protocol-determined vdaf representation

[jbr]

closes #410

[jbr]

blocked on divviup/janus#1743 no longer blocked, but will send mismatched histogram representations in dap-05 until that pr is merged

[inahga]

It otherwise looks good, I like the representation_for_protocol pattern.

[tgeoghegan]

Couple high level questions:

• This sets protocol = DAP-04 on existing paired aggregators, which makes sense. Do we require that all further aggregators someone pairs have to include their supported protocol in the capabilities API?
• Why store the protocol version in the tasks table? If each aggregator supports exactly one protocol version (which I think is the right thing to do), then a task's protocol version is implied by the protocol version of the aggregators it uses.

[jbr]

This sets protocol = DAP-04 on existing paired aggregators, which makes sense. Do we require that all further aggregators someone pairs have to include their supported protocol in the capabilities API?

Yes. If they do not, they will be treated as DAP-04 for posterity. This is a mandatory field in the config api now.

Edit: Wait, maybe I misunderstood. If they are in fact DAP-04, they don't need to advertise a protocol and it will work fine. There's a serde(default) on Protocol and the default is Dap04.

Why store the protocol version in the tasks table? If each aggregator supports exactly one protocol version (which I think is the right thing to do), then a task's protocol version is implied by the protocol version of the aggregators it uses.

I'm open to dropping it. I know aggregators cannot currently upgrade or support multiple DAP versions, but I was defensively programming against a future scenario in which it isn't always possible to determine from the aggregator pair. This is also why there's a variable named resolved_protocol — with the idea that there may be some further logic around protocol selection given a pair of aggregators. I think if I were reviewing this, I'd say that's currently unnecessary and we can add it and backfill it when/if we get there.

Edit: Removed the task protocol column

[tgeoghegan]

This all works for me:

• existing DAP-04 aggregators will be treated as supporting DAP-04, but any new aggregators that support DAP-05 will include the protocol version advertisement so this will all work
• inferring task protocol version from aggregator version also makes sense.