it has been replaced, we deque the event
switch (notification.getCause()) {
case REPLACED:
case EXPIRED:
dequeueMcastEvent(mcastEvent);
break;
default:
break;
}
}).build();
private void enqueueMcastEvent(McastEvent mcastEvent) {
// Retrieve, currentData, prevData and the group
final McastRouteUpdate mcastRouteUpdate = mcastEvent.subject();
final McastRouteUpdate mcastRoutePrevUpdate = mcastEvent.prevSubject();
final IpAddress group = mcastRoutePrevUpdate.route().group();
// Let's create the keys of the cache
ImmutableSet.Builder<HostId> sinksBuilder = ImmutableSet.builder();
if (mcastEvent.type() == SOURCES_ADDED ||
mcastEvent.type() == SOURCES_REMOVED) {
// FIXME To be addressed with multiple sources support
sinksBuilder.addAll(Collections.emptySet());
} else if (mcastEvent.type() == SINKS_ADDED) {
// We need to process the host id one by one
mcastRouteUpdate.sinks().forEach(((hostId, connectPoints) -> {
// Get the previous locations and verify if there are changes
Set<ConnectPoint> prevConnectPoints = mcastRoutePrevUpdate.sinks().get(hostId);
Set<ConnectPoint> changes = Sets.difference(connectPoints, prevConnectPoints != null ?
prevConnectPoints : Collections.emptySet());
if (!changes.isEmpty()) {
sinksBuilder.add(hostId);
}
}));
} else if (mcastEvent.type() == SINKS_REMOVED) {
// We need to process the host id one by one
mcastRoutePrevUpdate.sinks().forEach(((hostId, connectPoints) -> {
// Get the current locations and verify if there are changes
Set<ConnectPoint> currentConnectPoints = mcastRouteUpdate.sinks().get(hostId);
Set<ConnectPoint> changes = Sets.difference(connectPoints, currentConnectPoints != null ?
currentConnectPoints : Collections.emptySet());
if (!changes.isEmpty()) {
sinksBuilder.add(hostId);
}
}));
} else if (mcastEvent.type() == ROUTE_REMOVED) {
// Current subject is null, just take the previous host ids
sinksBuilder.addAll(mcastRoutePrevUpdate.sinks().keySet());
}
// Push the elements in the cache
sinksBuilder.build().forEach(sink -> {
McastCacheKey cacheKey = new McastCacheKey(group, sink);
mcastEventCache.put(cacheKey, mcastEvent);
});
}
private void dequeueMcastEvent(McastEvent mcastEvent) {
// Get new and old data
final McastRouteUpdate mcastUpdate = mcastEvent.subject();
final McastRouteUpdate mcastPrevUpdate = mcastEvent.prevSubject();
// Get source, mcast group
// FIXME To be addressed with multiple sources support
final ConnectPoint source = mcastPrevUpdate.sources()
.values()
.stream()
.flatMap(Collection::stream)
.findFirst()
.orElse(null);
IpAddress mcastIp = mcastPrevUpdate.route().group();
// Get all the previous sinks
Set<ConnectPoint> prevSinks = mcastPrevUpdate.sinks()
.values()
.stream()
.flatMap(Collection::stream)
.collect(Collectors.toSet());
// According to the event type let's call the proper method
switch (mcastEvent.type()) {
case SOURCES_ADDED:
// FIXME To be addressed with multiple sources support
// Get all the sinks
//Set<ConnectPoint> sinks = mcastRouteInfo.sinks();
// Compute the Mcast tree
//Map<ConnectPoint, List<Path>> mcasTree = computeSinkMcastTree(source.deviceId(), sinks);
// Process the given sinks using the pre-computed paths
//mcasTree.forEach((sink, paths) -> processSinkAddedInternal(source, sink, mcastIp, paths));
break;
case SOURCES_REMOVED:
// FIXME To be addressed with multiple sources support
// Get old source
//ConnectPoint oldSource = mcastEvent.prevSubject().source().orElse(null);
// Just the first cached element will be processed
//processSourceUpdatedInternal(mcastIp, source, oldSource);
break;
case ROUTE_REMOVED:
// Process the route removed, just the first cached element will be processed
processRouteRemovedInternal(source, mcastIp);
break;
case SINKS_ADDED:
// FIXME To be addressed with multiple sources support
processSinksAddedInternal(source, mcastIp,
mcastUpdate.sinks(), prevSinks);
break;
case SINKS_REMOVED:
// FIXME To be addressed with multiple sources support
processSinksRemovedInternal(source, mcastIp,
mcastUpdate.sinks(), mcastPrevUpdate.sinks());
break;
default:
break;
}
}
// Mcast lock to serialize local operations
private final Lock mcastLock = new ReentrantLock();
/**
* Acquires the lock used when making mcast changes.
*/
private void mcastLock() {
mcastLock.lock();
}
/**
* Releases the lock used when making mcast changes.
*/
private void mcastUnlock() {
mcastLock.unlock();
}
// Stability threshold for Mcast. Seconds
private static final long MCAST_STABLITY_THRESHOLD = 5;
// Last change done
private Instant lastMcastChange = Instant.now();
/**
* Determines if mcast in the network has been stable in the last
* MCAST_STABLITY_THRESHOLD seconds, by comparing the current time
* to the last mcast change timestamp.
*
* @return true if stable
*/
private boolean isMcastStable() {
long last = (long) (lastMcastChange.toEpochMilli() / 1000.0);
long now = (long) (Instant.now().toEpochMilli() / 1000.0);
log.trace("Mcast stable since {}s", now - last);
return (now - last) > MCAST_STABLITY_THRESHOLD;
}
// Verify interval for Mcast
private static final long MCAST_VERIFY_INTERVAL = 30;
// Executor for mcast bucket corrector
private ScheduledExecutorService executorService
= newScheduledThreadPool(1, groupedThreads("mcastWorker", "mcastWorker-%d", log));
/**
* Constructs the McastEventHandler.
*
* @param srManager Segment Routing manager
*/
public McastHandler(SegmentRoutingManager srManager) {
ApplicationId coreAppId = srManager.coreService.getAppId(CoreService.CORE_APP_NAME);
this.srManager = srManager;
this.topologyService = srManager.topologyService;
KryoNamespace.Builder mcastKryo = new KryoNamespace.Builder()
.register(KryoNamespaces.API)
.register(McastStoreKey.class)
.register(McastRole.class);
mcastNextObjStore = srManager.storageService
.<McastStoreKey, NextObjective>consistentMapBuilder()
.withName("onos-mcast-nextobj-store")
.withSerializer(Serializer.using(mcastKryo.build("McastHandler-NextObj")))
.build();
mcastRoleStore = srManager.storageService
.<McastStoreKey, McastRole>consistentMapBuilder()
.withName("onos-mcast-role-store")
.withSerializer(Serializer.using(mcastKryo.build("McastHandler-Role")))
.build();
// Let's create McastUtils object
mcastUtils = new McastUtils(srManager, coreAppId, log);
// Init the executor service and the buckets corrector
executorService.scheduleWithFixedDelay(new McastBucketCorrector(), 10,
MCAST_VERIFY_INTERVAL, TimeUnit.SECONDS);
// Schedule the clean up, this will allow the processing of the expired events
executorService.scheduleAtFixedRate(mcastEventCache::cleanUp, 0,
WAIT_TIME_MS, TimeUnit.MILLISECONDS);
}
/**
* Read initial multicast from mcast store.
*/
public void init() {
lastMcastChange = Instant.now();
mcastLock();
try {
srManager.multicastRouteService.getRoutes().forEach(mcastRoute -> {
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastRoute.group())) {
log.debug("Skip {} due to lack of leadership", mcastRoute.group());
return;
}
// FIXME To be addressed with multiple sources support
ConnectPoint source = srManager.multicastRouteService.sources(mcastRoute)
.stream()
.findFirst()
.orElse(null);
// Get all the sinks and process them
McastRouteData mcastRouteData = srManager.multicastRouteService.routeData(mcastRoute);
Set<ConnectPoint> sinks = processSinksToBeAdded(source, mcastRoute.group(), mcastRouteData.sinks());
// Filter out all the working sinks, we do not want to move them
sinks = sinks.stream()
.filter(sink -> {
McastStoreKey mcastKey = new McastStoreKey(mcastRoute.group(), sink.deviceId());
Versioned<NextObjective> verMcastNext = mcastNextObjStore.get(mcastKey);
return verMcastNext == null ||
!mcastUtils.getPorts(verMcastNext.value().next()).contains(sink.port());
})
.collect(Collectors.toSet());
// Compute the Mcast tree
Map<ConnectPoint, List<Path>> mcasTree = computeSinkMcastTree(source.deviceId(), sinks);
// Process the given sinks using the pre-computed paths
mcasTree.forEach((sink, paths) -> processSinkAddedInternal(source, sink,
mcastRoute.group(), paths));
});
} finally {
mcastUnlock();
}
}
/**
* Clean up when deactivating the application.
*/
public void terminate() {
mcastEventCache.invalidateAll();
executorService.shutdown();
mcastNextObjStore.destroy();
mcastRoleStore.destroy();
mcastUtils.terminate();
log.info("Terminated");
}
/**
* Processes the SOURCE_ADDED, SOURCE_UPDATED, SINK_ADDED,
* SINK_REMOVED and ROUTE_REMOVED events.
*
* @param event McastEvent with SOURCE_ADDED type
*/
public void processMcastEvent(McastEvent event) {
log.info("process {}", event);
// If it is a route added, we do not enqueue
if (event.type() == ROUTE_ADDED) {
// We need just to elect a leader
processRouteAddedInternal(event.subject().route().group());
} else {
// Just enqueue for now
enqueueMcastEvent(event);
}
}
/**
* Process the ROUTE_ADDED event.
*
* @param mcastIp the group address
*/
private void processRouteAddedInternal(IpAddress mcastIp) {
lastMcastChange = Instant.now();
mcastLock();
try {
log.debug("Processing route added for group {}", mcastIp);
// Just elect a new leader
mcastUtils.isLeader(mcastIp);
} finally {
mcastUnlock();
}
}
/**
* Removes the entire mcast tree related to this group.
*
* @param mcastIp multicast group IP address
*/
private void processRouteRemovedInternal(ConnectPoint source, IpAddress mcastIp) {
lastMcastChange = Instant.now();
mcastLock();
try {
log.debug("Processing route removed for group {}", mcastIp);
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastIp)) {
log.debug("Skip {} due to lack of leadership", mcastIp);
mcastUtils.withdrawLeader(mcastIp);
return;
}
// Find out the ingress, transit and egress device of the affected group
DeviceId ingressDevice = getDevice(mcastIp, INGRESS)
.stream().findAny().orElse(null);
Set<DeviceId> transitDevices = getDevice(mcastIp, TRANSIT);
Set<DeviceId> egressDevices = getDevice(mcastIp, EGRESS);
// If there are no egress devices, sinks could be only on the ingress
if (!egressDevices.isEmpty()) {
egressDevices.forEach(
deviceId -> removeGroupFromDevice(deviceId, mcastIp, mcastUtils.assignedVlan(null))
);
}
// Transit could be empty if sinks are on the ingress
if (!transitDevices.isEmpty()) {
transitDevices.forEach(
deviceId -> removeGroupFromDevice(deviceId, mcastIp, mcastUtils.assignedVlan(null))
);
}
// Ingress device should be not null
if (ingressDevice != null) {
removeGroupFromDevice(ingressDevice, mcastIp, mcastUtils.assignedVlan(source));
}
} finally {
mcastUnlock();
}
}
/**
* Process sinks to be removed.
*
* @param source the source connect point
* @param mcastIp the ip address of the group
* @param newSinks the new sinks to be processed
* @param prevSinks the previous sinks
*/
private void processSinksRemovedInternal(ConnectPoint source, IpAddress mcastIp,
Map<HostId, Set<ConnectPoint>> newSinks,
Map<HostId, Set<ConnectPoint>> prevSinks) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastIp)) {
log.debug("Skip {} due to lack of leadership", mcastIp);
return;
}
// Remove the previous ones
Set<ConnectPoint> sinksToBeRemoved = processSinksToBeRemoved(mcastIp, prevSinks,
newSinks);
sinksToBeRemoved.forEach(sink -> processSinkRemovedInternal(source, sink, mcastIp));
// Recover the dual-homed sinks
Set<ConnectPoint> sinksToBeRecovered = processSinksToBeRecovered(mcastIp, newSinks,
prevSinks);
sinksToBeRecovered.forEach(sink -> processSinkAddedInternal(source, sink, mcastIp, null));
} finally {
mcastUnlock();
}
}
/**
* Removes a path from source to sink for given multicast group.
*
* @param source connect point of the multicast source
* @param sink connection point of the multicast sink
* @param mcastIp multicast group IP address
*/
private void processSinkRemovedInternal(ConnectPoint source, ConnectPoint sink,
IpAddress mcastIp) {
lastMcastChange = Instant.now();
mcastLock();
try {
boolean isLast;
// When source and sink are on the same device
if (source.deviceId().equals(sink.deviceId())) {
// Source and sink are on even the same port. There must be something wrong.
if (source.port().equals(sink.port())) {
log.warn("Skip {} since sink {} is on the same port of source {}. Abort",
mcastIp, sink, source);
return;
}
isLast = removePortFromDevice(sink.deviceId(), sink.port(), mcastIp, mcastUtils.assignedVlan(source));
if (isLast) {
mcastRoleStore.remove(new McastStoreKey(mcastIp, sink.deviceId()));
}
return;
}
// Process the egress device
isLast = removePortFromDevice(sink.deviceId(), sink.port(), mcastIp, mcastUtils.assignedVlan(null));
if (isLast) {
mcastRoleStore.remove(new McastStoreKey(mcastIp, sink.deviceId()));
}
// If this is the last sink on the device, also update upstream
Optional<Path> mcastPath = getPath(source.deviceId(), sink.deviceId(),
mcastIp, null);
if (mcastPath.isPresent()) {
List<Link> links = Lists.newArrayList(mcastPath.get().links());
Collections.reverse(links);
for (Link link : links) {
if (isLast) {
isLast = removePortFromDevice(
link.src().deviceId(),
link.src().port(),
mcastIp,
mcastUtils.assignedVlan(link.src().deviceId().equals(source.deviceId()) ?
source : null)
);
if (isLast) {
mcastRoleStore.remove(new McastStoreKey(mcastIp, link.src().deviceId()));
}
}
}
}
} finally {
mcastUnlock();
}
}
/**
* Process sinks to be added.
*
* @param source the source connect point
* @param mcastIp the group IP
* @param newSinks the new sinks to be processed
* @param allPrevSinks all previous sinks
*/
private void processSinksAddedInternal(ConnectPoint source, IpAddress mcastIp,
Map<HostId, Set<ConnectPoint>> newSinks,
Set<ConnectPoint> allPrevSinks) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastIp)) {
log.debug("Skip {} due to lack of leadership", mcastIp);
return;
}
// Get the only sinks to be processed (new ones)
Set<ConnectPoint> sinksToBeAdded = processSinksToBeAdded(source, mcastIp, newSinks);
// Install new sinks
sinksToBeAdded = Sets.difference(sinksToBeAdded, allPrevSinks);
sinksToBeAdded.forEach(sink -> processSinkAddedInternal(source, sink, mcastIp, null));
} finally {
mcastUnlock();
}
}
/**
* Establishes a path from source to sink for given multicast group.
*
* @param source connect point of the multicast source
* @param sink connection point of the multicast sink
* @param mcastIp multicast group IP address
*/
private void processSinkAddedInternal(ConnectPoint source, ConnectPoint sink,
IpAddress mcastIp, List<Path> allPaths) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Process the ingress device
mcastUtils.addFilterToDevice(source.deviceId(), source.port(),
mcastUtils.assignedVlan(source), mcastIp, INGRESS);
// When source and sink are on the same device
if (source.deviceId().equals(sink.deviceId())) {
// Source and sink are on even the same port. There must be something wrong.
if (source.port().equals(sink.port())) {
log.warn("Skip {} since sink {} is on the same port of source {}. Abort",
mcastIp, sink, source);
return;
}
addPortToDevice(sink.deviceId(), sink.port(), mcastIp, mcastUtils.assignedVlan(source));
mcastRoleStore.put(new McastStoreKey(mcastIp, sink.deviceId()), INGRESS);
return;
}
// Find a path. If present, create/update groups and flows for each hop
Optional<Path> mcastPath = getPath(source.deviceId(), sink.deviceId(),
mcastIp, allPaths);
if (mcastPath.isPresent()) {
List<Link> links = mcastPath.get().links();
// Setup mcast role for ingress
mcastRoleStore.put(new McastStoreKey(mcastIp, source.deviceId()),
INGRESS);
// Setup properly the transit
links.forEach(link -> {
addPortToDevice(link.src().deviceId(), link.src().port(), mcastIp,
mcastUtils.assignedVlan(link.src().deviceId()
.equals(source.deviceId()) ? source : null));
mcastUtils.addFilterToDevice(link.dst().deviceId(), link.dst().port(),
mcastUtils.assignedVlan(null), mcastIp, null);
});
// Setup mcast role for the transit
links.stream()
.filter(link -> !link.dst().deviceId().equals(sink.deviceId()))
.forEach(link -> mcastRoleStore.put(new McastStoreKey(mcastIp, link.dst().deviceId()),
TRANSIT));
// Process the egress device
addPortToDevice(sink.deviceId(), sink.port(), mcastIp, mcastUtils.assignedVlan(null));
// Setup mcast role for egress
mcastRoleStore.put(new McastStoreKey(mcastIp, sink.deviceId()),
EGRESS);
} else {
log.warn("Unable to find a path from {} to {}. Abort sinkAdded",
source.deviceId(), sink.deviceId());
}
} finally {
mcastUnlock();
}
}
/**
* Processes the LINK_DOWN event.
*
* @param affectedLink Link that is going down
*/
public void processLinkDown(Link affectedLink) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Get groups affected by the link down event
getAffectedGroups(affectedLink).forEach(mcastIp -> {
// TODO Optimize when the group editing is in place
log.debug("Processing link down {} for group {}",
affectedLink, mcastIp);
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastIp)) {
log.debug("Skip {} due to lack of leadership", mcastIp);
return;
}
// Find out the ingress, transit and egress device of affected group
DeviceId ingressDevice = getDevice(mcastIp, INGRESS)
.stream().findAny().orElse(null);
Set<DeviceId> transitDevices = getDevice(mcastIp, TRANSIT);
Set<DeviceId> egressDevices = getDevice(mcastIp, EGRESS);
ConnectPoint source = mcastUtils.getSource(mcastIp);
// Do not proceed if ingress device or source of this group are missing
// If sinks are in other leafs, we have ingress, transit, egress, and source
// If sinks are in the same leaf, we have just ingress and source
if (ingressDevice == null || source == null) {
log.warn("Missing ingress {} or source {} for group {}",
ingressDevice, source, mcastIp);
return;
}
// Remove entire transit
transitDevices.forEach(transitDevice ->
removeGroupFromDevice(transitDevice, mcastIp,
mcastUtils.assignedVlan(null)));
// Remove transit-facing ports on the ingress device
removeIngressTransitPorts(mcastIp, ingressDevice, source);
// TODO create a shared procedure with DEVICE_DOWN
// Compute mcast tree for the the egress devices
Map<DeviceId, List<Path>> mcastTree = computeMcastTree(ingressDevice, egressDevices);
// We have to verify, if there are egresses without paths
Set<DeviceId> notRecovered = Sets.newHashSet();
mcastTree.forEach((egressDevice, paths) -> {
// Let's check if there is at least a path
Optional<Path> mcastPath = getPath(ingressDevice, egressDevice,
mcastIp, paths);
// No paths, we have to try with alternative location
if (!mcastPath.isPresent()) {
notRecovered.add(egressDevice);
// We were not able to find an alternative path for this egress
log.warn("Fail to recover egress device {} from link failure {}",
egressDevice, affectedLink);
removeGroupFromDevice(egressDevice, mcastIp,
mcastUtils.assignedVlan(null));
}
});
// Fast path, we can recover all the locations
if (notRecovered.isEmpty()) {
// Construct a new path for each egress device
mcastTree.forEach((egressDevice, paths) -> {
// We try to enforce the sinks path on the mcast tree
Optional<Path> mcastPath = getPath(ingressDevice, egressDevice,
mcastIp, paths);
// If a path is present, let's install it
if (mcastPath.isPresent()) {
installPath(mcastIp, source, mcastPath.get());
}
});
} else {
// Let's try to recover using alternate
recoverSinks(egressDevices, notRecovered, mcastIp,
ingressDevice, source, true);
}
});
} finally {
mcastUnlock();
}
}
/**
* Process the DEVICE_DOWN event.
*
* @param deviceDown device going down
*/
public void processDeviceDown(DeviceId deviceDown) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Get the mcast groups affected by the device going down
getAffectedGroups(deviceDown).forEach(mcastIp -> {
// TODO Optimize when the group editing is in place
log.debug("Processing device down {} for group {}",
deviceDown, mcastIp);
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastIp)) {
log.debug("Skip {} due to lack of leadership", mcastIp);
return;
}
// Find out the ingress, transit and egress device of affected group
DeviceId ingressDevice = getDevice(mcastIp, INGRESS)
.stream().findAny().orElse(null);
Set<DeviceId> transitDevices = getDevice(mcastIp, TRANSIT);
Set<DeviceId> egressDevices = getDevice(mcastIp, EGRESS);
ConnectPoint source = mcastUtils.getSource(mcastIp);
// Do not proceed if ingress device or source of this group are missing
// If sinks are in other leafs, we have ingress, transit, egress, and source
// If sinks are in the same leaf, we have just ingress and source
if (ingressDevice == null || source == null) {
log.warn("Missing ingress {} or source {} for group {}",
ingressDevice, source, mcastIp);
return;
}
// If it exists, we have to remove it in any case
if (!transitDevices.isEmpty()) {
// Remove entire transit
transitDevices.forEach(transitDevice ->
removeGroupFromDevice(transitDevice, mcastIp,
mcastUtils.assignedVlan(null)));
}
// If the ingress is down
if (ingressDevice.equals(deviceDown)) {
// Remove entire ingress
removeGroupFromDevice(ingressDevice, mcastIp, mcastUtils.assignedVlan(source));
// If other sinks different from the ingress exist
if (!egressDevices.isEmpty()) {
// Remove all the remaining egress
egressDevices.forEach(
egressDevice -> removeGroupFromDevice(egressDevice, mcastIp,
mcastUtils.assignedVlan(null))
);
}
} else {
// Egress or transit could be down at this point
// Get the ingress-transit ports if they exist
removeIngressTransitPorts(mcastIp, ingressDevice, source);
// One of the egress device is down
if (egressDevices.contains(deviceDown)) {
// Remove entire device down
removeGroupFromDevice(deviceDown, mcastIp, mcastUtils.assignedVlan(null));
// Remove the device down from egress
egressDevices.remove(deviceDown);
// If there are no more egress and ingress does not have sinks
if (egressDevices.isEmpty() && !hasSinks(ingressDevice, mcastIp)) {
// We have done
return;
}
}
// Compute mcast tree for the the egress devices
Map<DeviceId, List<Path>> mcastTree = computeMcastTree(ingressDevice, egressDevices);
// We have to verify, if there are egresses without paths
Set<DeviceId> notRecovered = Sets.newHashSet();
mcastTree.forEach((egressDevice, paths) -> {
// Let's check if there is at least a path
Optional<Path> mcastPath = getPath(ingressDevice, egressDevice,
mcastIp, paths);
// No paths, we have to try with alternative location
if (!mcastPath.isPresent()) {
notRecovered.add(egressDevice);
// We were not able to find an alternative path for this egress
log.warn("Fail to recover egress device {} from device down {}",
egressDevice, deviceDown);
removeGroupFromDevice(egressDevice, mcastIp, mcastUtils.assignedVlan(null));
}
});
// Fast path, we can recover all the locations
if (notRecovered.isEmpty()) {
// Construct a new path for each egress device
mcastTree.forEach((egressDevice, paths) -> {
// We try to enforce the sinks path on the mcast tree
Optional<Path> mcastPath = getPath(ingressDevice, egressDevice,
mcastIp, paths);
// If a path is present, let's install it
if (mcastPath.isPresent()) {
installPath(mcastIp, source, mcastPath.get());
}
});
} else {
// Let's try to recover using alternate
recoverSinks(egressDevices, notRecovered, mcastIp,
ingressDevice, source, false);
}
}
});
} finally {
mcastUnlock();
}
}
/**
* Try to recover sinks using alternate locations.
*
* @param egressDevices the original egress devices
* @param notRecovered the devices not recovered
* @param mcastIp the group address
* @param ingressDevice the ingress device
* @param source the source connect point
* @param isLinkFailure true if it is a link failure, otherwise false
*/
private void recoverSinks(Set<DeviceId> egressDevices, Set<DeviceId> notRecovered,
IpAddress mcastIp, DeviceId ingressDevice, ConnectPoint source,
boolean isLinkFailure) {
// Recovered devices
Set<DeviceId> recovered = Sets.difference(egressDevices, notRecovered);
// Total affected sinks
Set<ConnectPoint> totalAffectedSinks = Sets.newHashSet();
// Total sinks
Set<ConnectPoint> totalSinks = Sets.newHashSet();
// Let's compute all the affected sinks and all the sinks
notRecovered.forEach(deviceId -> {
totalAffectedSinks.addAll(
mcastUtils.getAffectedSinks(deviceId, mcastIp)
.values()
.stream()
.flatMap(Collection::stream)
.filter(connectPoint -> connectPoint.deviceId().equals(deviceId))
.collect(Collectors.toSet())
);
totalSinks.addAll(
mcastUtils.getAffectedSinks(deviceId, mcastIp)
.values()
.stream()
.flatMap(Collection::stream)
.collect(Collectors.toSet())
);
});
// Sinks to be added
Set<ConnectPoint> sinksToBeAdded = Sets.difference(totalSinks, totalAffectedSinks);
// New egress devices, filtering out the source
Set<DeviceId> newEgressDevice = sinksToBeAdded.stream()
.map(ConnectPoint::deviceId)
.collect(Collectors.toSet());
// Let's add the devices recovered from the previous round
newEgressDevice.addAll(recovered);
// Let's do a copy of the new egresses and filter out the source
Set<DeviceId> copyNewEgressDevice = ImmutableSet.copyOf(newEgressDevice);
newEgressDevice = newEgressDevice.stream()
.filter(deviceId -> !deviceId.equals(ingressDevice))
.collect(Collectors.toSet());
// Re-compute mcast tree for the the egress devices
Map<DeviceId, List<Path>> mcastTree = computeMcastTree(ingressDevice, newEgressDevice);
// if the source was originally in the new locations, add new sinks
if (copyNewEgressDevice.contains(ingressDevice)) {
sinksToBeAdded.stream()
.filter(connectPoint -> connectPoint.deviceId().equals(ingressDevice))
.forEach(sink -> processSinkAddedInternal(source, sink, mcastIp, ImmutableList.of()));
}
// Construct a new path for each egress device
mcastTree.forEach((egressDevice, paths) -> {
// We try to enforce the sinks path on the mcast tree
Optional<Path> mcastPath = getPath(ingressDevice, egressDevice,
mcastIp, paths);
// If a path is present, let's install it
if (mcastPath.isPresent()) {
// Using recovery procedure
if (recovered.contains(egressDevice)) {
installPath(mcastIp, source, mcastPath.get());
} else {
// otherwise we need to threat as new sink
sinksToBeAdded.stream()
.filter(connectPoint -> connectPoint.deviceId().equals(egressDevice))
.forEach(sink -> processSinkAddedInternal(source, sink, mcastIp, paths));
}
} else {
// We were not able to find an alternative path for this egress
log.warn("Fail to recover egress device {} from {} failure",
egressDevice, isLinkFailure ? "Link" : "Device");
removeGroupFromDevice(egressDevice, mcastIp, mcastUtils.assignedVlan(null));
}
});
}
/**
* Process all the sinks related to a mcast group and return
* the ones to be removed.
*
* @param mcastIp the group address
* @param prevsinks the previous sinks to be evaluated
* @param newSinks the new sinks to be evaluted
* @return the set of the sinks to be removed
*/
private Set<ConnectPoint> processSinksToBeRemoved(IpAddress mcastIp,
Map<HostId, Set<ConnectPoint>> prevsinks,
Map<HostId, Set<ConnectPoint>> newSinks) {
// Iterate over the sinks in order to build the set
// of the connect points to be removed from this group
final Set<ConnectPoint> sinksToBeProcessed = Sets.newHashSet();
prevsinks.forEach(((hostId, connectPoints) -> {
// We have to check with the existing flows
ConnectPoint sinkToBeProcessed = connectPoints.stream()
.filter(connectPoint -> isSink(mcastIp, connectPoint))
.findFirst().orElse(null);
if (sinkToBeProcessed != null) {
// If the host has been removed or location has been removed
if (!newSinks.containsKey(hostId) ||
!newSinks.get(hostId).contains(sinkToBeProcessed)) {
sinksToBeProcessed.add(sinkToBeProcessed);
}
}
}));
// We have done, return the set
return sinksToBeProcessed;
}
/**
* Process new locations and return the set of sinks to be added
* in the context of the recovery.
*
* @param newSinks the remaining sinks
* @param prevSinks the previous sinks
* @return the set of the sinks to be processed
*/
private Set<ConnectPoint> processSinksToBeRecovered(IpAddress mcastIp,
Map<HostId, Set<ConnectPoint>> newSinks,
Map<HostId, Set<ConnectPoint>> prevSinks) {
// Iterate over the sinks in order to build the set
// of the connect points to be served by this group
final Set<ConnectPoint> sinksToBeProcessed = Sets.newHashSet();
newSinks.forEach((hostId, connectPoints) -> {
// If it has more than 1 locations
if (connectPoints.size() > 1 || connectPoints.size() == 0) {
log.debug("Skip {} since sink {} has {} locations",
mcastIp, hostId, connectPoints.size());
return;
}
// If previously it had two locations, we need to recover it
// Filter out if the remaining location is already served
if (prevSinks.containsKey(hostId) && prevSinks.get(hostId).size() == 2) {
ConnectPoint sinkToBeProcessed = connectPoints.stream()
.filter(connectPoint -> !isSink(mcastIp, connectPoint))
.findFirst().orElse(null);
if (sinkToBeProcessed != null) {
sinksToBeProcessed.add(sinkToBeProcessed);
}
}
});
return sinksToBeProcessed;
}
/**
* Process all the sinks related to a mcast group and return
* the ones to be processed.
*
* @param source the source connect point
* @param mcastIp the group address
* @param sinks the sinks to be evaluated
* @return the set of the sinks to be processed
*/
private Set<ConnectPoint> processSinksToBeAdded(ConnectPoint source, IpAddress mcastIp,
Map<HostId, Set<ConnectPoint>> sinks) {
// Iterate over the sinks in order to build the set
// of the connect points to be served by this group
final Set<ConnectPoint> sinksToBeProcessed = Sets.newHashSet();
sinks.forEach(((hostId, connectPoints) -> {
// If it has more than 2 locations
if (connectPoints.size() > 2 || connectPoints.size() == 0) {
log.debug("Skip {} since sink {} has {} locations",
mcastIp, hostId, connectPoints.size());
return;
}
// If it has one location, just use it
if (connectPoints.size() == 1) {
sinksToBeProcessed.add(connectPoints.stream()
.findFirst().orElse(null));
return;
}
// We prefer to reuse existing flows
ConnectPoint sinkToBeProcessed = connectPoints.stream()
.filter(connectPoint -> isSink(mcastIp, connectPoint))
.findFirst().orElse(null);
if (sinkToBeProcessed != null) {
sinksToBeProcessed.add(sinkToBeProcessed);
return;
}
// Otherwise we prefer to reuse existing egresses
Set<DeviceId> egresses = getDevice(mcastIp, EGRESS);
sinkToBeProcessed = connectPoints.stream()
.filter(connectPoint -> egresses.contains(connectPoint.deviceId()))
.findFirst().orElse(null);
if (sinkToBeProcessed != null) {
sinksToBeProcessed.add(sinkToBeProcessed);
return;
}
// Otherwise we prefer a location co-located with the source (if it exists)
sinkToBeProcessed = connectPoints.stream()
.filter(connectPoint -> connectPoint.deviceId().equals(source.deviceId()))
.findFirst().orElse(null);
if (sinkToBeProcessed != null) {
sinksToBeProcessed.add(sinkToBeProcessed);
return;
}
// Finally, we randomly pick a new location
sinksToBeProcessed.add(connectPoints.stream()
.findFirst().orElse(null));
}));
// We have done, return the set
return sinksToBeProcessed;
}
/**
* Utility method to remove all the ingress transit ports.
*
* @param mcastIp the group ip
* @param ingressDevice the ingress device for this group
* @param source the source connect point
*/
private void removeIngressTransitPorts(IpAddress mcastIp, DeviceId ingressDevice,
ConnectPoint source) {
Set<PortNumber> ingressTransitPorts = ingressTransitPort(mcastIp);
ingressTransitPorts.forEach(ingressTransitPort -> {
if (ingressTransitPort != null) {
boolean isLast = removePortFromDevice(ingressDevice, ingressTransitPort,
mcastIp, mcastUtils.assignedVlan(source));
if (isLast) {
mcastRoleStore.remove(new McastStoreKey(mcastIp, ingressDevice));
}
}
});
}
/**
* Adds a port to given multicast group on given device. This involves the
* update of L3 multicast group and multicast routing table entry.
*
* @param deviceId device ID
* @param port port to be added
* @param mcastIp multicast group
* @param assignedVlan assigned VLAN ID
*/
private void addPortToDevice(DeviceId deviceId, PortNumber port,
IpAddress mcastIp, VlanId assignedVlan) {
McastStoreKey mcastStoreKey = new McastStoreKey(mcastIp, deviceId);
ImmutableSet.Builder<PortNumber> portBuilder = ImmutableSet.builder();
NextObjective newNextObj;
if (!mcastNextObjStore.containsKey(mcastStoreKey)) {
// First time someone request this mcast group via this device
portBuilder.add(port);
// New nextObj
newNextObj = mcastUtils.nextObjBuilder(mcastIp, assignedVlan,
portBuilder.build(), null).add();
// Store the new port
mcastNextObjStore.put(mcastStoreKey, newNextObj);
} else {
// This device already serves some subscribers of this mcast group
NextObjective nextObj = mcastNextObjStore.get(mcastStoreKey).value();
// Stop if the port is already in the nextobj
Set<PortNumber> existingPorts = mcastUtils.getPorts(nextObj.next());
if (existingPorts.contains(port)) {
log.info("NextObj for {}/{} already exists. Abort", deviceId, port);
return;
}
// Let's add the port and reuse the previous one
portBuilder.addAll(existingPorts).add(port);
// Reuse previous nextObj
newNextObj = mcastUtils.nextObjBuilder(mcastIp, assignedVlan,
portBuilder.build(), nextObj.id()).addToExisting();
// Store the final next objective and send only the difference to the driver
mcastNextObjStore.put(mcastStoreKey, newNextObj);
// Add just the new port
portBuilder = ImmutableSet.builder();
portBuilder.add(port);
newNextObj = mcastUtils.nextObjBuilder(mcastIp, assignedVlan,
portBuilder.build(), nextObj.id()).addToExisting();
}
// Create, store and apply the new nextObj and fwdObj
ObjectiveContext context = new DefaultObjectiveContext(
(objective) -> log.debug("Successfully add {} on {}/{}, vlan {}",
mcastIp, deviceId, port.toLong(), assignedVlan),
(objective, error) ->
log.warn("Failed to add {} on {}/{}, vlan {}: {}",
mcastIp, deviceId, port.toLong(), assignedVlan, error));
ForwardingObjective fwdObj = mcastUtils.fwdObjBuilder(mcastIp, assignedVlan,
newNextObj.id()).add(context);
srManager.flowObjectiveService.next(deviceId, newNextObj);
srManager.flowObjectiveService.forward(deviceId, fwdObj);
}
/**
* Removes a port from given multicast group on given device.
* This involves the update of L3 multicast group and multicast routing
* table entry.
*
* @param deviceId device ID
* @param port port to be added
* @param mcastIp multicast group
* @param assignedVlan assigned VLAN ID
* @return true if this is the last sink on this device
*/
private boolean removePortFromDevice(DeviceId deviceId, PortNumber port,
IpAddress mcastIp, VlanId assignedVlan) {
McastStoreKey mcastStoreKey =
new McastStoreKey(mcastIp, deviceId);
// This device is not serving this multicast group
if (!mcastNextObjStore.containsKey(mcastStoreKey)) {
log.warn("{} is not serving {} on port {}. Abort.", deviceId, mcastIp, port);
return false;
}
NextObjective nextObj = mcastNextObjStore.get(mcastStoreKey).value();
Set<PortNumber> existingPorts = mcastUtils.getPorts(nextObj.next());
// This port does not serve this multicast group
if (!existingPorts.contains(port)) {
log.warn("{} is not serving {} on port {}. Abort.", deviceId, mcastIp, port);
return false;
}
// Copy and modify the ImmutableSet
existingPorts = Sets.newHashSet(existingPorts);
existingPorts.remove(port);
NextObjective newNextObj;
ObjectiveContext context;
ForwardingObjective fwdObj;
if (existingPorts.isEmpty()) {
// If this is the last sink, remove flows and last bucket
// NOTE: Rely on GroupStore garbage collection rather than explicitly
// remove L3MG since there might be other flows/groups refer to
// the same L2IG
context = new DefaultObjectiveContext(
(objective) -> log.debug("Successfully remove {} on {}/{}, vlan {}",
mcastIp, deviceId, port.toLong(), assignedVlan),
(objective, error) ->
log.warn("Failed to remove {} on {}/{}, vlan {}: {}",
mcastIp, deviceId, port.toLong(), assignedVlan, error));
fwdObj = mcastUtils.fwdObjBuilder(mcastIp, assignedVlan, nextObj.id()).remove(context);
mcastNextObjStore.remove(mcastStoreKey);
} else {
// If this is not the last sink, update flows and groups
context = new DefaultObjectiveContext(
(objective) -> log.debug("Successfully update {} on {}/{}, vlan {}",
mcastIp, deviceId, port.toLong(), assignedVlan),
(objective, error) ->
log.warn("Failed to update {} on {}/{}, vlan {}: {}",
mcastIp, deviceId, port.toLong(), assignedVlan, error));
// Here we store the next objective with the remaining port
newNextObj = mcastUtils.nextObjBuilder(mcastIp, assignedVlan,
existingPorts, nextObj.id()).removeFromExisting();
fwdObj = mcastUtils.fwdObjBuilder(mcastIp, assignedVlan, newNextObj.id()).add(context);
mcastNextObjStore.put(mcastStoreKey, newNextObj);
}
// Let's modify the next objective removing the bucket
newNextObj = mcastUtils.nextObjBuilder(mcastIp, assignedVlan,
ImmutableSet.of(port), nextObj.id()).removeFromExisting();
srManager.flowObjectiveService.next(deviceId, newNextObj);
srManager.flowObjectiveService.forward(deviceId, fwdObj);
return existingPorts.isEmpty();
}
/**
* Removes entire group on given device.
*
* @param deviceId device ID
* @param mcastIp multicast group to be removed
* @param assignedVlan assigned VLAN ID
*/
private void removeGroupFromDevice(DeviceId deviceId, IpAddress mcastIp,
VlanId assignedVlan) {
McastStoreKey mcastStoreKey = new McastStoreKey(mcastIp, deviceId);
// This device is not serving this multicast group
if (!mcastNextObjStore.containsKey(mcastStoreKey)) {
log.warn("{} is not serving {}. Abort.", deviceId, mcastIp);
return;
}
NextObjective nextObj = mcastNextObjStore.get(mcastStoreKey).value();
// NOTE: Rely on GroupStore garbage collection rather than explicitly
// remove L3MG since there might be other flows/groups refer to
// the same L2IG
ObjectiveContext context = new DefaultObjectiveContext(
(objective) -> log.debug("Successfully remove {} on {}, vlan {}",
mcastIp, deviceId, assignedVlan),
(objective, error) ->
log.warn("Failed to remove {} on {}, vlan {}: {}",
mcastIp, deviceId, assignedVlan, error));
ForwardingObjective fwdObj = mcastUtils.fwdObjBuilder(mcastIp, assignedVlan, nextObj.id()).remove(context);
srManager.flowObjectiveService.forward(deviceId, fwdObj);
mcastNextObjStore.remove(mcastStoreKey);
mcastRoleStore.remove(mcastStoreKey);
}
private void installPath(IpAddress mcastIp, ConnectPoint source, Path mcastPath) {
// Get Links
List<Link> links = mcastPath.links();
// Setup new ingress mcast role
mcastRoleStore.put(new McastStoreKey(mcastIp, links.get(0).src().deviceId()),
INGRESS);
// For each link, modify the next on the source device adding the src port
// and a new filter objective on the destination port
links.forEach(link -> {
addPortToDevice(link.src().deviceId(), link.src().port(), mcastIp,
mcastUtils.assignedVlan(link.src().deviceId().equals(source.deviceId()) ? source : null));
mcastUtils.addFilterToDevice(link.dst().deviceId(), link.dst().port(),
mcastUtils.assignedVlan(null), mcastIp, null);
});
// Setup mcast role for the transit
links.stream()
.filter(link -> !link.src().deviceId().equals(source.deviceId()))
.forEach(link -> mcastRoleStore.put(new McastStoreKey(mcastIp, link.src().deviceId()),
TRANSIT));
}
/**
* Go through all the paths, looking for shared links to be used
* in the final path computation.
*
* @param egresses egress devices
* @param availablePaths all the available paths towards the egress
* @return shared links between egress devices
*/
private Set<Link> exploreMcastTree(Set<DeviceId> egresses,
Map<DeviceId, List<Path>> availablePaths) {
// Length of the shortest path
int minLength = Integer.MAX_VALUE;
int length;
// Current paths
List<Path> currentPaths;
// Verify the source can still reach all the egresses
for (DeviceId egress : egresses) {
// From the source we cannot reach all the sinks
// just continue and let's figure out after
currentPaths = availablePaths.get(egress);
if (currentPaths.isEmpty()) {
continue;
}
// Get the length of the first one available,
// update the min length
length = currentPaths.get(0).links().size();
if (length < minLength) {
minLength = length;
}
}
// If there are no paths
if (minLength == Integer.MAX_VALUE) {
return Collections.emptySet();
}
// Iterate looking for shared links
int index = 0;
// Define the sets for the intersection
Set<Link> sharedLinks = Sets.newHashSet();
Set<Link> currentSharedLinks;
Set<Link> currentLinks;
DeviceId egressToRemove = null;
// Let's find out the shared links
while (index < minLength) {
// Initialize the intersection with the paths related to the first egress
currentPaths = availablePaths.get(
egresses.stream()
.findFirst()
.orElse(null)
);
currentSharedLinks = Sets.newHashSet();
// Iterate over the paths and take the "index" links
for (Path path : currentPaths) {
currentSharedLinks.add(path.links().get(index));
}
// Iterate over the remaining egress
for (DeviceId egress : egresses) {
// Iterate over the paths and take the "index" links
currentLinks = Sets.newHashSet();
for (Path path : availablePaths.get(egress)) {
currentLinks.add(path.links().get(index));
}
// Do intersection
currentSharedLinks = Sets.intersection(currentSharedLinks, currentLinks);
// If there are no shared paths exit and record the device to remove
// we have to retry with a subset of sinks
if (currentSharedLinks.isEmpty()) {
egressToRemove = egress;
index = minLength;
break;
}
}
sharedLinks.addAll(currentSharedLinks);
index++;
}
// If the shared links is empty and there are egress
// let's retry another time with less sinks, we can
// still build optimal subtrees
if (sharedLinks.isEmpty() && egresses.size() > 1 && egressToRemove != null) {
egresses.remove(egressToRemove);
sharedLinks = exploreMcastTree(egresses, availablePaths);
}
return sharedLinks;
}
/**
* Build Mcast tree having as root the given source and as leaves the given egress points.
*
* @param source source of the tree
* @param sinks leaves of the tree
* @return the computed Mcast tree
*/
private Map<ConnectPoint, List<Path>> computeSinkMcastTree(DeviceId source,
Set<ConnectPoint> sinks) {
// Get the egress devices, remove source from the egress if present
Set<DeviceId> egresses = sinks.stream()
.map(ConnectPoint::deviceId)
.filter(deviceId -> !deviceId.equals(source))
.collect(Collectors.toSet());
Map<DeviceId, List<Path>> mcastTree = computeMcastTree(source, egresses);
// Build final tree and return it as it is
final Map<ConnectPoint, List<Path>> finalTree = Maps.newHashMap();
// We need to put back the source if it was originally present
sinks.forEach(sink -> {
List<Path> sinkPaths = mcastTree.get(sink.deviceId());
finalTree.put(sink, sinkPaths != null ? sinkPaths : ImmutableList.of());
});
return finalTree;
}
/**
* Build Mcast tree having as root the given source and as leaves the given egress.
*
* @param source source of the tree
* @param egresses leaves of the tree
* @return the computed Mcast tree
*/
private Map<DeviceId, List<Path>> computeMcastTree(DeviceId source,
Set<DeviceId> egresses) {
// Pre-compute all the paths
Map<DeviceId, List<Path>> availablePaths = Maps.newHashMap();
// No links to enforce
egresses.forEach(egress -> availablePaths.put(egress, getPaths(source, egress,
Collections.emptySet())));
// Explore the topology looking for shared links amongst the egresses
Set<Link> linksToEnforce = exploreMcastTree(Sets.newHashSet(egresses), availablePaths);
// Remove all the paths from the previous computation
availablePaths.clear();
// Build the final paths enforcing the shared links between egress devices
egresses.forEach(egress -> availablePaths.put(egress, getPaths(source, egress,
linksToEnforce)));
return availablePaths;
}
/**
* Gets path from src to dst computed using the custom link weigher.
*
* @param src source device ID
* @param dst destination device ID
* @return list of paths from src to dst
*/
private List<Path> getPaths(DeviceId src, DeviceId dst, Set<Link> linksToEnforce) {
// Takes a snapshot of the topology
final Topology currentTopology = topologyService.currentTopology();
// Build a specific link weigher for this path computation
final LinkWeigher linkWeigher = new SRLinkWeigher(srManager, src, linksToEnforce);
// We will use our custom link weigher for our path
// computations and build the list of valid paths
List<Path> allPaths = Lists.newArrayList(
topologyService.getPaths(currentTopology, src, dst, linkWeigher)
);
// If there are no valid paths, just exit
log.debug("{} path(s) found from {} to {}", allPaths.size(), src, dst);
return allPaths;
}
/**
* Gets a path from src to dst.
* If a path was allocated before, returns the allocated path.
* Otherwise, randomly pick one from available paths.
*
* @param src source device ID
* @param dst destination device ID
* @param mcastIp multicast group
* @param allPaths paths list
* @return an optional path from src to dst
*/
private Optional<Path> getPath(DeviceId src, DeviceId dst,
IpAddress mcastIp, List<Path> allPaths) {
// Firstly we get all the valid paths, if the supplied are null
if (allPaths == null) {
allPaths = getPaths(src, dst, Collections.emptySet());
}
// If there are no paths just exit
if (allPaths.isEmpty()) {
return Optional.empty();
}
// Create a map index of suitablity-to-list of paths. For example
// a path in the list associated to the index 1 shares only the
// first hop and it is less suitable of a path belonging to the index
// 2 that shares leaf-spine.
Map<Integer, List<Path>> eligiblePaths = Maps.newHashMap();
// Some init steps
int nhop;
McastStoreKey mcastStoreKey;
Link hop;
PortNumber srcPort;
Set<PortNumber> existingPorts;
NextObjective nextObj;
// Iterate over paths looking for eligible paths
for (Path path : allPaths) {
// Unlikely, it will happen...
if (!src.equals(path.links().get(0).src().deviceId())) {
continue;
}
nhop = 0;
// Iterate over the links
while (nhop < path.links().size()) {
// Get the link and verify if a next related
// to the src device exist in the store
hop = path.links().get(nhop);
mcastStoreKey = new McastStoreKey(mcastIp, hop.src().deviceId());
// It does not exist in the store, exit
if (!mcastNextObjStore.containsKey(mcastStoreKey)) {
break;
}
// Get the output ports on the next
nextObj = mcastNextObjStore.get(mcastStoreKey).value();
existingPorts = mcastUtils.getPorts(nextObj.next());
// And the src port on the link
srcPort = hop.src().port();
// the src port is not used as output, exit
if (!existingPorts.contains(srcPort)) {
break;
}
nhop++;
}
// n_hop defines the index
if (nhop > 0) {
eligiblePaths.compute(nhop, (index, paths) -> {
paths = paths == null ? Lists.newArrayList() : paths;
paths.add(path);
return paths;
});
}
}
// No suitable paths
if (eligiblePaths.isEmpty()) {
log.debug("No eligiblePath(s) found from {} to {}", src, dst);
// Otherwise, randomly pick a path
Collections.shuffle(allPaths);
return allPaths.stream().findFirst();
}
// Let's take the best ones
Integer bestIndex = eligiblePaths.keySet()
.stream()
.sorted(Comparator.reverseOrder())
.findFirst().orElse(null);
List<Path> bestPaths = eligiblePaths.get(bestIndex);
log.debug("{} eligiblePath(s) found from {} to {}",
bestPaths.size(), src, dst);
// randomly pick a path on the highest index
Collections.shuffle(bestPaths);
return bestPaths.stream().findFirst();
}
/**
* Gets device(s) of given role in given multicast group.
*
* @param mcastIp multicast IP
* @param role multicast role
* @return set of device ID or empty set if not found
*/
private Set<DeviceId> getDevice(IpAddress mcastIp, McastRole role) {
return mcastRoleStore.entrySet().stream()
.filter(entry -> entry.getKey().mcastIp().equals(mcastIp) &&
entry.getValue().value() == role)
.map(Entry::getKey).map(McastStoreKey::deviceId)
.collect(Collectors.toSet());
}
/**
* Gets groups which is affected by the link down event.
*
* @param link link going down
* @return a set of multicast IpAddress
*/
private Set<IpAddress> getAffectedGroups(Link link) {
DeviceId deviceId = link.src().deviceId();
PortNumber port = link.src().port();
return mcastNextObjStore.entrySet().stream()
.filter(entry -> entry.getKey().deviceId().equals(deviceId) &&
mcastUtils.getPorts(entry.getValue().value().next()).contains(port))
.map(Entry::getKey).map(McastStoreKey::mcastIp)
.collect(Collectors.toSet());
}
/**
* Gets groups which are affected by the device down event.
*
* @param deviceId device going down
* @return a set of multicast IpAddress
*/
private Set<IpAddress> getAffectedGroups(DeviceId deviceId) {
return mcastNextObjStore.entrySet().stream()
.filter(entry -> entry.getKey().deviceId().equals(deviceId))
.map(Entry::getKey).map(McastStoreKey::mcastIp)
.collect(Collectors.toSet());
}
/**
* Gets the spine-facing port on ingress device of given multicast group.
*
* @param mcastIp multicast IP
* @return spine-facing port on ingress device
*/
private Set<PortNumber> ingressTransitPort(IpAddress mcastIp) {
DeviceId ingressDevice = getDevice(mcastIp, INGRESS)
.stream().findAny().orElse(null);
ImmutableSet.Builder<PortNumber> portBuilder = ImmutableSet.builder();
if (ingressDevice != null) {
NextObjective nextObj = mcastNextObjStore
.get(new McastStoreKey(mcastIp, ingressDevice)).value();
Set<PortNumber> ports = mcastUtils.getPorts(nextObj.next());
// Let's find out all the ingress-transit ports
for (PortNumber port : ports) {
// Spine-facing port should have no subnet and no xconnect
if (srManager.deviceConfiguration() != null &&
srManager.deviceConfiguration().getPortSubnets(ingressDevice, port).isEmpty() &&
!srManager.xConnectHandler.hasXConnect(new ConnectPoint(ingressDevice, port))) {
portBuilder.add(port);
}
}
}
return portBuilder.build();
}
/**
* Verify if the given device has sinks
* for the multicast group.
*
* @param deviceId device Id
* @param mcastIp multicast IP
* @return true if the device has sink for the group.
* False otherwise.
*/
private boolean hasSinks(DeviceId deviceId, IpAddress mcastIp) {
if (deviceId != null) {
// Get the nextobjective
Versioned<NextObjective> versionedNextObj = mcastNextObjStore.get(
new McastStoreKey(mcastIp, deviceId)
);
// If it exists
if (versionedNextObj != null) {
NextObjective nextObj = versionedNextObj.value();
// Retrieves all the output ports
Set<PortNumber> ports = mcastUtils.getPorts(nextObj.next());
// Tries to find at least one port that is not spine-facing
for (PortNumber port : ports) {
// Spine-facing port should have no subnet and no xconnect
if (srManager.deviceConfiguration() != null &&
(!srManager.deviceConfiguration().getPortSubnets(deviceId, port).isEmpty() ||
srManager.xConnectHandler.hasXConnect(new ConnectPoint(deviceId, port)))) {
return true;
}
}
}
}
return false;
}
/**
* Verify if a given connect point is sink for this group.
*
* @param mcastIp group address
* @param connectPoint connect point to be verified
* @return true if the connect point is sink of the group
*/
private boolean isSink(IpAddress mcastIp, ConnectPoint connectPoint) {
// Let's check if we are already serving that location
McastStoreKey mcastStoreKey = new McastStoreKey(mcastIp, connectPoint.deviceId());
if (!mcastNextObjStore.containsKey(mcastStoreKey)) {
return false;
}
// Get next and check with the port
NextObjective mcastNext = mcastNextObjStore.get(mcastStoreKey).value();
return mcastUtils.getPorts(mcastNext.next()).contains(connectPoint.port());
}
/**
* Updates filtering objective for given device and port.
* It is called in general when the mcast config has been
* changed.
*
* @param deviceId device ID
* @param portNum ingress port number
* @param vlanId assigned VLAN ID
* @param install true to add, false to remove
*/
public void updateFilterToDevice(DeviceId deviceId, PortNumber portNum,
VlanId vlanId, boolean install) {
lastMcastChange = Instant.now();
mcastLock();
try {
// Iterates over the route and updates properly the filtering objective
// on the source device.
srManager.multicastRouteService.getRoutes().forEach(mcastRoute -> {
log.debug("Update filter for {}", mcastRoute.group());
// Verify leadership on the operation
if (!mcastUtils.isLeader(mcastRoute.group())) {
log.debug("Skip {} due to lack of leadership", mcastRoute.group());
return;
}
// FIXME To be addressed with multiple sources support
ConnectPoint source = srManager.multicastRouteService.sources(mcastRoute)
.stream()
.findFirst().orElse(null);
if (source.deviceId().equals(deviceId) && source.port().equals(portNum)) {
if (install) {
mcastUtils.addFilterToDevice(deviceId, portNum, vlanId, mcastRoute.group(), INGRESS);
} else {
mcastUtils.removeFilterToDevice(deviceId, portNum, vlanId, mcastRoute.group(), null);
}
}
});
} finally {
mcastUnlock();
}
}
/**
* Performs bucket verification operation for all mcast groups in the devices.
* Firstly, it verifies that mcast is stable before trying verification operation.
* Verification consists in creating new nexts with VERIFY operation. Actually,
* the operation is totally delegated to the driver.
*/
private final class McastBucketCorrector implements Runnable {
@Override
public void run() {
// Verify if the Mcast has been stable for MCAST_STABLITY_THRESHOLD
if (!isMcastStable()) {
return;
}
// Acquires lock
mcastLock();
try {
// Iterates over the routes and verify the related next objectives
srManager.multicastRouteService.getRoutes()
.stream()
.map(McastRoute::group)
.forEach(mcastIp -> {
log.trace("Running mcast buckets corrector for mcast group: {}",
mcastIp);
// For each group we get current information in the store
// and issue a check of the next objectives in place
DeviceId ingressDevice = getDevice(mcastIp, INGRESS)
.stream().findAny().orElse(null);
Set<DeviceId> transitDevices = getDevice(mcastIp, TRANSIT);
Set<DeviceId> egressDevices = getDevice(mcastIp, EGRESS);
// Get source and sinks from Mcast Route Service and warn about errors
ConnectPoint source = mcastUtils.getSource(mcastIp);
Set<ConnectPoint> sinks = mcastUtils.getSinks(mcastIp).values().stream()
.flatMap(Collection::stream)
.collect(Collectors.toSet());
// Do not proceed if ingress device or source of this group are missing
if (ingressDevice == null || source == null) {
if (!sinks.isEmpty()) {
log.warn("Unable to run buckets corrector. " +
"Missing ingress {} or source {} for group {}",
ingressDevice, source, mcastIp);
}
return;
}
// Continue only when this instance is the leader of the group
if (!mcastUtils.isLeader(mcastIp)) {
log.trace("Unable to run buckets corrector. " +
"Skip {} due to lack of leadership", mcastIp);
return;
}
// Create the set of the devices to be processed
ImmutableSet.Builder<DeviceId> devicesBuilder = ImmutableSet.builder();
devicesBuilder.add(ingressDevice);
if (!transitDevices.isEmpty()) {
devicesBuilder.addAll(transitDevices);
}
if (!egressDevices.isEmpty()) {
devicesBuilder.addAll(egressDevices);
}
Set<DeviceId> devicesToProcess = devicesBuilder.build();
// Iterate over the devices
devicesToProcess.forEach(deviceId -> {
McastStoreKey currentKey = new McastStoreKey(mcastIp, deviceId);
// If next exists in our store verify related next objective
if (mcastNextObjStore.containsKey(currentKey)) {
NextObjective currentNext = mcastNextObjStore.get(currentKey).value();
// Get current ports
Set<PortNumber> currentPorts = mcastUtils.getPorts(currentNext.next());
// Rebuild the next objective
currentNext = mcastUtils.nextObjBuilder(
mcastIp,
mcastUtils.assignedVlan(deviceId.equals(source.deviceId()) ?
source : null),
currentPorts,
currentNext.id()
).verify();
// Send to the flowobjective service
srManager.flowObjectiveService.next(deviceId, currentNext);
} else {
log.warn("Unable to run buckets corrector. " +
"Missing next for {} and group {}",
deviceId, mcastIp);
}
});
});
} finally {
// Finally, it releases the lock
mcastUnlock();
}
}
}
public Map<McastStoreKey, Integer> getMcastNextIds(IpAddress mcastIp) {
// If mcast ip is present
if (mcastIp != null) {
return mcastNextObjStore.entrySet().stream()
.filter(mcastEntry -> mcastIp.equals(mcastEntry.getKey().mcastIp()))
.collect(Collectors.toMap(Entry::getKey,
entry -> entry.getValue().value().id()));
}
// Otherwise take all the groups
return mcastNextObjStore.entrySet().stream()
.collect(Collectors.toMap(Entry::getKey,
entry -> entry.getValue().value().id()));
}
/**
* Returns the associated roles to the mcast groups or to the single
* group if mcastIp is present.
*
* @param mcastIp the group ip
* @return the mapping mcastIp-device to mcast role
*
* @deprecated in 1.12 ("Magpie") release.
*/
@Deprecated
public Map<McastStoreKey, McastRole> getMcastRoles(IpAddress mcastIp) {
// If mcast ip is present
if (mcastIp != null) {
return mcastRoleStore.entrySet().stream()
.filter(mcastEntry -> mcastIp.equals(mcastEntry.getKey().mcastIp()))
.collect(Collectors.toMap(Entry::getKey,
entry -> entry.getValue().value()));
}
// Otherwise take all the groups
return mcastRoleStore.entrySet().stream()
.collect(Collectors.toMap(Entry::getKey,
entry -> entry.getValue().value()));
}
/**
* Returns the associated paths to the mcast group.
*
* @param mcastIp the group ip
* @return the mapping egress point to mcast path
*
* @deprecated in 1.12 ("Magpie") release.
*/
@Deprecated
public Map<ConnectPoint, List<ConnectPoint>> getMcastPaths(IpAddress mcastIp) {
Map<ConnectPoint, List<ConnectPoint>> mcastPaths = Maps.newHashMap();
// Get the source
ConnectPoint source = mcastUtils.getSource(mcastIp);
// Source cannot be null, we don't know the starting point
if (source != null) {
// Init steps
Set<DeviceId> visited = Sets.newHashSet();
List<ConnectPoint> currentPath = Lists.newArrayList(
source
);
// Build recursively the mcast paths
buildMcastPaths(source.deviceId(), visited, mcastPaths, currentPath, mcastIp);
}
return mcastPaths;
}
/**
* Returns the associated trees to the mcast group.
*
* @param mcastIp the group ip
* @param sourcecp the source connect point
* @return the mapping egress point to mcast path
*/
public Multimap<ConnectPoint, List<ConnectPoint>> getMcastTrees(IpAddress mcastIp,
ConnectPoint sourcecp) {
Multimap<ConnectPoint, List<ConnectPoint>> mcastTrees = HashMultimap.create();
// Get the sources
Set<ConnectPoint> sources = mcastUtils.getSources(mcastIp);
// If we are providing the source, let's filter out
if (sourcecp != null) {
sources = sources.stream()
.filter(source -> source.equals(sourcecp))
.collect(Collectors.toSet());
}
// Source cannot be null, we don't know the starting point
if (!sources.isEmpty()) {
sources.forEach(source -> {
// Init steps
Map<ConnectPoint, List<ConnectPoint>> mcastPaths = Maps.newHashMap();
Set<DeviceId> visited = Sets.newHashSet();
List<ConnectPoint> currentPath = Lists.newArrayList(source);
// Build recursively the mcast paths
buildMcastPaths(source.deviceId(), visited, mcastPaths, currentPath, mcastIp);
mcastPaths.forEach(mcastTrees::put);
});
}
return mcastTrees;
}
/**
* Build recursively the mcast paths.
*
* @param toVisit the node to visit
* @param visited the visited nodes
* @param mcastPaths the current mcast paths
* @param currentPath the current path
* @param mcastIp the group ip
*/
private void buildMcastPaths(DeviceId toVisit, Set<DeviceId> visited,
Map<ConnectPoint, List<ConnectPoint>> mcastPaths,
List<ConnectPoint> currentPath, IpAddress mcastIp) {
// If we have visited the node to visit
// there is a loop
if (visited.contains(toVisit)) {
return;
}
// Visit next-hop
visited.add(toVisit);
McastStoreKey mcastStoreKey = new McastStoreKey(mcastIp, toVisit);
// Looking for next-hops
if (mcastNextObjStore.containsKey(mcastStoreKey)) {
// Build egress connectpoints
NextObjective nextObjective = mcastNextObjStore.get(mcastStoreKey).value();
// Get Ports
Set<PortNumber> outputPorts = mcastUtils.getPorts(nextObjective.next());
// Build relative cps
ImmutableSet.Builder<ConnectPoint> cpBuilder = ImmutableSet.builder();
outputPorts.forEach(portNumber -> cpBuilder.add(new ConnectPoint(toVisit, portNumber)));
Set<ConnectPoint> egressPoints = cpBuilder.build();
// Define other variables for the next steps
Set<Link> egressLinks;
List<ConnectPoint> newCurrentPath;
Set<DeviceId> newVisited;
DeviceId newToVisit;
for (ConnectPoint egressPoint : egressPoints) {
egressLinks = srManager.linkService.getEgressLinks(egressPoint);
// If it does not have egress links, stop
if (egressLinks.isEmpty()) {
// Add the connect points to the path
newCurrentPath = Lists.newArrayList(currentPath);
newCurrentPath.add(0, egressPoint);
// Save in the map
mcastPaths.put(egressPoint, newCurrentPath);
} else {
newVisited = Sets.newHashSet(visited);
// Iterate over the egress links for the next hops
for (Link egressLink : egressLinks) {
// Update to visit
newToVisit = egressLink.dst().deviceId();
// Add the connect points to the path
newCurrentPath = Lists.newArrayList(currentPath);
newCurrentPath.add(0, egressPoint);
newCurrentPath.add(0, egressLink.dst());
// Go to the next hop
buildMcastPaths(newToVisit, newVisited, mcastPaths, newCurrentPath, mcastIp);
}
}
}
}
}
/**
* Return the leaders of the mcast groups.
*
* @param mcastIp the group ip
* @return the mapping group-node
*/
public Map<IpAddress, NodeId> getMcastLeaders(IpAddress mcastIp) {
return mcastUtils.getMcastLeaders(mcastIp);
}
}