Just wondering if there are any thoughts/objections to the idea of making the max_fail_percentage a configurable option, either via the Ansible config files, or via an ANSIBLE_MAX_FAIL_PERCENTAGE environment variable?
I suggest this change because I’m currently working with a CI testing environment that is testing multiple ansible playbooks as part of a multinode cluster deployment and it would be useful to have the testing fail pretty much immediately, rather than have it continue until all the nodes fail.
While waiting til all have finished may fine more failures in a given CI run, it can also make it harder to diagnose things since some of the subsequent failures may be as a result of the earlier failure. Additionally failing on the first failure will release CI testing resources for re-use much sooner.
We could achieve this by setting max_fail_percentage as part of each of the plays in all of the playbooks, but that will be intrusive to the environment as a whole, which is spread across multiple projects and 10s of git repos.
Having it be configurable would allow the CI testing systems to set the max_fail_percentage attribute to 0 via the /etc/ansible/ansible.cfg file, or via ANSIBLE_MAX_FAIL_PERCENTAGE=0 in the environment for specific invocations.
A simple patch like the following, for the latest version (v1.9.1) available to install via python pip, provides this functionality with minimal changes:
`
diff -Npru ansible.orig/constants.py ansible/constants.py
— ansible.orig/constants.py 2015-06-15 10:29:50.000000000 -0400
+++ ansible/constants.py 2015-06-15 12:08:18.520155856 -0400
@@ -207,3 +207,6 @@ DEFAULT_SUBSET = None
DEFAULT_SU_PASS = None
VAULT_VERSION_MIN = 1.0
VAULT_VERSION_MAX = 1.0