Fix broker crash "confirmed N but only sent M" with managed agents running.

The broker's ManagementAgent caches schemas from managed agents.
This cache was not being replicated to new cluster members.

If an agent such as sesame was running and connected to a newly-joined
broker, that broker could send schema request messages which were not
sent by other brokers that had the schema in cache. This resulted in
the other brokers exiting with a "confirmed N but only sent M"
message.


git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk/qpid@897955 13f79535-47bb-0310-9956-ffa450edef68

1 files changed, 12 insertions, 1 deletions

diff --git a/cpp/src/qpid/cluster/UpdateClient.cpp b/cpp/src/qpid/cluster/UpdateClient.cpp
index b20cc907a2..d4bd4da7f8 100644
--- a/cpp/src/qpid/cluster/UpdateClient.cpp
+++ b/cpp/src/qpid/cluster/UpdateClient.cpp
@@ -138,10 +138,21 @@ void UpdateClient::update() {
     session.queueDelete(arg::queue=UPDATE);
     session.close();
 
-    // Update queue listeners: must come after sessions so consumerNumbering is populated.
+    // Update queue listeners: must come after sessions so consumerNumbering is populated
     b.getQueues().eachQueue(boost::bind(&UpdateClient::updateQueueListeners, this, _1));
 
     ClusterConnectionProxy(session).expiryId(expiry.getId());
+
+    // FIXME aconway 2010-01-08: we should enforce that all cluster members 
+    // have mgmt enabled or none of them do.
+
+    management::ManagementAgent* agent = updaterBroker.getManagementAgent();
+    if (agent) {
+        string schemaData;
+        agent->exportSchemas(schemaData);
+        ClusterConnectionProxy(session).managementSchema(schemaData);
+    }
+
     ClusterConnectionMembershipBody membership;
     map.toMethodBody(membership);
     AMQFrame frame(membership);