Attempt to wait but more intelligently

also added flaky to the test

tests/integration/utils/test_master.py

@@ -1,4 +1,7 @@
 # -*- coding: utf-8 -*-
+'''
+Test master code from utils
+'''
 from __future__ import absolute_import
 
 import os
@@ -11,6 +14,7 @@
 
 from tests.support.case import ShellTestCase
 from tests.support.paths import TMP_ROOT_DIR
+from tests.support.helpers import flaky
 
 DEFAULT_CONFIG = salt.config.master_config(None)
 DEFAULT_CONFIG['cachedir'] = os.path.join(TMP_ROOT_DIR, 'cache')
@@ -19,14 +23,31 @@
 class MasterUtilJobsTestCase(ShellTestCase):
 
     def setUp(self):
-        # Necessary so that the master pid health check
-        # passes as it looks for salt in cmdline
+        '''
+        Necessary so that the master pid health check
+        passes as it looks for salt in cmdline
+        '''
         setproctitle.setproctitle('salt')
 
+    @flaky
     def test_get_running_jobs(self):
+        '''
+        Test get running jobs
+        '''
         ret = self.run_run_plus("test.sleep", '90', asynchronous=True)
         jid = ret['jid']
-        time.sleep(20)
-        jobs = master.get_running_jobs(DEFAULT_CONFIG)
+
+        # Ran into a problem where the async jump was not seen until
+        # after the test had finished. This caused the test to fail
+        # because no job was present (not proc file). This attempts
+        # to wait a total of 20s before giving up.
+        attempt = 0
+        while attempt < 10:
+            jobs = master.get_running_jobs(DEFAULT_CONFIG)
+            if jobs:
+                break
+            time.sleep(2)
+            attempt += attempt + 1
+
         jids = [job['jid'] for job in jobs]
         assert jids.count(jid) == 1