Fixes stealing of agent responses in some cases

When deployment of environment fails with exception AgentListener.stop() is not
called and thus there remains a listener for RabbitMQ response queue.
Besides being a resource leak it introduces another problems: when that environment
get redeployed it becomes 2 listeners on the same queue and responses from agents
may be stolen by zombie listener making workflow wait forever or response
(hang deployment)

Change-Id: Ic4cedd323ab7b55690d095ed8addcb0dc3e335a7
Closes-Bug: #1369589

meta/io.murano/Classes/Environment.yaml

@@ -59,13 +59,15 @@ Methods:
   deploy:
     Usage: Action
     Body:
-      - $.agentListener.start()
-      - If: len($.applications) = 0
-        Then:
-          - $.stack.delete()
-        Else:
-          - $.applications.pselect($.deploy())
-      - $.agentListener.stop()
+      Try:
+        - $.agentListener.start()
+        - If: len($.applications) = 0
+          Then:
+            - $.stack.delete()
+          Else:
+            - $.applications.pselect($.deploy())
+      Finally:
+        - $.agentListener.stop()
 
   destroy:
     Body:

murano/dsl/executor.py

@@ -14,6 +14,7 @@
 
 import collections
 import inspect
+import sys
 import types
 import uuid
 
@@ -171,7 +172,7 @@ class MuranoDslExecutor(object):
                     return body(**params)
             except Exception as e:
                 raise dsl_exception.MuranoPlException.from_python_exception(
-                    e, context)
+                    e, context), None, sys.exc_info()[2]
         elif isinstance(body, expressions.DslExpression):
             return self.execute(
                 body, murano_class, this, context, **params)