import com.google.common.annotations.Beta;
import com.google.common.base.Preconditions;
import com.google.common.base.Verify;
+import java.util.Collection;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
import javax.annotation.Nonnull;
import javax.annotation.Nullable;
import javax.annotation.concurrent.GuardedBy;
+import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
+import org.opendaylight.controller.cluster.access.commands.OutOfSequenceEnvelopeException;
import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
import org.opendaylight.controller.cluster.access.concepts.FailureEnvelope;
import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
import org.opendaylight.yangtools.concepts.WritableIdentifier;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
+import scala.concurrent.duration.FiniteDuration;
/**
* A behavior, which handles messages sent to a {@link AbstractClientActor}.
* @param enqueuedEntries Previously-enqueued entries
* @return A {@link ReconnectForwarder} to handle any straggler messages which arrive after this method returns.
*/
- @Nonnull ReconnectForwarder finishReconnect(@Nonnull Iterable<ConnectionEntry> enqueuedEntries);
+ @Nonnull ReconnectForwarder finishReconnect(@Nonnull Collection<ConnectionEntry> enqueuedEntries);
}
private static final Logger LOG = LoggerFactory.getLogger(ClientActorBehavior.class);
+ private static final FiniteDuration RESOLVE_RETRY_DURATION = FiniteDuration.apply(5, TimeUnit.SECONDS);
/**
* Map of connections to the backend. This map is concurrent to allow lookups, but given complex operations
}
}
+ private AbstractClientConnection<T> getConnection(final ResponseEnvelope<?> response) {
+ // Always called from actor context: no locking required
+ return connections.get(extractCookie(response.getMessage().getTarget()));
+ }
+
@SuppressWarnings("unchecked")
@Override
final ClientActorBehavior<T> onReceiveCommand(final Object command) {
}
private void onResponse(final ResponseEnvelope<?> response) {
- final long cookie = extractCookie(response.getMessage().getTarget());
- final AbstractClientConnection<T> connection = connections.get(cookie);
+ final AbstractClientConnection<T> connection = getConnection(response);
if (connection != null) {
connection.receiveResponse(response);
} else {
poison(cause);
return null;
}
+ if (cause instanceof NotLeaderException) {
+ final AbstractClientConnection<T> conn = getConnection(command);
+ if (conn instanceof ReconnectingClientConnection) {
+ // Already reconnecting, do not churn the logs
+ return this;
+ } else if (conn != null) {
+ LOG.info("{}: connection {} indicated no leadership, reconnecting it", persistenceId(), conn, cause);
+ return conn.reconnect(this, cause);
+ }
+ }
+ if (cause instanceof OutOfSequenceEnvelopeException) {
+ final AbstractClientConnection<T> conn = getConnection(command);
+ if (conn instanceof ReconnectingClientConnection) {
+ // Already reconnecting, do not churn the logs
+ return this;
+ } else if (conn != null) {
+ LOG.info("{}: connection {} indicated no sequencing mismatch on {} sequence {}, reconnecting it",
+ persistenceId(), conn, failure.getTarget(), failure.getSequence(), cause);
+ return conn.reconnect(this, cause);
+ }
+ }
return onRequestFailure(command);
}
private void backendConnectFinished(final Long shard, final AbstractClientConnection<T> conn,
final T backend, final Throwable failure) {
if (failure != null) {
+ if (failure instanceof TimeoutException) {
+ if (!conn.equals(connections.get(shard))) {
+ // AbstractClientConnection will remove itself when it decides there is no point in continuing,
+ // at which point we want to stop retrying
+ LOG.info("{}: stopping resolution of shard {} on stale connection {}", persistenceId(), shard, conn,
+ failure);
+ return;
+ }
+
+ LOG.debug("{}: timed out resolving shard {}, scheduling retry in {}", persistenceId(), shard,
+ RESOLVE_RETRY_DURATION, failure);
+ context().executeInActor(b -> {
+ resolveConnection(shard, conn);
+ return b;
+ }, RESOLVE_RETRY_DURATION);
+ return;
+ }
+
LOG.error("{}: failed to resolve shard {}", persistenceId(), shard, failure);
- conn.poison(new RuntimeRequestException("Failed to resolve shard " + shard, failure));
+ final RequestException cause;
+ if (failure instanceof RequestException) {
+ cause = (RequestException) failure;
+ } else {
+ cause = new RuntimeRequestException("Failed to resolve shard " + shard, failure);
+ }
+
+ conn.poison(cause);
return;
}
- LOG.debug("{}: resolved shard {} to {}", persistenceId(), shard, backend);
+ LOG.info("{}: resolved shard {} to {}", persistenceId(), shard, backend);
final long stamp = connectionsLock.writeLock();
try {
// Create a new connected connection
final ConnectedClientConnection<T> newConn = new ConnectedClientConnection<>(conn.context(),
conn.cookie(), backend);
- LOG.debug("{}: resolving connection {} to {}", persistenceId(), conn, newConn);
+ LOG.info("{}: resolving connection {} to {}", persistenceId(), conn, newConn);
// Start reconnecting without the old connection lock held
final ConnectionConnectCohort cohort = Verify.verifyNotNull(connectionUp(newConn));
// Lock the old connection and get a reference to its entries
- final Iterable<ConnectionEntry> replayIterable = conn.startReplay();
+ final Collection<ConnectionEntry> replayIterable = conn.startReplay();
// Finish the connection attempt
final ReconnectForwarder forwarder = Verify.verifyNotNull(cohort.finishReconnect(replayIterable));
conn.finishReplay(forwarder);
// Make sure new lookups pick up the new connection
- connections.replace(shard, conn, newConn);
- LOG.debug("{}: replaced connection {} with {}", persistenceId(), conn, newConn);
+ if (!connections.replace(shard, conn, newConn)) {
+ final AbstractClientConnection<T> existing = connections.get(conn.cookie());
+ LOG.warn("{}: old connection {} does not match existing {}, new connection {} in limbo",
+ persistenceId(), conn, existing, newConn);
+ } else {
+ LOG.info("{}: replaced connection {} with {}", persistenceId(), conn, newConn);
+ }
} finally {
connectionsLock.unlockWrite(stamp);
}
}
void removeConnection(final AbstractClientConnection<?> conn) {
- connections.remove(conn.cookie(), conn);
- LOG.debug("{}: removed connection {}", persistenceId(), conn);
+ final long stamp = connectionsLock.writeLock();
+ try {
+ if (!connections.remove(conn.cookie(), conn)) {
+ final AbstractClientConnection<T> existing = connections.get(conn.cookie());
+ if (existing != null) {
+ LOG.warn("{}: failed to remove connection {}, as it was superseded by {}", persistenceId(), conn,
+ existing);
+ } else {
+ LOG.warn("{}: failed to remove connection {}, as it was not tracked", persistenceId(), conn);
+ }
+ } else {
+ LOG.info("{}: removed connection {}", persistenceId(), conn);
+ }
+ } finally {
+ connectionsLock.unlockWrite(stamp);
+ }
}
@SuppressWarnings("unchecked")
void reconnectConnection(final ConnectedClientConnection<?> oldConn,
final ReconnectingClientConnection<?> newConn) {
final ReconnectingClientConnection<T> conn = (ReconnectingClientConnection<T>)newConn;
- connections.replace(oldConn.cookie(), (AbstractClientConnection<T>)oldConn, conn);
- LOG.debug("{}: connection {} reconnecting as {}", persistenceId(), oldConn, newConn);
+ LOG.info("{}: connection {} reconnecting as {}", persistenceId(), oldConn, newConn);
+
+ final long stamp = connectionsLock.writeLock();
+ try {
+ final boolean replaced = connections.replace(oldConn.cookie(), (AbstractClientConnection<T>)oldConn, conn);
+ if (!replaced) {
+ final AbstractClientConnection<T> existing = connections.get(oldConn.cookie());
+ if (existing != null) {
+ LOG.warn("{}: failed to replace connection {}, as it was superseded by {}", persistenceId(), conn,
+ existing);
+ } else {
+ LOG.warn("{}: failed to replace connection {}, as it was not tracked", persistenceId(), conn);
+ }
+ }
+ } finally {
+ connectionsLock.unlockWrite(stamp);
+ }
final Long shard = oldConn.cookie();
+ LOG.info("{}: refreshing backend for shard {}", persistenceId(), shard);
resolver().refreshBackendInfo(shard, conn.getBackendInfo().get()).whenComplete(
(backend, failure) -> context().executeInActor(behavior -> {
backendConnectFinished(shard, conn, backend, failure);
private ConnectingClientConnection<T> createConnection(final Long shard) {
final ConnectingClientConnection<T> conn = new ConnectingClientConnection<>(context(), shard);
+ resolveConnection(shard, conn);
+ return conn;
+ }
+ private void resolveConnection(final Long shard, final AbstractClientConnection<T> conn) {
+ LOG.debug("{}: resolving shard {} connection {}", persistenceId(), shard, conn);
resolver().getBackendInfo(shard).whenComplete((backend, failure) -> context().executeInActor(behavior -> {
backendConnectFinished(shard, conn, backend, failure);
return behavior;
}));
-
- return conn;
}
}
Code Review / controller.git / blobdiff