Bryan Boreham

I tried a couple of settings but couldn’t find one that suited. Maybe me opening a fork would be ok. However Weaveworks is now out of Travis credits so might...

Now it appears that Travis is not being kicked off either from PRs or from merges to master. At https://travis-ci.com/plans it says "upon validated request placed with our Support Team...

I am now thinking this is not a bug, on the basis: * there is one lock per outgoing socket * the `Send()` on the socket will block all callers...

I kind-of get what you are aiming at here, but I have a concern about the way it's done. This PR adds an extra check into the most complicated part...

Thanks for the PR! I wonder if it would be simpler to use a rate-limiter, like in #106 (call `Wait()` not `Allow()`)?

Another thought strikes me: we can turn off the "discovery" feature under Kubernetes, because we update the peer list from `kube-utils` every time a node is deleted, and a new...

Yes, basically I am happier to think that `golang.org/x/time/rate` is battle-tested and covers odd cases better than a locally-coded version. Ack on your specific use-case.

After trying lots of different things in PRs, I no longer think the golang.org rate-limiter would be a good fit for this case.

Compare https://github.com/Sirupsen/logrus/blob/master/logrus.go#L97 (not claiming I can see an easy middle ground) See also https://github.com/weaveworks/weave/commit/c38dcb1d73349eaf5336d6541552e1edb15ff7db

Good point. I agree it is redundant; however I can't see an easy way for one node to _know_ it is redundant. Something like: "all the nodes I am about...