import akka.actor.ActorRef;
import akka.util.Timeout;
import com.google.common.base.Preconditions;
-import com.google.common.base.Throwables;
import com.google.common.primitives.UnsignedLong;
+import java.util.Set;
+import java.util.concurrent.CompletableFuture;
import java.util.concurrent.CompletionStage;
+import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
import java.util.concurrent.atomic.AtomicLong;
+import java.util.function.Consumer;
import javax.annotation.Nonnull;
import javax.annotation.Nullable;
import javax.annotation.concurrent.GuardedBy;
import org.opendaylight.controller.cluster.access.client.BackendInfoResolver;
import org.opendaylight.controller.cluster.access.commands.ConnectClientRequest;
import org.opendaylight.controller.cluster.access.commands.ConnectClientSuccess;
+import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
import org.opendaylight.controller.cluster.access.concepts.RequestFailure;
import org.opendaylight.controller.cluster.common.actor.ExplicitAsk;
+import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
+import org.opendaylight.controller.cluster.datastore.exceptions.NotInitializedException;
+import org.opendaylight.controller.cluster.datastore.exceptions.PrimaryNotFoundException;
+import org.opendaylight.controller.cluster.datastore.messages.PrimaryShardInfo;
import org.opendaylight.controller.cluster.datastore.utils.ActorContext;
+import org.opendaylight.yangtools.concepts.Registration;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import scala.Function1;
return result;
}
- private synchronized void onStageResolved(final ShardBackendInfo result, final Throwable failure) {
+ private synchronized void onStageResolved(final ShardBackendInfo info, final Throwable failure) {
if (failure == null) {
- this.result = Preconditions.checkNotNull(result);
+ this.result = Preconditions.checkNotNull(info);
} else {
LOG.warn("Failed to resolve shard", failure);
}
private static final Logger LOG = LoggerFactory.getLogger(AbstractShardBackendResolver.class);
/**
- * Fall-over-dead timeout. If we do not make progress in this long, just fall over and propagate the failure.
- * All users are expected to fail, possibly attempting to recover by restarting. It is fair to remain
- * non-operational.
+ * Connect request timeout. If the shard does not respond within this interval, we retry the lookup and connection.
*/
// TODO: maybe make this configurable somehow?
- private static final Timeout DEAD_TIMEOUT = Timeout.apply(15, TimeUnit.MINUTES);
+ private static final Timeout CONNECT_TIMEOUT = Timeout.apply(5, TimeUnit.SECONDS);
private final AtomicLong nextSessionId = new AtomicLong();
private final Function1<ActorRef, ?> connectFunction;
private final ActorContext actorContext;
+ private final Set<Consumer<Long>> staleBackendInfoCallbacks = ConcurrentHashMap.newKeySet();
// FIXME: we really need just ActorContext.findPrimaryShardAsync()
AbstractShardBackendResolver(final ClientIdentifier clientId, final ActorContext actorContext) {
ABIVersion.current()));
}
+ @Override
+ public Registration notifyWhenBackendInfoIsStale(final Consumer<Long> callback) {
+ staleBackendInfoCallbacks.add(callback);
+ return () -> staleBackendInfoCallbacks.remove(callback);
+ }
+
+ protected void notifyStaleBackendInfoCallbacks(Long cookie) {
+ staleBackendInfoCallbacks.forEach(callback -> callback.accept(cookie));
+ }
+
+ protected ActorContext actorContext() {
+ return actorContext;
+ }
+
protected final void flushCache(final String shardName) {
actorContext.getPrimaryShardInfoCache().remove(shardName);
}
protected final ShardState resolveBackendInfo(final String shardName, final long cookie) {
LOG.debug("Resolving cookie {} to shard {}", cookie, shardName);
- return new ShardState(FutureConverters.toJava(actorContext.findPrimaryShardAsync(shardName)).thenCompose(i -> {
- LOG.debug("Looking up primary info for {} from {}", shardName, i);
- return FutureConverters.toJava(ExplicitAsk.ask(i.getPrimaryShardActor(), connectFunction, DEAD_TIMEOUT));
- }).thenApply(response -> {
- if (response instanceof RequestFailure) {
- final RequestFailure<?, ?> failure = (RequestFailure<?, ?>) response;
- LOG.debug("Connect request failed {}", failure, failure.getCause());
- throw Throwables.propagate(failure.getCause());
+ final CompletableFuture<ShardBackendInfo> future = new CompletableFuture<>();
+ FutureConverters.toJava(actorContext.findPrimaryShardAsync(shardName)).whenComplete((info, failure) -> {
+ if (failure == null) {
+ connectShard(shardName, cookie, info, future);
+ return;
+ }
+
+ LOG.debug("Shard {} failed to resolve", shardName, failure);
+ if (failure instanceof NoShardLeaderException) {
+ future.completeExceptionally(wrap("Shard has no current leader", failure));
+ } else if (failure instanceof NotInitializedException) {
+ // FIXME: this actually is an exception we can retry on
+ LOG.info("Shard {} has not initialized yet", shardName);
+ future.completeExceptionally(failure);
+ } else if (failure instanceof PrimaryNotFoundException) {
+ LOG.info("Failed to find primary for shard {}", shardName);
+ future.completeExceptionally(failure);
+ } else {
+ future.completeExceptionally(failure);
}
+ });
+
+ return new ShardState(future);
+ }
+
+ private static TimeoutException wrap(final String message, final Throwable cause) {
+ final TimeoutException ret = new TimeoutException(message);
+ ret.initCause(Preconditions.checkNotNull(cause));
+ return ret;
+ }
- LOG.debug("Resolved backend information to {}", response);
+ private void connectShard(final String shardName, final long cookie, final PrimaryShardInfo info,
+ final CompletableFuture<ShardBackendInfo> future) {
+ LOG.debug("Shard {} resolved to {}, attempting to connect", shardName, info);
- Preconditions.checkArgument(response instanceof ConnectClientSuccess, "Unhandled response {}", response);
- final ConnectClientSuccess success = (ConnectClientSuccess) response;
+ FutureConverters.toJava(ExplicitAsk.ask(info.getPrimaryShardActor(), connectFunction, CONNECT_TIMEOUT))
+ .whenComplete((response, failure) -> onConnectResponse(shardName, cookie, future, response, failure));
+ }
+
+ private void onConnectResponse(final String shardName, final long cookie,
+ final CompletableFuture<ShardBackendInfo> future, final Object response, final Throwable failure) {
+ if (failure != null) {
+ LOG.debug("Connect attempt to {} failed, will retry", shardName, failure);
+ future.completeExceptionally(wrap("Connection attempt failed", failure));
+ return;
+ }
+ if (response instanceof RequestFailure) {
+ final Throwable cause = ((RequestFailure<?, ?>) response).getCause().unwrap();
+ LOG.debug("Connect attempt to {} failed to process", shardName, cause);
+ final Throwable result = cause instanceof NotLeaderException
+ ? wrap("Leader moved during establishment", cause) : cause;
+ future.completeExceptionally(result);
+ return;
+ }
- return new ShardBackendInfo(success.getBackend(),
- nextSessionId.getAndIncrement(), success.getVersion(), shardName, UnsignedLong.fromLongBits(cookie),
- success.getDataTree(), success.getMaxMessages());
- }));
+ LOG.debug("Resolved backend information to {}", response);
+ Preconditions.checkArgument(response instanceof ConnectClientSuccess, "Unhandled response %s",
+ response);
+ final ConnectClientSuccess success = (ConnectClientSuccess) response;
+ future.complete(new ShardBackendInfo(success.getBackend(), nextSessionId.getAndIncrement(),
+ success.getVersion(), shardName, UnsignedLong.fromLongBits(cookie), success.getDataTree(),
+ success.getMaxMessages()));
}
}
Code Review / controller.git / blobdiff