autonat: don't change status on dial request refused

[sukunrt]

We don't want to update autonat reachability status from public to unknown or from private to unknown when peers reply to dial requests with status dial refused.

[sukunrt]

@marten-seemann
Another way to do this would be to keep a flag on autonat probeImmediately which would schedule the next probe immediately.
I can change this if you think that's a better approach

[marten-seemann]

Yes, I'd like to schedule the next attempt earlier. Not sure if a bool on the struct would be the nicest way to solve this though, can we just do this in the loop?

[marten-seemann]

1. Why are we decreasing the confidence when a dial is refused?
2. I don't see any logic here to schedule the next probe sooner.

[sukunrt]

this hacky way is also used here
If confidence is not maximum the next probe will be done at retryInterval(1 minute) and not refreshTime(15minute)

I'll see if I can make that happen without keeping a bool around.

[sukunrt]

I changed it to retry immediately. Is this approach better?

[marten-seemann]

Can we have a short delay? I’d like to avoid the situation where all servers are at their limit and clients just keep churning through their list.

[sukunrt]

a bool seems like the simplest way to achive this 😅
that's what I've implemented.

[marten-seemann]

Why not just pass this as a parameter to scheduleProbe?

[sukunrt]

I changed it to pass it as a parameter but why is this better?

scheduleProbe is already making the scheduling decision based on params set on the struct like as.lastInbound why handle this info separately?

[marten-seemann]

Usually I try to keep as little members as possible, but you're right, there seems to be precedent here for not following this principle.

[sukunrt]

I agree with your logic, it is better to have this state have as little visibility as possible. We can keep this and change the other "precedents" later when we touch that piece.