Provide takes ~2.5 minutes

[Stebalien]

Worse, it provides in two steps:

1. It finds the closest peers.
2. It finally sends out provider records.

That means one can't even reasonably set a timeout to perform a best-effort provide. We should probably set a shorter timeout on GetClosestPeers, reserving ~5s to send actual provide records.

(I mean, we should try to make GetClosestPeers faster but...)

[tobowers]

This is interesting. We have a 7 node system doing 75k block adds (quickly) and it's using up all the CPU on most of the nodes. It seems this dies down after the 36h timeout, but continues on until then unbounded. Could this be the reason?

[Stebalien]

Probably.

[Stebalien]

It's now taking 30-50 seconds and keeping this issue open likely isn't going to help.