BUG-8445: check sessionId before propagating failures

[controller.git] / opendaylight / md-sal / cds-access-client / src / main / java / org / opendaylight / controller / cluster / access / client / ClientActorBehavior.java

diff --git a/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/ClientActorBehavior.java b/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/ClientActorBehavior.java
index 9ba118ed6b9c95a1d5c914a4b681413af4cce95d..ccfbba6cd3d94f92ef9757999352893e1b0f7270 100644 (file)
--- a/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/ClientActorBehavior.java
+++ b/opendaylight/md-sal/cds-access-client/src/main/java/org/opendaylight/controller/cluster/access/client/ClientActorBehavior.java
@@ -9,9 +9,11 @@ package org.opendaylight.controller.cluster.access.client;
 
 import com.google.common.annotations.Beta;
 import com.google.common.base.Preconditions;
+import com.google.common.base.Stopwatch;
 import com.google.common.base.Verify;
 import java.util.Collection;
 import java.util.Map;
+import java.util.Optional;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.TimeUnit;
 import java.util.concurrent.TimeoutException;
@@ -159,6 +161,25 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
     }
 
     private ClientActorBehavior<T> internalOnRequestFailure(final FailureEnvelope command) {
+        final AbstractClientConnection<T> conn = getConnection(command);
+        if (conn != null) {
+            /*
+             * We are talking to multiple actors, which may be lagging behind our state significantly. This has
+             * the effect that we may be receiving responses from a previous connection after we have created a new
+             * one to a different actor.
+             *
+             * Since we are already replaying requests to the new actor, we want to ignore errors reported on the old
+             * connection -- for example NotLeaderException, which must not cause a new reconnect. Check the envelope's
+             * sessionId and if it does not match our current connection just ignore it.
+             */
+            final Optional<T> optBackend = conn.getBackendInfo();
+            if (optBackend.isPresent() && optBackend.get().getSessionId() != command.getSessionId()) {
+                LOG.debug("{}: Mismatched current connection {} and envelope {}, ignoring response", persistenceId(),
+                    conn, command);
+                return this;
+            }
+        }
+
         final RequestFailure<?, ?> failure = command.getMessage();
         final RequestException cause = failure.getCause();
         if (cause instanceof RetiredGenerationException) {
@@ -168,7 +189,6 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
             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;
@@ -178,7 +198,6 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
             }
         }
         if (cause instanceof OutOfSequenceEnvelopeException) {
-            final AbstractClientConnection<T> conn = getConnection(command);
             if (conn instanceof ReconnectingClientConnection) {
                 // Already reconnecting, do not churn the logs
                 return this;
@@ -278,6 +297,8 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
         LOG.info("{}: resolved shard {} to {}", persistenceId(), shard, backend);
         final long stamp = connectionsLock.writeLock();
         try {
+            final Stopwatch sw = Stopwatch.createStarted();
+
             // Create a new connected connection
             final ConnectedClientConnection<T> newConn = new ConnectedClientConnection<>(conn.context(),
                     conn.cookie(), backend);
@@ -301,7 +322,7 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
                 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);
+                LOG.info("{}: replaced connection {} with {} in {}", persistenceId(), conn, newConn, sw);
             }
         } finally {
             connectionsLock.unlockWrite(stamp);