Rough work on protocol refactoring

[Stef Walter]

So as we discussed elsewhere ... past proof-of-concept, our internal
protocol needs a bit better framing in order to handle the modularity
stuff and be extensible.

* Use only one Websocket for efficiency
 * Connect out to multiple DBus services (not just cockpitd)
 * Support multiple payload types in future (not only dbus-json)
 * Alternate credentials (ie: initial setup)

Here's a first whack at documenting what it might look like:

https://github.com/stefwalter/cockpit/blob/protocol-refactor/doc/protocol.md

I'm working on some cleanup patches to dbus-server, and building some
testing around it before getting into implementing this.

Stef

[Jan Safranek]

On 02/13/2014 03:26 PM, Stef Walter wrote:
> So as we discussed elsewhere ... past proof-of-concept, our internal
> protocol needs a bit better framing in order to handle the modularity
> stuff and be extensible.
> Use only one Websocket for efficiency
> Connect out to multiple DBus services (not just cockpitd)
> Support multiple payload types in future (not only dbus-json)
> Alternate credentials (ie: initial setup)
>
> Here's a first whack at documenting what it might look like:
>
> https://github.com/stefwalter/cockpit/blob/protocol-refactor/doc/protocol.md

The protocol looks usable, I would only add also 'protocol version'
field, this can save a lot of headache when you consider moving from
json to something else.

Out of curiosity, are there any alternative already defined/used protocols?

Jan

[Marius Vollmer]

Jan Safranek  writes:
> The protocol looks usable, I would only add also 'protocol version'
> field, this can save a lot of headache when you consider moving from
> json to something else.

I think we can include that in the "protocol" string, if necessary.

[Stef Walter]

On 17.02.2014 09:02, Marius Vollmer wrote:
> Jan Safranek  writes:
> > The protocol looks usable, I would only add also 'protocol
> > version' field, this can save a lot of headache when you consider
> > moving from json to something else.
>
> I think we can include that in the "protocol" string, if necessary.

Yes. Did that. Although it's now called "payload" to make it clearer
what it's about.

Updated documentation here:

https://github.com/stefwalter/cockpit/blob/protocol-refactor/doc/protocol.md

Merge request here for initial work on this:

https://github.com/cockpit-project/cockpit/pull/230
> Out of curiosity, are there any alternative already defined/used protocols?

This is really about framing and control commands, over a text based
transport (ie: old WebSockets). It ends up being pretty Cockpit
specific. Nothing else comes to mind that would fit well here.

Stef

[Marius Vollmer]

Stef Walter  writes:
> Here's a first whack at documenting what it might look like:
>
> https://github.com/stefwalter/cockpit/blob/protocol-refactor/doc/protocol.md

Nice!