Webhook server and metrics server use different values to turn off #3604

What

controller-runtime turns its servers off in two different ways:

So a project that turns everything off locally passes
--metrics-bind-address=0 --health-probe-bind-address=0 --webhook-port=-1.
Users who expect --webhook-port=0 to work like the other flags get a
webhook server on 9443 instead, with no warning.

Why it matters

It is not homogeneous, which makes it very difficult and confusing for end users. Ideally, everything should work in the same way.
Having a clear convention here would be very helpful in reducing confusion and avoiding misleading behavior.

Why the webhook server cannot use 0 as it is

webhook.Options.Port is an int, and 0 is its zero value, so an unset portis 0. If 0 turned the server off, every caller that relies on the default 9443 would silently stop serving webhooks which can be harder yet for end users.

Proposal

Use one format for every server: replace the webhook server's Host and Port
with a BindAddress, the same option the metrics server and the health probe
already have.

• add webhook.Options.BindAddress: "0" turns the webhook server off, and any
  other value is the address to listen on, such as :9443;
• deprecate Host and Port in the same release, and remove them in a later
  one. Until then, Port < 0 keeps turning the server off and an unset
  BindAddress keeps today's default, :9443, so nothing breaks;
• document the rule once for all servers: "0" turns the server off.

If replacing Port is too large a change, an alternative is to make it a
*int, so that an unset port keeps the default 9443 and 0 turns the server
off, matching the other servers.