parallel multi-keyring design thoughts

[mattsb42-aws]

Collecting some thoughts here for the parallel multi-keyring whenever that happens.

• how do you specify the generator?
  • My inclination is that parallel multi-keyrings MUST NOT generate. The reason for this is because you would always need to generate before you can fan out. The right way to identify the generator keyring would be to build a multi-keyring with that keyring as the generator and a parallel multi-keyring as a child keyring.
• Controls that we will need to consider include:
  • parallelism : how many child keyrings will operate concurrently?
  • ordering/grouping/priority : should there be any requirements about one group of child keyrings finishing before another group start?
  • My inclination is that if you want this kind of grouping behavior you should get that by combining multiple parallel multi-keyrings as children of a multi-keyring.

[robin-aws]

+1 on both inclinations. I think the combination of the sequential, generation-supporting multi-keyring with the fully-parallel non-generating multi-keyring should support all possible concurrency scenarios.

It's also worth thinking about how other flavours of multi-keyrings would interact with this though (such as N-of-M). Should there be separate sequential and parallel versions of that?

[mattsb42-aws]

N-of-M decrypt is a completely different beast and (at least as I envision it in my head) will result in a single "EDK". From the perspective of anything outside of it, it will manifest as a single keyring, not a collection of keyrings.

N-of-M encrypt is an interesting question, because it would need to answer similar questions as both the multi-keyring and the parallel multi-keyring, but also additional questions (ex: is one group required? what is the threshold?) I suspect that some of these will fall into similar collapsing patterns as explored above. Maybe it would collapse into a "require N EDKs" keyring that just passes through to an underlying keyring and analyses the results and a "allow encrypt to fail" keyring that allows encrypt to proceed when the keyring it wraps fails?