From f371dd71b259908a9f5fd49727e46b24f397a356 Mon Sep 17 00:00:00 2001
From: AlanCoding <arominge@redhat.com>
Date: Mon, 19 Jun 2017 12:07:40 -0400
Subject: [PATCH] Run isolated heartbeat against all hosts at once

Previously we were running the playbook on a host-by-host
basis, but this changes it to pass in the list of all
isolated isntances the machine is responsible for.
Using the `json` Ansible stdout module, we are able to
parse the output for information on each host.
---
 awx/main/isolated/isolated_manager.py | 33 ++++++++++++++++-----------
 awx/main/tasks.py                     |  7 +++---
 2 files changed, 23 insertions(+), 17 deletions(-)

diff --git a/awx/main/isolated/isolated_manager.py b/awx/main/isolated/isolated_manager.py
index 44aa11e1a3..1a08b98549 100644
--- a/awx/main/isolated/isolated_manager.py
+++ b/awx/main/isolated/isolated_manager.py
@@ -305,26 +305,29 @@ class IsolatedManager(object):
             logger.warning('Cleanup from isolated job encountered error, output:\n{}'.format(buff.getvalue()))
 
     @staticmethod
-    def health_check(instance, cutoff_pk=0):
+    def health_check(instance_qs, cutoff_pk=0):
         '''
-        :param instance:            Django object representing the isolated instance
+        :param instance_qs:         List of Django objects representing the
+                                    isolated instances to manage
         :param cutoff_pk:           Job id of the oldest job still in the running state
         Method logic not yet written.
         returns the instance's capacity or None if it is not reachable
         '''
-        start_delimiter = 'wNqCXG6uul'
-        end_delimiter = 'n6kmoFyyAP'
         extra_vars = dict(
             cutoff_pk=cutoff_pk,
-            start_delimiter=start_delimiter,
-            end_delimiter=end_delimiter
         )
+        hostname_string = ''
+        for instance in instance_qs:
+            hostname_string += '{},'.format(instance.hostname)
         args = ['ansible-playbook', '-u', settings.AWX_ISOLATED_USERNAME, '-i',
-                '%s,' % instance.hostname, 'heartbeat_isolated.yml', '-e',
+                hostname_string, 'heartbeat_isolated.yml', '-e',
                 json.dumps(extra_vars)]
         module_path = os.path.join(os.path.dirname(awx.__file__), 'lib', 'management_modules')
         playbook_path = os.path.join(os.path.dirname(awx.__file__), 'playbooks')
-        env = {'ANSIBLE_LIBRARY': module_path}
+        env = {
+            'ANSIBLE_LIBRARY': module_path,
+            'ANSIBLE_STDOUT_CALLBACK': 'json'
+        }
         buff = cStringIO.StringIO()
         status, rc = run.run_pexpect(
             args, playbook_path, env, buff,
@@ -332,11 +335,15 @@ class IsolatedManager(object):
             pexpect_timeout=5
         )
         output = buff.getvalue()
-        if status != 'successful':
-            return 0  # recognized by task manager as 'unreachable'
-        result = re.search('{}(.*){}'.format(start_delimiter, end_delimiter), output)
-        cap = result.group(1)
-        return cap
+        output = output[output.find('{'):]  # Remove starting log statements
+        result = json.loads(output)
+        for instance in instance_qs:
+            task_result = result['plays'][0]['tasks'][0]['hosts'][instance.hostname]
+            if 'capacity' in task_result:
+                instance.capacity = int(task_result['capacity'])
+                instance.save(update_fields=['capacity'])
+            elif 'msg' in task_result:
+                logger.warning('Could not update capacity of {}, msg={}'.format(instance.hostname, task_result['msg']))
 
     @staticmethod
     def wrap_stdout_handle(instance, private_data_dir, stdout_handle):
diff --git a/awx/main/tasks.py b/awx/main/tasks.py
index 27f16a5d38..09e32c25f5 100644
--- a/awx/main/tasks.py
+++ b/awx/main/tasks.py
@@ -218,10 +218,9 @@ def tower_isolated_heartbeat(self):
         return
     cutoff_pk = UnifiedJob.lowest_running_id()
     # Slow pass looping over isolated IGs and their isolated instances
-    for isolated_instance in isolated_instance_qs:
-        logger.debug("Managing isolated instance {}.".format(isolated_instance.hostname))
-        isolated_instance.capacity = isolated_manager.IsolatedManager.health_check(isolated_instance, cutoff_pk=cutoff_pk)
-        isolated_instance.save(update_fields=['capacity'])
+    if len(isolated_instance_qs) > 0:
+        logger.debug("Managing isolated instances {}.".format(','.join([inst.hostname for inst in isolated_instance_qs])))
+        isolated_manager.IsolatedManager.health_check(isolated_instance_qs, cutoff_pk=cutoff_pk)
 
 
 @task(bind=True, queue='tower')