QPID-4850: HA Sporadic stall in ha_tests.test_failover_send_receive

Fixed a deadlock in the test: test was killing a backup broker then waiting to
pass a checkpoint: however if the backup was expected by the primary, the primary
holds queues till it connects, so we deadlock.

git-svn-id: https://svn.apache.org/repos/asf/qpid/trunk/qpid@1483107 13f79535-47bb-0310-9956-ffa450edef68

3 files changed, 4 insertions, 9 deletions

diff --git a/cpp/src/qpid/ha/QueueReplicator.cpp b/cpp/src/qpid/ha/QueueReplicator.cpp
index 3580c49826..946831319c 100644
--- a/cpp/src/qpid/ha/QueueReplicator.cpp
+++ b/cpp/src/qpid/ha/QueueReplicator.cpp
@@ -162,6 +162,7 @@ QueueReplicator::~QueueReplicator() {}
 
 // Called from Queue::destroyed()
 void QueueReplicator::destroy() {
+    QPID_LOG(debug, logPrefix << " destroyed");
     boost::shared_ptr<Bridge> bridge2; // To call outside of lock
     {
         Mutex::ScopedLock l(lock);
diff --git a/cpp/src/qpid/ha/ReplicatingSubscription.h b/cpp/src/qpid/ha/ReplicatingSubscription.h
index 7fcb4ccf13..2780f4fd00 100644
--- a/cpp/src/qpid/ha/ReplicatingSubscription.h
+++ b/cpp/src/qpid/ha/ReplicatingSubscription.h
@@ -61,8 +61,8 @@ class QueueGuard;
  *
  * Lifecycle: broker::Queue holds shared_ptrs to this as a consumer.
  *
- * Lock Hierarchy: ReplicatingSubscription MUS NOT call QueueGuard with it's lock held
- * QueueGuard MAY call ReplicatingSubscription with it's lock held.
+ * Lock Hierarchy: ReplicatingSubscription MUST NOT call QueueGuard with its
+ * lock held QueueGuard MAY call ReplicatingSubscription with its lock held.
  */
 class ReplicatingSubscription : public broker::SemanticState::ConsumerImpl
 {
diff --git a/cpp/src/tests/ha_tests.py b/cpp/src/tests/ha_tests.py
index 3f870ba5b8..8b4ff80dd0 100755
--- a/cpp/src/tests/ha_tests.py
+++ b/cpp/src/tests/ha_tests.py
@@ -972,19 +972,13 @@ class LongTests(HaBrokerTest):
                         brokers.bounce(victim) # Next one is promoted
                         primary = next
                     else:
-                        brokers.kill(victim, promote_next=False, final=False)
-                        dead = victim
+                        brokers.bounce(victim, promote_next=False)
 
                     # Make sure we are not stalled
                     map(wait_passed, receivers, checkpoint)
                     # Run another checkpoint to ensure things work in this configuration
                     checkpoint = [ r.received+100 for r in receivers ]
                     map(wait_passed, receivers, checkpoint)
-
-                    if dead is not None:
-                        brokers.restart(dead) # Restart backup
-                        brokers[dead].ready()
-                        dead = None
                     i += 1
             except:
                 traceback.print_exc()