Automatic relay selection

[Mrc527]

Regarding ticket #678, I would like to propose a small enhancement.

It's a bit annoying you have to setup on all the clients which relay to use. How about communicating relays availability via lighthouse and letting the client to automatically select the best relay? (ping based maybe? Or via periodic ping/throughput tests?)

This would help massively, streamlining the deployment in environments where you have considerable amount of nodes.

P.S. the relay function is MEGA! Love it.

[rawdigits]

We are considering doing this and it was discussed at length when we worked through the goals here. The primary reason it wasn't done is that we are trying to avoid a stampeding herd fallback scenario, especially one that might affect lighthouses if people use their lighthouses as their relays. If enough things hit the same relay, it could harm network availability, which is our top priority.

I would say this relay config follows our existing design, where we let clients configure their own firewall rules (generally done by config management systems) as opposed to trying to distribute them out.

Another system that I rather like (but we haven't tried/worked on) is allowing a dns name for relays, so that you could round robin relays with simple dns queries. The answers can be nebula IPs, but it would be uncomplicated. It would be up to the implementor whether they wanted to round robin dns answers or some other method...

[Mrc527]

But actually, exactly because

If enough things hit the same relay, it could harm network availability, which is our top priority.

I would try to setup some sort of automatic balancing.

Otherwise a human being (myself) is setting a fixed relay for all of the clients because "it's easier" to handle on the config side and I crash the node.

Love the DNS-based round robin idea but again, I would vote for an automatic way. If you let the user do that, in my opinion it will not happen 😄

[jstasiak]

I'm hitting a similar problem but not quite the same.

In my case clients have three relays set up: R1, R2, R3. R1 should be the most preferred one, R2 should only be used when R1 is not available and R3 is the relay of last resort.

Is there a way to express this in Nebula configuration at the moment? Is the ordering of the relay array in the configuration affecting anything?

[johnmaguire]

Hi @jstasiak - currently Nebula will try all relays "at the same time." There is no way to express an ordered preference of relays currently, sorry!

[jstasiak]

That's fair, thank you for the information!

[victorrodriguez1984]

It was on this product, but I detected bad performance
https://kuebler-it.github.io/MeshVPN/

It wuold be good a real full mesh and secret shared to subscriptors

[IPQualityScore]

Sorry to dig up a old thread, but this is something we'd like to see implemented.

[johnmaguire]

@IPQualityScore The best way to show your interest in a feature request is to click the 👍 reaction on the first post in the issue. This allows us to sort by upvotes. Thanks!