Races in dial queue tests

[anacrolix]

@raulk There's races in the dial queue tests around the manipulation of the dial queue variables. I'm also able to trigger panic: sync: negative WaitGroup in TestDialQueueMutePeriodHonored when I was messing around trying to fix them if the scaling mute period is set to 0. I think that might indicate some assumptions about timing.

[raulk]

I can see this happening. Weirdly, it didn’t error locally on any execution, nor on 2 CIs. I guess tests execute in a different direction down under, just like the drains circling :P

I’ll build on your latest PR and make all params adjustable via a struct. Also paves the way to make some form of this queue exportable to other areas of the system that might benefit.

[anacrolix]

Sounds ideal.

[anacrolix]

Actually you should just merge #241 now, and use a struct in another PR.

[anacrolix]

This will stop the CI from failing so we can move on in the short term.

[raulk]

@anacrolix sounds like a plan. Merged.

[anacrolix]

Do you want to try a test for a scaling mute period of 0? As mentioned some of the tests fail with this, so I suspect a logic error somewhere. (There may be some timing assumptions, that will eventually cause problems in the wild.)

[anacrolix]

With #248 I can no longer reproduce the other failures I found.