Workaround for dnf+docker version race condition

In some versions of dnf the rpm database can require maintenance after
usage. Unfortunately, this trips up get_version_output in the
openshift_facts library as the maintenance may interrupt the STDOUT
stream. This makes the resulting data returned from get_version_output
malformed and unparsable. This change adds an rpm --rebuilddb command
before requesting docker version output when the ansible package manager
is dnf. This causes the STDOUT maintenance messages to no longer
interrupt the expected yaml stream as well as releases locks in the rpm
database.

roles/openshift_facts/library/openshift_facts.py

@@ -1256,6 +1256,13 @@ def is_service_running(service):
     return service_running
 
 
+def rpm_rebuilddb():
+    """
+    Runs rpm --rebuilddb to ensure the db is in good shape.
+    """
+    module.run_command(['/usr/bin/rpm', '--rebuilddb'])  # noqa: F405
+
+
 def get_version_output(binary, version_cmd):
     """ runs and returns the version output for a command """
     cmd = []
@@ -1966,6 +1973,11 @@ class OpenShiftFacts(object):
         if 'docker' in roles:
             docker = dict(disable_push_dockerhub=False,
                           options='--log-driver=json-file --log-opt max-size=50m')
+            # NOTE: This is a workaround for a dnf output racecondition that can occur in
+            # some situations. See https://bugzilla.redhat.com/show_bug.cgi?id=918184
+            if self.system_facts['ansible_pkg_mgr'] == 'dnf':
+                rpm_rebuilddb()
+
             version_info = get_docker_version_info()
             if version_info is not None:
                 docker['api_version'] = version_info['api_version']