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=da016bae885be4ea9fc15035f8e05f6e62c6d058;hb=b74c6012092e47430a8f4d6f4ddeb1d3e2b1b7df;hp=d37893bd7c3a9862ad89fb9d88b94337d0355f13;hpb=e1c283de301355cb8fa3f7d4fa28a6dd0af501eb;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 d37893bd7c..da016bae88 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 @@ -13,6 +13,7 @@ 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.Optional; import java.util.concurrent.TimeUnit; import java.util.concurrent.locks.Lock; @@ -40,14 +41,31 @@ import scala.concurrent.duration.FiniteDuration; public abstract class AbstractClientConnection { private static final Logger LOG = LoggerFactory.getLogger(AbstractClientConnection.class); - // Keep these constants in nanoseconds, as that prevents unnecessary conversions in the fast path + /* + * Timers involved in communication with the backend. There are three tiers which are spaced out to allow for + * recovery at each tier. Keep these constants in nanoseconds, as that prevents unnecessary conversions in the fast + * path. + */ + /** + * Backend aliveness timer. This is reset whenever we receive a response from the backend and kept armed whenever + * we have an outstanding request. If when this time expires, we tear down this connection and attept to reconnect + * it. + */ @VisibleForTesting - static final long NO_PROGRESS_TIMEOUT_NANOS = TimeUnit.MINUTES.toNanos(15); + static final long BACKEND_ALIVE_TIMEOUT_NANOS = TimeUnit.SECONDS.toNanos(30); + + /** + * Request timeout. If the request fails to complete within this time since it was originally enqueued, we time + * the request out. + */ @VisibleForTesting - static final long REQUEST_TIMEOUT_NANOS = TimeUnit.SECONDS.toNanos(30); + static final long REQUEST_TIMEOUT_NANOS = TimeUnit.MINUTES.toNanos(2); - private static final FiniteDuration REQUEST_TIMEOUT_DURATION = FiniteDuration.apply(REQUEST_TIMEOUT_NANOS, - TimeUnit.NANOSECONDS); + /** + * No progress timeout. A client fails to make any forward progress in this time, it will terminate itself. + */ + @VisibleForTesting + static final long NO_PROGRESS_TIMEOUT_NANOS = TimeUnit.MINUTES.toNanos(15); private final Lock lock = new ReentrantLock(); private final ClientActorContext context; @@ -58,6 +76,11 @@ public abstract class AbstractClientConnection { @GuardedBy("lock") private boolean haveTimer; + /** + * Time reference when we saw any activity from the backend. + */ + private long lastReceivedTicks; + private volatile RequestException poisoned; // Do not allow subclassing outside of this package @@ -66,6 +89,7 @@ public abstract class AbstractClientConnection { this.context = Preconditions.checkNotNull(context); this.cookie = Preconditions.checkNotNull(cookie); this.queue = Preconditions.checkNotNull(queue); + this.lastReceivedTicks = currentTime(); } // Do not allow subclassing outside of this package @@ -73,6 +97,7 @@ public abstract class AbstractClientConnection { this.context = oldConnection.context; this.cookie = oldConnection.cookie; this.queue = new TransmitQueue.Halted(targetQueueSize); + this.lastReceivedTicks = oldConnection.lastReceivedTicks; } public final ClientActorContext context() { @@ -131,9 +156,9 @@ public abstract class AbstractClientConnection { public abstract Optional getBackendInfo(); - final Iterable startReplay() { + final Collection startReplay() { lock.lock(); - return queue.asIterable(); + return queue.drain(); } @GuardedBy("lock") @@ -159,8 +184,8 @@ public abstract class AbstractClientConnection { } if (queue.isEmpty()) { - // The queue is becoming non-empty, schedule a timer - scheduleTimer(REQUEST_TIMEOUT_DURATION); + // The queue is becoming non-empty, schedule a timer. + scheduleTimer(entry.getEnqueuedTicks() + REQUEST_TIMEOUT_NANOS - now); } return queue.enqueue(entry, now); } finally { @@ -183,7 +208,7 @@ public abstract class AbstractClientConnection { * @param delay Delay, in nanoseconds */ @GuardedBy("lock") - private void scheduleTimer(final FiniteDuration delay) { + private void scheduleTimer(final long delay) { if (haveTimer) { LOG.debug("{}: timer already scheduled", context.persistenceId()); return; @@ -192,8 +217,14 @@ public abstract class AbstractClientConnection { LOG.debug("{}: connection has successor, not scheduling timer", context.persistenceId()); return; } - LOG.debug("{}: scheduling timeout in {}", context.persistenceId(), delay); - context.executeInActor(this::runTimer, delay); + + // If the delay is negative, we need to schedule an action immediately. While the caller could have checked + // for that condition and take appropriate action, but this is more convenient and less error-prone. + final long normalized = delay <= 0 ? 0 : Math.min(delay, BACKEND_ALIVE_TIMEOUT_NANOS); + + final FiniteDuration dur = FiniteDuration.fromNanos(normalized); + LOG.debug("{}: scheduling timeout in {}", context.persistenceId(), dur); + context.executeInActor(this::runTimer, dur); haveTimer = true; } @@ -206,13 +237,14 @@ public abstract class AbstractClientConnection { */ @VisibleForTesting final ClientActorBehavior runTimer(final ClientActorBehavior current) { - final Optional delay; + final Optional delay; lock.lock(); try { haveTimer = false; final long now = currentTime(); // 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 >= NO_PROGRESS_TIMEOUT_NANOS) { LOG.error("Queue {} has not seen progress in {} seconds, failing all requests", this, @@ -244,7 +276,7 @@ public abstract class AbstractClientConnection { } @VisibleForTesting - final Optional checkTimeout(final long now) { + final Optional checkTimeout(final long now) { lock.lock(); try { return lockedCheckTimeout(now); @@ -262,19 +294,38 @@ public abstract class AbstractClientConnection { @SuppressFBWarnings(value = "NP_OPTIONAL_RETURN_NULL", justification = "Returning null Optional is documented in the API contract.") @GuardedBy("lock") - private Optional lockedCheckTimeout(final long now) { - final ConnectionEntry head = queue.peek(); - if (head == null) { + private Optional lockedCheckTimeout(final long now) { + if (queue.isEmpty()) { return Optional.empty(); } - final long beenOpen = now - head.getEnqueuedTicks(); - if (beenOpen >= REQUEST_TIMEOUT_NANOS) { - LOG.debug("Connection {} has a request not completed for {} nanoseconds, timing out", this, beenOpen); + final long backendSilentTicks = now - lastReceivedTicks; + if (backendSilentTicks >= BACKEND_ALIVE_TIMEOUT_NANOS) { + LOG.debug("Connection {} has not seen activity from backend for {} nanoseconds, timing out", this, + backendSilentTicks); return null; } - return Optional.of(FiniteDuration.apply(REQUEST_TIMEOUT_NANOS - beenOpen, TimeUnit.NANOSECONDS)); + int tasksTimedOut = 0; + for (ConnectionEntry head = queue.peek(); head != null; head = queue.peek()) { + final long beenOpen = now - head.getEnqueuedTicks(); + if (beenOpen < REQUEST_TIMEOUT_NANOS) { + return Optional.of(REQUEST_TIMEOUT_NANOS - beenOpen); + } + + tasksTimedOut++; + queue.remove(now); + LOG.debug("Connection {} timed out entryt {}", this, head); + head.complete(head.getRequest().toRequestFailure( + new RequestTimeoutException("Timed out after " + beenOpen + "ns"))); + } + + LOG.debug("Connection {} timed out {} tasks", this, tasksTimedOut); + if (tasksTimedOut != 0) { + queue.tryTransmit(now); + } + + return Optional.empty(); } final void poison(final RequestException cause) { @@ -299,6 +350,7 @@ public abstract class AbstractClientConnection { final void receiveResponse(final ResponseEnvelope envelope) { final long now = currentTime(); + lastReceivedTicks = now; final Optional maybeEntry; lock.lock();