V3.0 branch

[thrawn01]

Purpose

Prep Gubernator for a v3.0 release. This is a long running branch where we will merge V3 related feature once they are ready.

Removing GRPC Support

GRPC turned out to be more trouble than it was worth. Although GRPC has consistent semantics like flow control, request cancellation, and error handling, and decent performance. However, it is not without its issues.

• GRPC is more complex than is necessary for high-performance, distributed environments. Like Request Queuing, and built-in concurrency limits.
• GRPC implementations can be slower than expected (Slower than standard HTTP in some cases)
• Using GRPC can result in more code than using standard HTTP
• GRPC is not suitable for the public facing web based APIs
• GRPC has proxy and service mesh issues

Merged

• Replaced GRPC with DUH-RPC client for improved performance
  • Updated tests with new client

  • 15

  • 21

TODO

See V3 Milestone -->

[pete-woods]

I made this PR #34 to try and help with my issue #32.

V3 seems like the right time for a breaking API change like this.

[pete-woods]

I dunno if you've considered it, but this cache lib https://github.com/dgraph-io/ristretto is really fast (I see you've been looking into other cache implementations for v3)

(I see now otter implementor claiming theirs is much better than even ristretto, though :) )

[thrawn01]

@pete-woods Yes, I've tried to use ristretto for gubernator. It appears to be "eventual" in that if you write a value to the cache, then immediately retrieve it, it's possible the value you just wrote doesn't exist, as the write path hasn't synchronized with the underlying store. This occurs even though the API confirmed the write of the value as successful.

For this reason I used otter.