The "if" checks had duplicated code. For example, they all called
tftbase.get_manifest(), which can be done at the end of the block.

Also, we already chose a unique and suitable name for the pod. For the
generated YAML file, we should just use the same name. Previously, the
generated name would often differ from the pod name for no reason.

Signed-off-by: Thomas Haller <thaller@redhat.com>

There are two cases for calling Task.render_file():

- we render the file of the pod itself, that is, using
  self.in_file_template and self.out_file_yaml.
- we render other resources, where the caller specifies
  the in and out file parameters.

In no case, does the caller only want to specify either the
in_file_template or the out_file_yaml parameter.

This is confusing. Add Task.render_pod_file().

Also, change the log string for plugin pods from "Server Pod Yaml"
to "Plugin Pod Yaml".

Signed-off-by: Thomas Haller <thaller@redhat.com>

The property is always directly dependent on the pod_name.
Don't track it separately. Calculate it when needed.

Signed-off-by: Thomas Haller <thaller@redhat.com>

Preserves ASCII lower case letters and digits. But other characters
are escaped using "-".

The goal is that the input unambiguously maps to a result string (which
is reversible in theory).

Also, leading and trailing '-' are escaped too, because those are not
valid in a DNS label.

We will use this for the pod-name.

Signed-off-by: Thomas Haller <thaller@redhat.com>

Use tftbase.str_sanitize() for this.

Signed-off-by: Thomas Haller <thaller@redhat.com>

This seems more pythonic. The getter has no side effects and is cheap to
evaluate.

Also, it is based on the "node_name". That property is based on
immutable attributes and thus yields the identical reference on each
call. This makes "node_name" stateless and simple to reason about. As
such, "node_name_sanitized" also gives always an equal value, albeit a
newly allocated string reference.

Signed-off-by: Thomas Haller <thaller@redhat.com>

We use the node name to derive the pod_name. For that usage we have
"node_name_sanitize", which handles any characters.

Otherwise, we also use the node name for the pod selector:

  nodeSelector:
    kubernetes.io/hostname: {{ node_name }}

This hostname is usually the same as the node name. The node name must
be a valid DNS name. Technically, it seems you could configure the
hostname to differ from the node name, and then the hostname is not
required to be a valid DNS name. But let's not support that.

So this change breaks support for systems that rename the hostname of a
node to no longer be a valid DNS name. Just don't do that. Not
supported.

What this gives is that we know that the node name does not contain
bogus characters. For example, our naive Jinja2 replacement in the pod
files is not aware that it needs to insert a YAML string. Now we at
least know that the node name won't contain characters like '"' or '['
that would break the YAML.

Signed-off-by: Thomas Haller <thaller@redhat.com>

Jinja is not aware of what it is templating. So it does not know that we
want to set a string in YAML, or how to quote that.

Luckily, we already ensure that the node name is a valid DNS name. The
form of the pod name is even more strict. So they cannot contain quotes
that could break our YAML. Still, a node name "1" would be interpreted
as a plain number in YAML. We always want a YAML string.

Quote the value in the template.

In this case, this is safe, because we know that the node name and the
pod name only contain certain characters.

Signed-off-by: Thomas Haller <thaller@redhat.com>