2 * Copyright (c) 2021 PANTHEON.tech, s.r.o. and others. All rights reserved.
4 * This program and the accompanying materials are made available under the
5 * terms of the Eclipse Public License v1.0 which accompanies this distribution,
6 * and is available at http://www.eclipse.org/legal/epl-v10.html
8 package org.opendaylight.controller.eos.akka.owner.supervisor;
10 import static com.google.common.base.Verify.verifyNotNull;
11 import static java.util.Objects.requireNonNull;
13 import akka.actor.typed.ActorRef;
14 import akka.actor.typed.Behavior;
15 import akka.actor.typed.javadsl.ActorContext;
16 import akka.actor.typed.javadsl.Behaviors;
17 import akka.actor.typed.javadsl.Receive;
18 import akka.cluster.ClusterEvent;
19 import akka.cluster.ClusterEvent.CurrentClusterState;
20 import akka.cluster.Member;
21 import akka.cluster.ddata.LWWRegister;
22 import akka.cluster.ddata.LWWRegisterKey;
23 import akka.cluster.ddata.ORMap;
24 import akka.cluster.ddata.ORSet;
25 import akka.cluster.ddata.SelfUniqueAddress;
26 import akka.cluster.ddata.typed.javadsl.DistributedData;
27 import akka.cluster.ddata.typed.javadsl.Replicator;
28 import akka.cluster.ddata.typed.javadsl.ReplicatorMessageAdapter;
29 import akka.cluster.typed.Cluster;
30 import akka.cluster.typed.Subscribe;
31 import akka.pattern.StatusReply;
32 import com.google.common.collect.HashMultimap;
33 import com.google.common.collect.ImmutableList;
34 import com.google.common.collect.ImmutableSet;
35 import com.google.common.collect.Multimap;
36 import com.google.common.collect.Sets;
37 import java.time.Duration;
38 import java.util.ArrayList;
39 import java.util.Collection;
40 import java.util.HashSet;
41 import java.util.List;
44 import java.util.function.BiPredicate;
45 import java.util.stream.Collectors;
46 import java.util.stream.StreamSupport;
47 import org.opendaylight.controller.eos.akka.owner.supervisor.command.AbstractEntityRequest;
48 import org.opendaylight.controller.eos.akka.owner.supervisor.command.CandidatesChanged;
49 import org.opendaylight.controller.eos.akka.owner.supervisor.command.ClearCandidates;
50 import org.opendaylight.controller.eos.akka.owner.supervisor.command.ClearCandidatesForMember;
51 import org.opendaylight.controller.eos.akka.owner.supervisor.command.DataCenterDeactivated;
52 import org.opendaylight.controller.eos.akka.owner.supervisor.command.DeactivateDataCenter;
53 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntitiesBackendReply;
54 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntitiesBackendRequest;
55 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntityBackendReply;
56 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntityBackendRequest;
57 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntityOwnerBackendReply;
58 import org.opendaylight.controller.eos.akka.owner.supervisor.command.GetEntityOwnerBackendRequest;
59 import org.opendaylight.controller.eos.akka.owner.supervisor.command.MemberDownEvent;
60 import org.opendaylight.controller.eos.akka.owner.supervisor.command.MemberReachableEvent;
61 import org.opendaylight.controller.eos.akka.owner.supervisor.command.MemberUnreachableEvent;
62 import org.opendaylight.controller.eos.akka.owner.supervisor.command.MemberUpEvent;
63 import org.opendaylight.controller.eos.akka.owner.supervisor.command.OwnerChanged;
64 import org.opendaylight.controller.eos.akka.owner.supervisor.command.OwnerSupervisorCommand;
65 import org.opendaylight.controller.eos.akka.registry.candidate.CandidateRegistry;
66 import org.opendaylight.mdsal.eos.dom.api.DOMEntity;
67 import org.opendaylight.yangtools.binding.data.codec.api.BindingInstanceIdentifierCodec;
68 import org.slf4j.Logger;
69 import org.slf4j.LoggerFactory;
70 import scala.collection.JavaConverters;
73 * Responsible for tracking candidates and assigning ownership of entities. This behavior is subscribed to the candidate
74 * registry in distributed-data and picks entity owners based on the current cluster state and registered candidates.
75 * On cluster up/down etc. events the owners are reassigned if possible.
77 public final class OwnerSupervisor extends AbstractSupervisor {
79 private static final Logger LOG = LoggerFactory.getLogger(OwnerSupervisor.class);
80 private static final String DATACENTER_PREFIX = "dc-";
82 private final ReplicatorMessageAdapter<OwnerSupervisorCommand, LWWRegister<String>> ownerReplicator;
84 // Our own clock implementation so we do not have to rely on synchronized clocks. This basically functions as an
85 // increasing counter which is fine for our needs as we only ever have a single writer since t supervisor is
86 // running in a cluster-singleton
87 private static final LWWRegister.Clock<String> CLOCK = (currentTimestamp, value) -> currentTimestamp + 1;
89 private final Cluster cluster;
90 private final SelfUniqueAddress node;
91 private final String dataCenter;
93 private final Set<String> activeMembers;
95 // currently registered candidates
96 private final Map<DOMEntity, Set<String>> currentCandidates;
98 private final Map<DOMEntity, String> currentOwners;
99 // reverse lookup of owner to entity
100 private final Multimap<String, DOMEntity> ownerToEntity = HashMultimap.create();
102 // only reassign owner for those entities that lost this candidate or is not reachable
103 private final BiPredicate<DOMEntity, String> reassignPredicate = (entity, candidate) ->
104 !isActiveCandidate(candidate) || !isCandidateFor(entity, candidate);
106 private final BindingInstanceIdentifierCodec iidCodec;
108 private OwnerSupervisor(final ActorContext<OwnerSupervisorCommand> context,
109 final Map<DOMEntity, Set<String>> currentCandidates,
110 final Map<DOMEntity, String> currentOwners,
111 final BindingInstanceIdentifierCodec iidCodec) {
113 this.iidCodec = requireNonNull(iidCodec);
115 final DistributedData distributedData = DistributedData.get(context.getSystem());
116 final ActorRef<Replicator.Command> replicator = distributedData.replicator();
118 cluster = Cluster.get(context.getSystem());
119 ownerReplicator = new ReplicatorMessageAdapter<>(context, replicator, Duration.ofSeconds(5));
120 dataCenter = extractDatacenterRole(cluster.selfMember());
122 node = distributedData.selfUniqueAddress();
123 activeMembers = getActiveMembers();
125 this.currentCandidates = currentCandidates;
126 this.currentOwners = currentOwners;
128 for (final Map.Entry<DOMEntity, String> entry : currentOwners.entrySet()) {
129 ownerToEntity.put(entry.getValue(), entry.getKey());
132 // check whether we have any unreachable/missing owners
133 reassignUnreachableOwners();
134 assignMissingOwners();
136 final ActorRef<ClusterEvent.MemberEvent> memberEventAdapter =
137 context.messageAdapter(ClusterEvent.MemberEvent.class, event -> {
138 if (event instanceof ClusterEvent.MemberUp) {
139 return new MemberUpEvent(event.member().address(), event.member().getRoles());
141 return new MemberDownEvent(event.member().address(), event.member().getRoles());
144 cluster.subscriptions().tell(Subscribe.create(memberEventAdapter, ClusterEvent.MemberEvent.class));
146 final ActorRef<ClusterEvent.ReachabilityEvent> reachabilityEventAdapter =
147 context.messageAdapter(ClusterEvent.ReachabilityEvent.class, event -> {
148 if (event instanceof ClusterEvent.ReachableMember) {
149 return new MemberReachableEvent(event.member().address(), event.member().getRoles());
151 return new MemberUnreachableEvent(event.member().address(), event.member().getRoles());
154 cluster.subscriptions().tell(Subscribe.create(reachabilityEventAdapter, ClusterEvent.ReachabilityEvent.class));
156 candidateReplicator.subscribe(CandidateRegistry.KEY, CandidatesChanged::new);
158 LOG.debug("Owner Supervisor started");
161 public static Behavior<OwnerSupervisorCommand> create(final Map<DOMEntity, Set<String>> currentCandidates,
162 final Map<DOMEntity, String> currentOwners, final BindingInstanceIdentifierCodec iidCodec) {
163 return Behaviors.setup(ctx -> new OwnerSupervisor(ctx, currentCandidates, currentOwners, iidCodec));
167 public Receive<OwnerSupervisorCommand> createReceive() {
168 return newReceiveBuilder()
169 .onMessage(CandidatesChanged.class, this::onCandidatesChanged)
170 .onMessage(DeactivateDataCenter.class, this::onDeactivateDatacenter)
171 .onMessage(OwnerChanged.class, this::onOwnerChanged)
172 .onMessage(MemberUpEvent.class, this::onPeerUp)
173 .onMessage(MemberDownEvent.class, this::onPeerDown)
174 .onMessage(MemberReachableEvent.class, this::onPeerReachable)
175 .onMessage(MemberUnreachableEvent.class, this::onPeerUnreachable)
176 .onMessage(GetEntitiesBackendRequest.class, this::onGetEntities)
177 .onMessage(GetEntityBackendRequest.class, this::onGetEntity)
178 .onMessage(GetEntityOwnerBackendRequest.class, this::onGetEntityOwner)
179 .onMessage(ClearCandidatesForMember.class, this::onClearCandidatesForMember)
180 .onMessage(ClearCandidates.class, this::finishClearCandidates)
184 private Behavior<OwnerSupervisorCommand> onDeactivateDatacenter(final DeactivateDataCenter command) {
185 LOG.debug("Deactivating Owner Supervisor on {}", cluster.selfMember());
186 command.getReplyTo().tell(DataCenterDeactivated.INSTANCE);
187 return IdleSupervisor.create(iidCodec);
190 private Behavior<OwnerSupervisorCommand> onOwnerChanged(final OwnerChanged command) {
191 LOG.debug("Owner has changed for {}", command.getResponse().key());
195 private void reassignUnreachableOwners() {
196 final Set<String> ownersToReassign = new HashSet<>();
197 for (final String owner : ownerToEntity.keys()) {
198 if (!isActiveCandidate(owner)) {
199 ownersToReassign.add(owner);
203 for (final String owner : ownersToReassign) {
204 reassignCandidatesFor(owner, ImmutableList.copyOf(ownerToEntity.get(owner)), reassignPredicate);
208 private void assignMissingOwners() {
209 for (final Map.Entry<DOMEntity, Set<String>> entry : currentCandidates.entrySet()) {
210 if (!currentOwners.containsKey(entry.getKey())) {
211 assignOwnerFor(entry.getKey());
216 private Behavior<OwnerSupervisorCommand> onCandidatesChanged(final CandidatesChanged message) {
217 LOG.debug("onCandidatesChanged {}", message.getResponse());
218 if (message.getResponse() instanceof Replicator.Changed) {
219 final Replicator.Changed<ORMap<DOMEntity, ORSet<String>>> changed =
220 (Replicator.Changed<ORMap<DOMEntity, ORSet<String>>>) message.getResponse();
221 processCandidateChanges(changed.get(CandidateRegistry.KEY));
226 private void processCandidateChanges(final ORMap<DOMEntity, ORSet<String>> candidates) {
227 final Map<DOMEntity, ORSet<String>> entries = candidates.getEntries();
228 for (final Map.Entry<DOMEntity, ORSet<String>> entry : entries.entrySet()) {
229 processCandidatesFor(entry.getKey(), entry.getValue());
233 private void processCandidatesFor(final DOMEntity entity, final ORSet<String> receivedCandidates) {
234 LOG.debug("Processing candidates for : {}, new value: {}", entity, receivedCandidates.elements());
236 final Set<String> candidates = JavaConverters.asJava(receivedCandidates.elements());
237 // only insert candidates if there are any to insert, otherwise we would generate unnecessary notification with
239 if (!currentCandidates.containsKey(entity) && !candidates.isEmpty()) {
240 LOG.debug("Candidates missing for entity: {} adding all candidates", entity);
241 currentCandidates.put(entity, new HashSet<>(candidates));
243 LOG.debug("Current state for {} : {}", entity, currentCandidates.get(entity).toString());
244 assignOwnerFor(entity);
249 final Set<String> currentlyPresent = currentCandidates.getOrDefault(entity, Set.of());
250 final Set<String> difference = ImmutableSet.copyOf(Sets.symmetricDifference(currentlyPresent, candidates));
252 LOG.debug("currently present candidates: {}", currentlyPresent);
253 LOG.debug("difference: {}", difference);
255 final List<String> ownersToReassign = new ArrayList<>();
257 // first add/remove candidates from entities
258 for (final String toCheck : difference) {
259 if (!currentlyPresent.contains(toCheck)) {
261 LOG.debug("Adding new candidate for entity: {} : {}", entity, toCheck);
262 currentCandidates.get(entity).add(toCheck);
264 final String currentOwner = currentOwners.get(entity);
266 if (currentOwner == null || !activeMembers.contains(currentOwner)) {
267 // might as well assign right away when we don't have an owner or its unreachable
268 assignOwnerFor(entity);
271 LOG.debug("Current state for entity: {} : {}", entity, currentCandidates.get(entity).toString());
275 if (!candidates.contains(toCheck)) {
277 LOG.debug("Removing candidate from entity: {} - {}", entity, toCheck);
278 currentCandidates.get(entity).remove(toCheck);
279 if (ownerToEntity.containsKey(toCheck)) {
280 ownersToReassign.add(toCheck);
285 // then reassign those that need new owners
286 for (final String toReassign : ownersToReassign) {
287 reassignCandidatesFor(toReassign, ImmutableList.copyOf(ownerToEntity.get(toReassign)),
291 if (currentCandidates.get(entity) == null) {
292 LOG.debug("Last candidate removed for {}", entity);
294 LOG.debug("Current state for entity: {} : {}", entity, currentCandidates.get(entity).toString());
298 private void reassignCandidatesFor(final String oldOwner, final Collection<DOMEntity> entities,
299 final BiPredicate<DOMEntity, String> predicate) {
300 LOG.debug("Reassigning owners for {}", entities);
301 for (final DOMEntity entity : entities) {
302 if (predicate.test(entity, oldOwner)) {
304 if (!isActiveCandidate(oldOwner) && isCandidateFor(entity, oldOwner) && hasSingleCandidate(entity)) {
305 // only skip new owner assignment, only if unreachable, still is a candidate and is the ONLY
307 LOG.debug("{} is the only candidate for {}. Skipping reassignment.", oldOwner, entity);
310 ownerToEntity.remove(oldOwner, entity);
311 assignOwnerFor(entity);
316 private boolean isActiveCandidate(final String candidate) {
317 return activeMembers.contains(candidate);
320 private boolean isCandidateFor(final DOMEntity entity, final String candidate) {
321 return currentCandidates.getOrDefault(entity, Set.of()).contains(candidate);
324 private boolean hasSingleCandidate(final DOMEntity entity) {
325 return currentCandidates.getOrDefault(entity, Set.of()).size() == 1;
328 private void assignOwnerFor(final DOMEntity entity) {
329 final Set<String> candidatesForEntity = currentCandidates.get(entity);
330 if (candidatesForEntity.isEmpty()) {
331 LOG.debug("No candidates present for entity: {}", entity);
336 String pickedCandidate = null;
337 for (final String candidate : candidatesForEntity) {
338 if (activeMembers.contains(candidate)) {
339 pickedCandidate = candidate;
343 if (pickedCandidate == null) {
344 LOG.debug("No candidate is reachable for {}, activeMembers: {}, currentCandidates: {}",
345 entity, activeMembers, currentCandidates.get(entity));
346 // no candidate is reachable so only remove owner if necessary
350 ownerToEntity.put(pickedCandidate, entity);
352 LOG.debug("Entity {} new owner: {}", entity, pickedCandidate);
353 currentOwners.put(entity, pickedCandidate);
354 writeNewOwner(entity, pickedCandidate);
357 private void removeOwner(final DOMEntity entity) {
358 if (currentOwners.containsKey(entity)) {
359 // assign empty owner to dd, as we cannot delete data for a key since that would prevent
360 // writes for the same key
361 currentOwners.remove(entity);
363 writeNewOwner(entity, "");
367 private void writeNewOwner(final DOMEntity entity, final String candidate) {
368 ownerReplicator.askUpdate(
369 askReplyTo -> new Replicator.Update<>(
370 new LWWRegisterKey<>(entity.toString()),
371 new LWWRegister<>(node.uniqueAddress(), candidate, 0),
372 Replicator.writeLocal(),
374 register -> register.withValue(node, candidate, CLOCK)),
378 private Behavior<OwnerSupervisorCommand> onPeerUp(final MemberUpEvent event) {
379 LOG.debug("Received MemberUp : {}", event);
381 handleReachableEvent(event.getRoles());
385 private Behavior<OwnerSupervisorCommand> onPeerReachable(final MemberReachableEvent event) {
386 LOG.debug("Received MemberReachable : {}", event);
388 handleReachableEvent(event.getRoles());
392 private Behavior<OwnerSupervisorCommand> onGetEntities(final GetEntitiesBackendRequest request) {
393 request.getReplyTo().tell(StatusReply.success(new GetEntitiesBackendReply(currentOwners, currentCandidates)));
397 private Behavior<OwnerSupervisorCommand> onGetEntity(final GetEntityBackendRequest request) {
398 final DOMEntity entity = extractEntity(request);
399 request.getReplyTo().tell(StatusReply.success(
400 new GetEntityBackendReply(currentOwners.get(entity), currentCandidates.get(entity))));
404 private Behavior<OwnerSupervisorCommand> onGetEntityOwner(final GetEntityOwnerBackendRequest request) {
405 request.getReplyTo().tell(
406 StatusReply.success(new GetEntityOwnerBackendReply(currentOwners.get(extractEntity(request)))));
410 private void handleReachableEvent(final Set<String> roles) {
411 if (roles.contains(dataCenter)) {
412 activeMembers.add(extractRole(roles));
413 assignMissingOwners();
415 LOG.debug("Received reachable event from a foreign datacenter, Ignoring... Roles: {}", roles);
419 private Behavior<OwnerSupervisorCommand> onPeerDown(final MemberDownEvent event) {
420 LOG.debug("Received MemberDown : {}", event);
422 handleUnreachableEvent(event.getRoles());
426 private Behavior<OwnerSupervisorCommand> onPeerUnreachable(final MemberUnreachableEvent event) {
427 LOG.debug("Received MemberUnreachable : {}", event);
429 handleUnreachableEvent(event.getRoles());
433 private void handleUnreachableEvent(final Set<String> roles) {
434 if (roles.contains(dataCenter)) {
435 activeMembers.remove(extractRole(roles));
436 reassignUnreachableOwners();
438 LOG.debug("Received unreachable event from a foreign datacenter, Ignoring... Roles: {}", roles);
442 private Set<String> getActiveMembers() {
443 final CurrentClusterState clusterState = cluster.state();
444 final Set<String> unreachableRoles = clusterState.getUnreachable().stream()
445 .map(OwnerSupervisor::extractRole)
446 .collect(Collectors.toSet());
448 return StreamSupport.stream(clusterState.getMembers().spliterator(), false)
449 // We are evaluating the set of roles for each member
450 .map(Member::getRoles)
451 // Filter out any members which do not share our dataCenter
452 .filter(roles -> roles.contains(dataCenter))
453 // Find first legal role
454 .map(OwnerSupervisor::extractRole)
455 // filter out unreachable roles
456 .filter(role -> !unreachableRoles.contains(role))
457 .collect(Collectors.toSet());
460 private DOMEntity extractEntity(final AbstractEntityRequest<?> request) {
461 final var name = request.getName();
462 final var iid = name.getInstanceIdentifier();
464 return new DOMEntity(request.getType().getValue(), iidCodec.fromBinding(iid));
466 final var str = verifyNotNull(name.getString(), "Unhandled entity name %s", name);
467 return new DOMEntity(request.getType().getValue(), str);
470 private static String extractRole(final Member member) {
471 return extractRole(member.getRoles());
474 private static String extractRole(final Set<String> roles) {
475 return roles.stream().filter(role -> !role.startsWith(DATACENTER_PREFIX))
476 .findFirst().orElseThrow(() -> new IllegalArgumentException("No valid role found."));
479 private static String extractDatacenterRole(final Member member) {
480 return member.getRoles().stream().filter(role -> role.startsWith(DATACENTER_PREFIX))
481 .findFirst().orElseThrow(() -> new IllegalArgumentException("No valid role found."));