Drew Erny

I've opened a proposal for this, which is at moby/moby#39447. PTAL if you're interested in this feature, and confirm that the proposal meets your needs as a user.

TODO list for the stuff we _need_ to do to ship this - [x] Write new top-level allocator - [x] Ensure `manager/allocator/allocator_test.go` runs against the new allocator - [x] Go...

@aaronlehmann you wrote some of my favorite code in swarmkit, so I'm really interested in your high-level opinion of this new code, even if you don't have time for a...

Ugh, I just realized that the allocator doesn't handle the case of node-local networks, because I've segregated allocating the driver into the `driver` package and allocating IP addresses into the...

When I plug the new allocator into the main code base and run the integration tests, tehy fail. I believe this is due to the new allocator being substantially too...

> However, we do seem to be doing some validity checks here, even for host-mode ports. yes, we are. there is no good reason for this. i am a liar.

i'm unsure how useful TLA+ is going to be for the allocator, which is not distributed and is not yet even concurrent.

I've updated the PR to include some major differences: 1. The new allocator can be used in swarmkit by setting the environment variable `SWARMKIT_USE_NEW_ALLOCATOR="iknowtherisk"`. The new allocator is gated behind...

The problem with two identical ports is that we can't know which ports on the spec map to ports on the object, which we need to know in order to...

Yeah, I forgot to regenerate. I was going to just put off doing so until my next push >