Bugfix: HA kill process recovery has a conflict

It happens in Nokia SUT when running in the Plugfest. The problem happens when
the start_process start to recover the killed process (like nova-api), but the
self-cured mechanism already recovered. And somehow it lead to a conflict and
has problems.
So the recover of the HA attack-recover should be improved to only recover it
when it needs to.

JIRA: YARDSTICK-1222

Change-Id: I1acb5a7d59d6fe4e0de0b0c5942fa89e051dd1ff
Signed-off-by: rexlee8776 <limingjiang@huawei.com>
(cherry picked from commit 736f3fa5d52345d6fe5174b83de043f779fa0600)

diff --git a/yardstick/benchmark/scenarios/availability/serviceha.py b/yardstick/benchmark/scenarios/availability/serviceha.py
index b6e8401..ee2eeb0 100755 (executable)
--- a/yardstick/benchmark/scenarios/availability/serviceha.py
+++ b/yardstick/benchmark/scenarios/availability/serviceha.py
@@ -29,6 +29,7 @@ class ServiceHA(base.Scenario):
         self.context_cfg = context_cfg
         self.setup_done = False
         self.data = {}
+        self.sla_pass = False
 
     def setup(self):
         """scenario setup"""
@@ -69,23 +70,25 @@ class ServiceHA(base.Scenario):
         self.monitorMgr.wait_monitors()
         LOG.info("Monitor '%s' stop!", self.__scenario_type__)
 
-        sla_pass = self.monitorMgr.verify_SLA()
+        self.sla_pass = self.monitorMgr.verify_SLA()
         for k, v in self.data.items():
             if v == 0:
-                sla_pass = False
+                self.sla_pass = False
                 LOG.info("The service process (%s) not found in the host environment", k)
 
-        result['sla_pass'] = 1 if sla_pass else 0
+        result['sla_pass'] = 1 if self.sla_pass else 0
         self.monitorMgr.store_result(result)
 
-        assert sla_pass is True, "The HA test case NOT pass the SLA"
+        assert self.sla_pass is True, "The HA test case NOT pass the SLA"
 
         return
 
     def teardown(self):
         """scenario teardown"""
-        for attacker in self.attackers:
-            attacker.recover()
+        # only recover when sla not pass
+        if not self.sla_pass:
+            for attacker in self.attackers:
+                attacker.recover()
 
 
 def _test():    # pragma: no cover

diff --git a/yardstick/tests/unit/benchmark/scenarios/availability/test_serviceha.py b/yardstick/tests/unit/benchmark/scenarios/availability/test_serviceha.py
index dd656fb..5c9bda5 100644 (file)
--- a/yardstick/tests/unit/benchmark/scenarios/availability/test_serviceha.py
+++ b/yardstick/tests/unit/benchmark/scenarios/availability/test_serviceha.py
@@ -42,6 +42,13 @@ class ServicehaTestCase(unittest.TestCase):
         }
         sla = {"outage_time": 5}
         self.args = {"options": options, "sla": sla}
+        self.test__serviceha = serviceha.ServiceHA(self.args, self.ctx)
+
+    def test___init__(self):
+
+        self.assertEqual(self.test__serviceha.data, {})
+        self.assertFalse(self.test__serviceha.setup_done)
+        self.assertFalse(self.test__serviceha.sla_pass)
 
     # NOTE(elfoley): This should be split into test_setup and test_run
     # NOTE(elfoley): This should explicitly test outcomes and states
@@ -73,3 +80,18 @@ class ServicehaTestCase(unittest.TestCase):
         ret = {}
         self.assertRaises(AssertionError, p.run, ret)
         self.assertEqual(ret['sla_pass'], 0)
+
+    @mock.patch.object(serviceha, 'baseattacker')
+    @mock.patch.object(serviceha, 'basemonitor')
+    def test__serviceha_no_teardown_when_sla_pass(self, mock_monitor,
+                                                  *args):
+        p = serviceha.ServiceHA(self.args, self.ctx)
+        p.setup()
+        self.assertTrue(p.setup_done)
+        mock_monitor.MonitorMgr().verify_SLA.return_value = True
+        ret = {}
+        p.run(ret)
+        attacker = mock.Mock()
+        p.attackers = [attacker]
+        p.teardown()
+        attacker.recover.assert_not_called()