We're being too strict - there is a third possibility, which is that a
user will have defined the OS_* environment variables and expect them to
pass through.
There is a common pattern in modules where some parameters are required
only if another parameter is present AND set to a particular value. For
instance, if a cloud server state is "present" it's important to
indicate the image to be used, but if it's "absent", the image that was
used to launch it is not necessary. Provide a check that takes as an
input a list of 3-element tuples containing parameter to depend on, the
value it should be set to, and a list of parameters which are required
if the required parameter is set to the required value.
With this fix, we get a friendly error message:
failed: [localhost] => {"failed": true}
msg: value of argument start_port is not of type int and we were unable to automatically convert
This PR adds the option to retry failed ssh executions, if the failure
is caused by ssh itself, not the remote command. This can be helpful if
there are transient network issues. Retries are only implemented in the
openssh connection plugin and are disabled by default. Retries are
enabled by setting ssh_connection > retries to an integer greater
than 0.
Running a long series of playbooks, or a short playbook against a large
cluster may result in transient ssh failures, some examples logged
[here](https://trello.com/c/1yh6csEQ/13-ssh-errors).
Ansible should be able to retry an ssh connection in order to survive
transient failures.
Ansible marks a host as failed the first time it fails to contact it.