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%2FClientActorBehavior.java;h=ccfbba6cd3d94f92ef9757999352893e1b0f7270;hb=773ee5e88f6ea658e4a57634976c500b0ef8b4a2;hp=8f71edb6ca2e36c2a7a006a8e3b462a60f96aa94;hpb=930747a6ba5d888d2fbe54473132680e4621d858;p=controller.git

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 8f71edb6ca..ccfbba6cd3 100644
--- 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);
@@ -296,16 +317,35 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
             conn.finishReplay(forwarder);
 
             // Make sure new lookups pick up the new connection
-            connections.replace(shard, conn, newConn);
-            LOG.info("{}: 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 {} in {}", persistenceId(), conn, newConn, sw);
+            }
         } 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")
@@ -314,11 +354,20 @@ public abstract class ClientActorBehavior<T extends BackendInfo> extends
         final ReconnectingClientConnection<T> conn = (ReconnectingClientConnection<T>)newConn;
         LOG.info("{}: connection {} reconnecting as {}", persistenceId(), oldConn, newConn);
 
-        final boolean replaced = connections.replace(oldConn.cookie(), (AbstractClientConnection<T>)oldConn, conn);
-        if (!replaced) {
-            final AbstractClientConnection<T> existing = connections.get(oldConn.cookie());
-            LOG.warn("{}: old connection {} does not match existing {}, new connection {} in limbo", persistenceId(),
-                oldConn, existing, 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();