Max percentage failure support

In this patch we update the new strategies so they support a max failure
percentage on a play. Which will be useful for the tripleo roles so we
can tolerate some % of failure (e.g. on computes) during a deployment.

Co-Authored-By: Alex Schultz <aschultz@redhat.com>
Co-Authored-By: Kevin Carter <kecarter@redhat.com>
Co-Authored-By: Emilien Macchi <emilien@redhat.com>

Change-Id: I30a930e4f7b8200ffa67c17b967db8a1fb60b4c0

tripleo_ansible/ansible_plugins/strategy/tripleo_base.py

@@ -47,6 +47,7 @@ class TripleoBase(StrategyBase):
         self._play_context = None
         self._strat_results = []
         self.noop_task = None
+        self._fail_cache = {}
         # these were defined in 2.9
         self._has_hosts_cache = False
         self._has_hosts_cache_all = False
@@ -84,6 +85,50 @@ class TripleoBase(StrategyBase):
         task.name = name
         self._callback_sent = True
 
+    def _get_fail_percent(self, host):
+        """Return maximum percentage failure per role"""
+        if host and host in self._fail_cache:
+            return self._fail_cache[host]
+
+        fail_vars = self._variable_manager.get_vars(play=self._iterator._play,
+                                                    host=host,
+                                                    task=None)
+        percent = fail_vars.get('max_fail_percentage', 0)
+        role = fail_vars.get('tripleo_role_name', 'default')
+        self._fail_cache[host] = (percent, role)
+        return (percent, role)
+
+    def _check_fail_percent(self, host, current_failures):
+        """Check if max fail pourcentage was reached
+
+       When a failure occurs for a host, check if we reached
+       the max percentage of failure for the group in which
+       the host is part from.
+       """
+        percent, role = self._get_fail_percent(host)
+        current_failed = current_failures.get(role, 1)
+
+        groups = self._inventory.get_groups_dict()
+        group_count = len(groups.get(role, []))
+        if group_count == 0:
+            return True
+        failed_percent = (current_failed / group_count) * 100
+        if failed_percent > percent:
+            return True
+        return False
+
+    def _get_current_failures(self):
+        """Return the number of failures per role"""
+        failures = {}
+        for host, _ in self._iterator.get_failed_hosts().items():
+            host_obj = self._inventory.get_host(host)
+            per, role = self._get_fail_percent(host_obj)
+            if role in failures:
+                failures[role] += 1
+            else:
+                failures[role] = 1
+        return failures
+
     def process_includes(self, host_results, noop=False):
         """Handle includes
 

tripleo_ansible/ansible_plugins/strategy/tripleo_free.py

@@ -123,10 +123,12 @@ class StrategyModule(BASE.TripleoBase):
         function returns True if there were failures and False if
         there are no failures.
         """
+        fail_lookup = self._get_current_failures()
         if self._any_errors_fatal:
             for res in results:
                 if ((res.is_failed() or res._task.action == 'meta')
-                        and self._iterator.is_failed(res._host)):
+                        and self._iterator.is_failed(res._host)
+                        and self._check_fail_percent(res._host, fail_lookup)):
                     return True
         return False
 

tripleo_ansible/ansible_plugins/strategy/tripleo_linear.py

@@ -334,15 +334,20 @@ class StrategyModule(BASE.TripleoBase):
 
                 failed_hosts = []
                 unreachable_hosts = []
+                fail_lookup = self._get_current_failures()
                 for res in self._strat_results:
                     if ((res.is_failed() or res._task.action == 'meta')
                             and self._iterator.is_failed(res._host)):
-                        failed_hosts.append(res._host.name)
+                        failed_hosts.append(res._host)
                     elif res.is_unreachable():
-                        unreachable_hosts.append(res._host.name)
+                        unreachable_hosts.append(res._host)
 
-                # TODO(mwhahaha): handle max_fail_percentage by tripleo role
+                errored = False
-                if (self._any_errors_fatal
+                for host in set(failed_hosts + unreachable_hosts):
+                    errored = self._check_fail_percent(host, fail_lookup)
+                    if errored:
+                        break
+                if (errored and self._any_errors_fatal
                         and (len(failed_hosts) > 0
                              or len(unreachable_hosts) > 0)):
                     result = self._process_failures()