added autodetect of interfaces on private networks

[aldernero]

What this PR does:
This PR sets the interface lists in cfg.InfNames to contain all network interfaces that have an address on a private network, conformant to RFC1918. This replaces the static []string{"en0", "eth0"} that is currently used.

If the new function, PrivateNetworkInterfaces doesn't find any interfaces matching the RFC1918 criteria, it defaults to [en0, eth0].

Which issue(s) this PR fixes:

Fixes #881

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[pstibrany]

LGTM, but the change in dskit needs to happen in github.com/grafana/dskit repository first.

[pstibrany]

Files under vendor/ directory are library code, updated using go mod vendor. This one lives at github.com/grafana/dskit, and needs to be updated there. Note that dskit cannot use code from Mimir, the dependency is from Mimir to dskit instead.

[pstibrany]

(This implies that util.PrivateNetworkInterfaces() may need to live in dskit).

[simonswine]

Thanks for providing the draft PR. I think it is going into the right direction. I agree with all that @pstibrany said.

I also do think we need to look at the order of the interfaces that are returned as that list is gonna be used later to determine an IP address with https://github.com/grafana/dskit/blob/0bff37e60ff46c6f7e634abd42366aeebe0784bf/ring/util.go#L176

For my local machine the list is:

[wlp0s20f3 (my wifi) br-240fb45644c4 (another docker brdige) docker0 enp0s20f0u4u4 (my ethernet that has the default route)]

So it would always choose the IP address of the wifi adapter rather than the ethernet.

I think compared to what we have currently (hardcoded list of interfaces) this is only a minor problem. But maybe there is a way listing the route table and take the interface names from there. (Might be not that os independent, I don't think it exists in net)

[simonswine]

As a small addition, this is my routing table:

$ ip route
default via 172.20.74.254 dev enp0s20f0u4u4 proto dhcp metric 100
default via 172.20.74.254 dev wlp0s20f3 proto dhcp metric 600
172.17.0.0/16 dev docker0 proto kernel scope link src 172.17.0.1 linkdown
172.18.0.0/16 dev br-240fb45644c4 proto kernel scope link src 172.18.0.1 linkdown
172.20.74.0/24 dev enp0s20f0u4u4 proto kernel scope link src 172.20.74.176 metric 100
172.20.74.0/24 dev wlp0s20f3 proto kernel scope link src 172.20.74.200 metric 600

With those rules, we should be able to determine the interfaces:

• Select interface routes with destination network in the private space
• Failing that use the default route
• The interface with the smallest metric for destination is preferred

Not too sure if that is a over engineered, but I think that approach could lead to quite good results

[pracucci]

Thanks for working on this! I've some comments here, even if final code will move to dskit.

I see @simonswine feedback as a nice addition. I would not block this PR on that, and I would try to work on it as a follow up PR.

[pracucci]

This should be removed.

[pracucci]

I would move the logic to a private function that takes the list of interfaces in input, so that you can test it. My comment is about adding unit tests to this function :)

[pracucci]

I'm a bit dubious about this log because:

1. It gets logged each time we call PrivateNetworkInterfaces()
2. It gets logged even if the user overrides the config, so the result of PrivateNetworkInterfaces() doesn't really matter for them (this comment applies to other logs too)

[simonswine]

I see @simonswine feedback as a nice addition. I would not block this PR on that, and I would try to work on it as a follow up PR.

Totally agree with this. What this PR contains right now, is already a significant improvement, let's get in first.

[aldernero]

I'm going to close this PR and open a new one that covers integrating the recent dskit change (that replaced much of this).