The linked bug show another user successfully saying "please opt me out", and Google building the feature to do that in a week.
Drew has for some reason chosen not to ask for an opt out, even though it appears trivial and would probably be fixed by the weekend if he asked for it.
It seems like this Git hosting service's workload-per-Q has increased due to the Go proxy change, which (IMO) is a cause for concern.
If that's the root of the issue though, most of the article ("no one will get promoted for prioritizing that at Google.", "the go team has not prioritized it", etc.) is wrong. They may not have addressed the issue in the precise way he wanted, but I also think it's rather unreasonable to expect someone to not only address your issue, but address it in the precise way you suggest.
That's rarely globally optimal and feels rather entitled.
Maybe they should request 500 times a day because that's what the proxy is doing.
This same question comes up for ~any CI service or any number of other services that regularly build from source. There's a reason that github maintains a cache and that repo clones are served from that instead of by invoking the git command on the server.
A git hosting service that can't handle a trivial clone qps isn't actually a git hosting service.
Also traffic is not free, so even if they maintain a cache somewhere there is a cost to this.
Ok so let me answer your previous question more explicitly:
> So the solution is that everyone who does not want that behavior should request that they turn of refresh traffic for their domain?
Yes.
> How long would it take until they would complain about people spamming their issue tracker if everyone who self-hosts a go module would start requesting this?
We have an answer to this. It is, empirically, a very long time, as there seem to be a handful (possibly only 2, certainly less than 5) users impacted negatively enough to report an issue. It does not usually make sense to solve an issue that doesn't exist. Presumably dozens of users reporting the same issue would cause a reprioritization, but if no one is complaining, there's no reason to change priorities.
You have enough information to understand that it is a problem, regardless of how many people are actually complaining to you. I self host git, if Google was chewing up 120gb a month of my bandwidth I'd definitely feel it but I wouldn't figure out what was going on easily, let alone figure how how to complain and stop it.
The behavior is rude and bad engineering. Just because people aren't yelling at you for leaving dog poop in their yard doesn't mean it isn't a problem for them too.
The other user who accepted that proposed fix had a single module that had a single user, so they can tolerate some weird caching behavior for their toy project.
There comes a point (and it isn't a particularly far-off point) where disabling this would result in significantly increased traffic to sourcehut. That's a problem, but it's a problem with sourcehut's scalability.
They mention that requests still are served from the cache when refresh is disabled.