X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fcds-access-client%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Faccess%2Fclient%2FAbstractClientConnection.java;h=e0739dd6c5a36cdff73e22ff64cb261baf8f1373;hb=a8fa0eeb173402295e92a986022a61d9cad090d0;hp=3cc8d7073ae3467d04fed6468f1af943b7c15b0f;hpb=90c86de54d7bf10bfb9ffb0a8ad6a818aeecc895;p=controller.git diff --git a/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/AbstractClientConnection.java b/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/AbstractClientConnection.java index 3cc8d7073a..e0739dd6c5 100644 --- a/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/AbstractClientConnection.java +++ b/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/AbstractClientConnection.java @@ -7,22 +7,24 @@ */ package org.opendaylight.controller.cluster.access.client; +import static java.util.Objects.requireNonNull; + import akka.actor.ActorRef; import com.google.common.annotations.VisibleForTesting; import com.google.common.base.MoreObjects; import com.google.common.base.MoreObjects.ToStringHelper; -import com.google.common.base.Preconditions; import edu.umd.cs.findbugs.annotations.SuppressFBWarnings; import java.util.Collection; +import java.util.List; import java.util.Optional; import java.util.concurrent.TimeUnit; import java.util.concurrent.TimeoutException; import java.util.concurrent.locks.Lock; import java.util.concurrent.locks.ReentrantLock; import java.util.function.Consumer; -import javax.annotation.Nonnull; import javax.annotation.concurrent.GuardedBy; import javax.annotation.concurrent.NotThreadSafe; +import org.eclipse.jdt.annotation.NonNull; import org.opendaylight.controller.cluster.access.concepts.Request; import org.opendaylight.controller.cluster.access.concepts.RequestException; import org.opendaylight.controller.cluster.access.concepts.Response; @@ -77,7 +79,8 @@ public abstract class AbstractClientConnection { private final ClientActorContext context; @GuardedBy("lock") private final TransmitQueue queue; - private final Long cookie; + private final @NonNull Long cookie; + private final String backendName; @GuardedBy("lock") private boolean haveTimer; @@ -90,19 +93,23 @@ public abstract class AbstractClientConnection { private volatile RequestException poisoned; // Private constructor to avoid code duplication. - private AbstractClientConnection(final AbstractClientConnection oldConn, final TransmitQueue newQueue) { - this.context = Preconditions.checkNotNull(oldConn.context); - this.cookie = Preconditions.checkNotNull(oldConn.cookie); - this.queue = Preconditions.checkNotNull(newQueue); + private AbstractClientConnection(final AbstractClientConnection oldConn, final TransmitQueue newQueue, + final String backendName) { + this.context = requireNonNull(oldConn.context); + this.cookie = requireNonNull(oldConn.cookie); + this.backendName = requireNonNull(backendName); + this.queue = requireNonNull(newQueue); // Will be updated in finishReplay if needed. this.lastReceivedTicks = oldConn.lastReceivedTicks; } // This constructor is only to be called by ConnectingClientConnection constructor. // Do not allow subclassing outside of this package - AbstractClientConnection(final ClientActorContext context, final Long cookie, final int queueDepth) { - this.context = Preconditions.checkNotNull(context); - this.cookie = Preconditions.checkNotNull(cookie); + AbstractClientConnection(final ClientActorContext context, final Long cookie, final String backendName, + final int queueDepth) { + this.context = requireNonNull(context); + this.cookie = requireNonNull(cookie); + this.backendName = requireNonNull(backendName); this.queue = new TransmitQueue.Halted(queueDepth); this.lastReceivedTicks = currentTime(); } @@ -110,20 +117,22 @@ public abstract class AbstractClientConnection { // This constructor is only to be called (indirectly) by ReconnectingClientConnection constructor. // Do not allow subclassing outside of this package AbstractClientConnection(final AbstractClientConnection oldConn) { - this(oldConn, new TransmitQueue.Halted(oldConn.queue, oldConn.currentTime())); + this(oldConn, new TransmitQueue.Halted(oldConn.queue, oldConn.currentTime()), oldConn.backendName); } // This constructor is only to be called (indirectly) by ConnectedClientConnection constructor. // Do not allow subclassing outside of this package - AbstractClientConnection(final AbstractClientConnection oldConn, final T newBackend, final int queueDepth) { - this(oldConn, new TransmitQueue.Transmitting(oldConn.queue, queueDepth, newBackend, oldConn.currentTime())); + AbstractClientConnection(final AbstractClientConnection oldConn, final T newBackend, + final int queueDepth) { + this(oldConn, new TransmitQueue.Transmitting(oldConn.queue, queueDepth, newBackend, oldConn.currentTime(), + requireNonNull(oldConn.context).messageSlicer()), newBackend.getName()); } public final ClientActorContext context() { return context; } - public final @Nonnull Long cookie() { + public final @NonNull Long cookie() { return cookie; } @@ -261,7 +270,7 @@ public abstract class AbstractClientConnection { TimeUnit.NANOSECONDS.sleep(delay); } catch (InterruptedException e) { Thread.currentThread().interrupt(); - LOG.debug("Interrupted after sleeping {}ns", e, currentTime() - now); + LOG.debug("Interrupted after sleeping {}ns", currentTime() - now, e); } } @@ -309,9 +318,10 @@ public abstract class AbstractClientConnection { */ @VisibleForTesting final ClientActorBehavior runTimer(final ClientActorBehavior current) { - final Optional delay; - lock.lock(); + + final List poisonEntries; + final NoProgressException poisonCause; try { haveTimer = false; final long now = currentTime(); @@ -321,36 +331,38 @@ public abstract class AbstractClientConnection { // The following line is only reliable when queue is not forwarding, but such state should not last long. // FIXME: BUG-8422: this may not be accurate w.r.t. replayed entries final long ticksSinceProgress = queue.ticksStalling(now); - if (ticksSinceProgress >= context.config().getNoProgressTimeout()) { - LOG.error("Queue {} has not seen progress in {} seconds, failing all requests", this, - TimeUnit.NANOSECONDS.toSeconds(ticksSinceProgress)); + if (ticksSinceProgress < context.config().getNoProgressTimeout()) { + // Requests are always scheduled in sequence, hence checking for timeout is relatively straightforward. + // Note we use also inquire about the delay, so we can re-schedule if needed, hence the unusual + // tri-state return convention. + final Optional delay = lockedCheckTimeout(now); + if (delay == null) { + // We have timed out. There is no point in scheduling a timer + LOG.debug("{}: connection {} timed out", context.persistenceId(), this); + return lockedReconnect(current, new RuntimeRequestException("Backend connection timed out", + new TimeoutException())); + } - lockedPoison(new NoProgressException(ticksSinceProgress)); - current.removeConnection(this); - return current; - } + if (delay.isPresent()) { + // If there is new delay, schedule a timer + scheduleTimer(delay.get()); + } else { + LOG.debug("{}: not scheduling timeout on {}", context.persistenceId(), this); + } - // Requests are always scheduled in sequence, hence checking for timeout is relatively straightforward. - // Note we use also inquire about the delay, so we can re-schedule if needed, hence the unusual tri-state - // return convention. - delay = lockedCheckTimeout(now); - if (delay == null) { - // We have timed out. There is no point in scheduling a timer - LOG.debug("{}: connection {} timed out", context.persistenceId(), this); - return lockedReconnect(current, new RuntimeRequestException("Backend connection timed out", - new TimeoutException())); + return current; } - if (delay.isPresent()) { - // If there is new delay, schedule a timer - scheduleTimer(delay.get()); - } else { - LOG.debug("{}: not scheduling timeout on {}", context.persistenceId(), this); - } + LOG.error("Queue {} has not seen progress in {} seconds, failing all requests", this, + TimeUnit.NANOSECONDS.toSeconds(ticksSinceProgress)); + poisonCause = new NoProgressException(ticksSinceProgress); + poisonEntries = lockedPoison(poisonCause); + current.removeConnection(this); } finally { lock.unlock(); } + poison(poisonEntries, poisonCause); return current; } @@ -402,9 +414,7 @@ public abstract class AbstractClientConnection { queue.remove(now); LOG.debug("{}: Connection {} timed out entry {}", context.persistenceId(), this, head); - final double time = beenOpen * 1.0 / 1_000_000_000; - head.complete(head.getRequest().toRequestFailure( - new RequestTimeoutException("Timed out after " + time + "seconds"))); + timeoutEntry(head, beenOpen); } LOG.debug("Connection {} timed out {} tasks", this, tasksTimedOut); @@ -415,19 +425,46 @@ public abstract class AbstractClientConnection { return Optional.empty(); } + private void timeoutEntry(final ConnectionEntry entry, final long beenOpen) { + // Timeouts needs to be re-scheduled on actor thread because we are holding the lock on the current queue, + // which may be the tail of a successor chain. This is a problem if the callback attempts to send a request + // because that will attempt to lock the chain from the start, potentially causing a deadlock if there is + // a concurrent attempt to transmit. + context.executeInActor(current -> { + final double time = beenOpen * 1.0 / 1_000_000_000; + entry.complete(entry.getRequest().toRequestFailure( + new RequestTimeoutException(entry.getRequest() + " timed out after " + time + + " seconds. The backend for " + backendName + " is not available."))); + return current; + }); + } + final void poison(final RequestException cause) { + final List entries; + lock.lock(); try { - lockedPoison(cause); + entries = lockedPoison(cause); } finally { lock.unlock(); } + + poison(entries, cause); + } + + // Do not hold any locks while calling this + private static void poison(final Collection entries, final RequestException cause) { + for (ConnectionEntry e : entries) { + final Request request = e.getRequest(); + LOG.trace("Poisoning request {}", request, cause); + e.complete(request.toRequestFailure(cause)); + } } @GuardedBy("lock") - private void lockedPoison(final RequestException cause) { + private List lockedPoison(final RequestException cause) { poisoned = enrichPoison(cause); - queue.poison(cause); + return queue.poison(); } RequestException enrichPoison(final RequestException ex) {