Fix etcd scaleup plays

This commit ensures that only the proper host groups
have sanity checks run during etcd scaleup.

This commit also adds additional debugging statements
to sanity_checks.py to make it easier to debug when
an error occurs.

Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1543771

2 files changed, 19 insertions, 1 deletions

diff --git a/playbooks/openshift-etcd/scaleup.yml b/playbooks/openshift-etcd/scaleup.yml
index 3e2fca8d4..1b2229baa 100644
--- a/playbooks/openshift-etcd/scaleup.yml
+++ b/playbooks/openshift-etcd/scaleup.yml
@@ -45,6 +45,7 @@
   vars:
     skip_version: True
     l_init_fact_hosts: "oo_masters_to_config:oo_etcd_to_config:oo_lb_to_config:oo_new_etcd_to_config"
+    l_sanity_check_hosts: "{{ groups['oo_new_etcd_to_config'] | union(groups['oo_masters_to_config']) | union(groups['oo_etcd_to_config']) }}"
     l_openshift_version_set_hosts: "all:!all"
     l_openshift_version_check_hosts: "all:!all"
   when:
diff --git a/roles/lib_utils/action_plugins/sanity_checks.py b/roles/lib_utils/action_plugins/sanity_checks.py
index 09ce55e8f..ce54debc2 100644
--- a/roles/lib_utils/action_plugins/sanity_checks.py
+++ b/roles/lib_utils/action_plugins/sanity_checks.py
@@ -54,6 +54,12 @@ class ActionModule(ActionBase):
     def template_var(self, hostvars, host, varname):
         """Retrieve a variable from hostvars and template it.
            If undefined, return None type."""
+        # We will set the current host and variable checked for easy debugging
+        # if there are any unhandled exceptions.
+        # pylint: disable=W0201
+        self.last_checked_var = varname
+        # pylint: disable=W0201
+        self.last_checked_host = host
         res = hostvars[host].get(varname)
         if res is None:
             return None
@@ -156,6 +162,11 @@ class ActionModule(ActionBase):
         # pylint: disable=W0201
         self.task_vars = task_vars or {}
 
+        # pylint: disable=W0201
+        self.last_checked_host = "none"
+        # pylint: disable=W0201
+        self.last_checked_var = "none"
+
         # self._task.args holds task parameters.
         # check_hosts is a parameter to this plugin, and should provide
         # a list of hosts.
@@ -172,7 +183,13 @@ class ActionModule(ActionBase):
 
         # We loop through each host in the provided list check_hosts
         for host in check_hosts:
-            self.run_checks(hostvars, host)
+            try:
+                self.run_checks(hostvars, host)
+            except Exception as uncaught_e:
+                msg = "last_checked_host: {}, last_checked_var: {};"
+                msg = msg.format(self.last_checked_host, self.last_checked_var)
+                msg += str(uncaught_e)
+                raise errors.AnsibleModuleError(msg)
 
         result["changed"] = False
         result["failed"] = False