Fix cluster race condition with connections closed by broker while in use.

If a client is using a connection that is closed at the broker end
because of an error, there is a race condition that allows the
connection to be incorrectly re-created on replica brokers which can
cause those brokers to exit with an error that does not occur on the
directly connected broker.

The fix: explicitly announce new connections, shadow connections are no
longer implicitly created on first use. Make error-check a cluster
control so it can be handled independently of the lifecycle of the
connection where an error initially occured.


git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk/qpid@789947 13f79535-47bb-0310-9956-ffa450edef68

1 files changed, 1 insertions, 1 deletions

diff --git a/cpp/src/qpid/cluster/Multicaster.cpp b/cpp/src/qpid/cluster/Multicaster.cpp
index fee13c92c8..4d02d58efe 100644
--- a/cpp/src/qpid/cluster/Multicaster.cpp
+++ b/cpp/src/qpid/cluster/Multicaster.cpp
@@ -62,7 +62,7 @@ void Multicaster::mcast(const Event& e) {
     {
         sys::Mutex::ScopedLock l(lock);
         LATENCY_TRACK(cpgLatency.start());
-        if (e.getType() == DATA && e.isConnection() && holding) {
+        if (e.isConnection() && holding) {
             holdingQueue.push_back(e); 
             return;
         }