import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import scala.compat.java8.FutureConverters;
-import scala.concurrent.Await;
import scala.concurrent.Future;
import scala.concurrent.Promise;
import scala.concurrent.duration.FiniteDuration;
private final DOMDataTreePrefixTable<DOMDataTreeShardRegistration<DOMDataTreeShard>> shards =
DOMDataTreePrefixTable.create();
- private final EnumMap<LogicalDatastoreType, DistributedShardRegistration> defaultShardRegistrations =
- new EnumMap<>(LogicalDatastoreType.class);
-
private final EnumMap<LogicalDatastoreType, Entry<DataStoreClient, ActorRef>> configurationShardMap =
new EnumMap<>(LogicalDatastoreType.class);
public void init() {
// create our writers to the configuration
try {
- LOG.debug("{} - starting config shard lookup.",
- distributedConfigDatastore.getActorContext().getCurrentMemberName());
+ LOG.debug("{} - starting config shard lookup.", memberName);
// We have to wait for prefix config shards to be up and running
// so we can create datastore clients for them
handleConfigShardLookup().get(SHARD_FUTURE_TIMEOUT_DURATION.length(), SHARD_FUTURE_TIMEOUT_DURATION.unit());
-
- LOG.debug("Prefix configuration shards ready - creating clients");
-
} catch (InterruptedException | ExecutionException | TimeoutException e) {
throw new IllegalStateException("Prefix config shards not found", e);
}
try {
- LOG.debug("Prefix configuration shards ready - creating clients");
+ LOG.debug("{}: Prefix configuration shards ready - creating clients", memberName);
configurationShardMap.put(LogicalDatastoreType.CONFIGURATION,
createDatastoreClient(ClusterUtils.PREFIX_CONFIG_SHARD_ID,
distributedConfigDatastore.getActorContext()));
//create shard registration for DEFAULT_SHARD
try {
- defaultShardRegistrations.put(LogicalDatastoreType.CONFIGURATION,
- initDefaultShard(LogicalDatastoreType.CONFIGURATION));
+ initDefaultShard(LogicalDatastoreType.CONFIGURATION);
} catch (final InterruptedException | ExecutionException e) {
throw new IllegalStateException("Unable to create default shard frontend for config shard", e);
}
try {
- defaultShardRegistrations.put(LogicalDatastoreType.OPERATIONAL,
- initDefaultShard(LogicalDatastoreType.OPERATIONAL));
+ initDefaultShard(LogicalDatastoreType.OPERATIONAL);
} catch (final InterruptedException | ExecutionException e) {
throw new IllegalStateException("Unable to create default shard frontend for operational shard", e);
}
@Nonnull
@Override
public DOMDataTreeProducer createProducer(@Nonnull final Collection<DOMDataTreeIdentifier> subtrees) {
- LOG.debug("{} - Creating producer for {}",
- distributedConfigDatastore.getActorContext().getClusterWrapper().getCurrentMemberName(), subtrees);
+ LOG.debug("{} - Creating producer for {}", memberName, subtrees);
final DOMDataTreeProducer producer = shardedDOMDataTree.createProducer(subtrees);
final Object response = distributedConfigDatastore.getActorContext()
.executeOperation(shardedDataTreeActor, new ProducerCreated(subtrees));
if (response == null) {
- LOG.debug("{} - Received success from remote nodes, creating producer:{}",
- distributedConfigDatastore.getActorContext().getClusterWrapper().getCurrentMemberName(), subtrees);
+ LOG.debug("{} - Received success from remote nodes, creating producer:{}", memberName, subtrees);
return new ProxyProducer(producer, subtrees, shardedDataTreeActor,
distributedConfigDatastore.getActorContext(), shards);
}
}
void resolveShardAdditions(final Set<DOMDataTreeIdentifier> additions) {
- LOG.debug("Member {}: Resolving additions : {}", memberName, additions);
+ LOG.debug("{}: Resolving additions : {}", memberName, additions);
final ArrayList<DOMDataTreeIdentifier> list = new ArrayList<>(additions);
// we need to register the shards from top to bottom, so we need to atleast make sure the ordering reflects that
Collections.sort(list, (o1, o2) -> {
}
void resolveShardRemovals(final Set<DOMDataTreeIdentifier> removals) {
- LOG.debug("Member {}: Resolving removals : {}", memberName, removals);
+ LOG.debug("{}: Resolving removals : {}", memberName, removals);
// do we need to go from bottom to top?
removals.forEach(this::despawnShardFrontend);
}
private void createShardFrontend(final DOMDataTreeIdentifier prefix) {
- LOG.debug("Member {}: Creating CDS shard for prefix: {}", memberName, prefix);
+ LOG.debug("{}: Creating CDS shard for prefix: {}", memberName, prefix);
final String shardName = ClusterUtils.getCleanShardName(prefix.getRootIdentifier());
final AbstractDataStore distributedDataStore =
prefix.getDatastoreType().equals(org.opendaylight.mdsal.common.api.LogicalDatastoreType.CONFIGURATION)
}
private void despawnShardFrontend(final DOMDataTreeIdentifier prefix) {
- LOG.debug("Member {}: Removing CDS shard for prefix: {}", memberName, prefix);
+ LOG.debug("{}: Removing CDS shard for prefix: {}", memberName, prefix);
final DOMDataTreePrefixTableEntry<DOMDataTreeShardRegistration<DOMDataTreeShard>> lookup;
synchronized (shards) {
lookup = shards.lookup(prefix);
}
if (lookup == null || !lookup.getValue().getPrefix().equals(prefix)) {
- LOG.debug("Member {}: Received despawn for non-existing CDS shard frontend, prefix: {}, ignoring..",
+ LOG.debug("{}: Received despawn for non-existing CDS shard frontend, prefix: {}, ignoring..",
memberName, prefix);
return;
}
final String shardName, final ActorContext actorContext)
throws DOMDataTreeShardCreationFailedException {
- LOG.debug("Creating distributed datastore client for shard {}", shardName);
+ LOG.debug("{}: Creating distributed datastore client for shard {}", memberName, shardName);
final Props distributedDataStoreClientProps =
SimpleDataStoreClientActor.props(memberName, "Shard-" + shardName, actorContext, shardName);
return new SimpleEntry<>(SimpleDataStoreClientActor
.getDistributedDataStoreClient(clientActor, 30, TimeUnit.SECONDS), clientActor);
} catch (final Exception e) {
- LOG.error("Failed to get actor for {}", distributedDataStoreClientProps, e);
+ LOG.error("{}: Failed to get actor for {}", distributedDataStoreClientProps, memberName, e);
clientActor.tell(PoisonPill.getInstance(), noSender());
throw new DOMDataTreeShardCreationFailedException(
"Unable to create datastore client for shard{" + shardName + "}", e);
}
@SuppressWarnings("checkstyle:IllegalCatch")
- private DistributedShardRegistration initDefaultShard(final LogicalDatastoreType logicalDatastoreType)
+ private void initDefaultShard(final LogicalDatastoreType logicalDatastoreType)
throws ExecutionException, InterruptedException {
- final Collection<MemberName> names =
- distributedConfigDatastore.getActorContext().getConfiguration().getUniqueMemberNamesForAllShards();
final PrefixedShardConfigWriter writer = writerMap.get(logicalDatastoreType);
if (writer.checkDefaultIsPresent()) {
- LOG.debug("Default shard for {} is already present in the config. Possibly saved in snapshot.",
- logicalDatastoreType);
- return new DistributedShardRegistrationImpl(
- new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY),
- shardedDataTreeActor, this);
+ LOG.debug("{}: Default shard for {} is already present in the config. Possibly saved in snapshot.",
+ memberName, logicalDatastoreType);
} else {
try {
- // There can be situation when there is already started default shard
- // because it is present in modules.conf. In that case we have to create
- // just frontend for default shard, but not shard itself
- // TODO we don't have to do it for config and operational default shard
- // separately. Just one of them should be enough
+ // Currently the default shard configuration is present in the out-of-box modules.conf and is
+ // expected to be present. So look up the local default shard here and create the frontend.
+
+ // TODO we don't have to do it for config and operational default shard separately. Just one of them
+ // should be enough
final ActorContext actorContext = logicalDatastoreType == LogicalDatastoreType.CONFIGURATION
? distributedConfigDatastore.getActorContext() : distributedOperDatastore.getActorContext();
actorContext.findLocalShard(ClusterUtils.getCleanShardName(YangInstanceIdentifier.EMPTY));
if (defaultLocalShardOptional.isPresent()) {
- LOG.debug("{} Default shard is already started, creating just frontend", logicalDatastoreType);
+ LOG.debug("{}: Default shard for {} is already started, creating just frontend", memberName,
+ logicalDatastoreType);
createShardFrontend(new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY));
- return new DistributedShardRegistrationImpl(
- new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY),
- shardedDataTreeActor, this);
}
- // we should probably only have one node create the default shards
- return Await.result(FutureConverters.toScala(createDistributedShard(
- new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY), names)),
- SHARD_FUTURE_TIMEOUT_DURATION);
- } catch (DOMDataTreeShardingConflictException e) {
- LOG.debug("Default shard already registered, possibly due to other node doing it faster");
- return new DistributedShardRegistrationImpl(
- new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY),
- shardedDataTreeActor, this);
+ // The local shard isn't present - we assume that means the local member isn't in the replica list
+ // and will be dynamically created later via an explicit add-shard-replica request. This is the
+ // bootstrapping mechanism to add a new node into an existing cluster. The following code to create
+ // the default shard as a prefix shard is problematic in this scenario so it is commented out. Since
+ // the default shard is a module-based shard by default, it makes sense to always treat it as such,
+ // ie bootstrap it in the same manner as the special prefix-configuration and EOS shards.
+// final Collection<MemberName> names = distributedConfigDatastore.getActorContext().getConfiguration()
+// .getUniqueMemberNamesForAllShards();
+// Await.result(FutureConverters.toScala(createDistributedShard(
+// new DOMDataTreeIdentifier(logicalDatastoreType, YangInstanceIdentifier.EMPTY), names)),
+// SHARD_FUTURE_TIMEOUT_DURATION);
+// } catch (DOMDataTreeShardingConflictException e) {
+// LOG.debug("{}: Default shard for {} already registered, possibly due to other node doing it faster",
+// memberName, logicalDatastoreType);
} catch (Exception e) {
- LOG.error("{} default shard initialization failed", logicalDatastoreType, e);
+ LOG.error("{}: Default shard initialization for {} failed", memberName, logicalDatastoreType, e);
throw new RuntimeException(e);
}
}
private final Cluster cluster;
- private Map<DOMDataTreeIdentifier, PrefixShardConfiguration> currentConfiguration = new HashMap<>();
+ private final Map<DOMDataTreeIdentifier, PrefixShardConfiguration> currentConfiguration = new HashMap<>();
ShardedDataTreeActor(final ShardedDataTreeActorCreator builder) {
LOG.debug("Creating ShardedDataTreeActor on {}", builder.getClusterWrapper().getCurrentMemberName());
// schedule a notification task for the reply
actorSystem.scheduler().scheduleOnce(SHARD_LOOKUP_TASK_INTERVAL,
new ConfigShardLookupTask(
- actorSystem, getSender(), context, clusterWrapper, message, lookupTaskMaxRetries),
+ actorSystem, getSender(), context, message, lookupTaskMaxRetries),
actorSystem.dispatcher());
}
private final ActorSystem system;
private final ActorRef replyTo;
private final ActorContext context;
- private final ClusterWrapper clusterWrapper;
- private final int lookupTaskMaxRetries;
ConfigShardLookupTask(final ActorSystem system,
final ActorRef replyTo,
final ActorContext context,
- final ClusterWrapper clusterWrapper,
final StartConfigShardLookup message,
final int lookupMaxRetries) {
super(replyTo, lookupMaxRetries);
this.system = system;
this.replyTo = replyTo;
this.context = context;
- this.clusterWrapper = clusterWrapper;
- this.lookupTaskMaxRetries = lookupMaxRetries;
}
@Override
if (!localShard.isPresent()) {
tryReschedule(null);
} else {
- LOG.debug("Local backend for prefix configuration shard lookup successful, starting leader lookup..");
- system.scheduler().scheduleOnce(
- SHARD_LOOKUP_TASK_INTERVAL,
- new ConfigShardReadinessTask(
- system, replyTo, context, clusterWrapper, localShard.get(), lookupTaskMaxRetries),
- system.dispatcher());
+ LOG.debug("Local backend for prefix configuration shard lookup successful");
+ replyTo.tell(new Status.Success(null), noSender());
}
}
}
}
private void initEmptyDatastores() throws Exception {
+ initEmptyDatastores(MODULE_SHARDS_CONFIG);
+ }
+
+ private void initEmptyDatastores(String moduleShardsConfig) throws Exception {
leaderTestKit = new IntegrationTestKit(leaderSystem, leaderDatastoreContextBuilder);
leaderConfigDatastore = leaderTestKit.setupDistributedDataStore(
- "config", MODULE_SHARDS_CONFIG, true,
+ "config", moduleShardsConfig, true,
SchemaContextHelper.distributedShardedDOMDataTreeSchemaContext());
leaderOperDatastore = leaderTestKit.setupDistributedDataStore(
- "operational", MODULE_SHARDS_CONFIG, true,
+ "operational", moduleShardsConfig, true,
SchemaContextHelper.distributedShardedDOMDataTreeSchemaContext());
leaderShardFactory = new DistributedShardedDOMDataTree(leaderSystemProvider,
followerTestKit = new IntegrationTestKit(followerSystem, followerDatastoreContextBuilder);
followerConfigDatastore = followerTestKit.setupDistributedDataStore(
- "config", MODULE_SHARDS_CONFIG, true, SchemaContextHelper.distributedShardedDOMDataTreeSchemaContext());
+ "config", moduleShardsConfig, true, SchemaContextHelper.distributedShardedDOMDataTreeSchemaContext());
followerOperDatastore = followerTestKit.setupDistributedDataStore(
- "operational", MODULE_SHARDS_CONFIG, true,
+ "operational", moduleShardsConfig, true,
SchemaContextHelper.distributedShardedDOMDataTreeSchemaContext());
followerShardFactory = new DistributedShardedDOMDataTree(followerSystemProvider,
followerTestKit.waitForMembersUp("member-1");
+ LOG.info("Initializing leader DistributedShardedDOMDataTree");
leaderShardFactory.init();
- followerShardFactory.init();
leaderTestKit.waitUntilLeader(leaderConfigDatastore.getActorContext(),
ClusterUtils.getCleanShardName(YangInstanceIdentifier.EMPTY));
leaderTestKit.waitUntilLeader(leaderOperDatastore.getActorContext(),
ClusterUtils.getCleanShardName(YangInstanceIdentifier.EMPTY));
+
+ LOG.info("Initializing follower DistributedShardedDOMDataTree");
+ followerShardFactory.init();
}
@Test
LOG.info("testMultipleRegistrationsAtOnePrefix ending");
}
+
+ @Test
+ public void testInitialBootstrappingWithNoModuleShards() throws Exception {
+ LOG.info("testInitialBootstrappingWithNoModuleShards starting");
+ initEmptyDatastores("module-shards-default-member-1.conf");
+
+ // We just verify the DistributedShardedDOMDataTree initialized without error.
+ }
}