What does your ideal git host do when one of its (non-paying) clients uses bots to clones repos so hard that its (paying) human clients are unable to clone their repos?
Improve infrastructure. Apparently they are trivially DDOSable by anyone renting a few servers and running git clone in a loop? That's a problem they should solve. And no, blocking by user agent is not adequate protection.
You're not paying for the traffic load though, Microsoft does. I think it's unfair to compare a forge run by a giant and one run by a tiny company, or for that matter, self hosting users.
- let’s round and say 2000 requests at 1gb = 2000 gb/hour = 48000 gb/day
- AWS bandwidth at $0.02 / gb = $960 / day = $28,800 / month
So, one of the richest companies in the world is charging you nearly $30k monthly because they cannot be bothered to be polite. Would you be ok with that situation?
From what I can tell, per Drew's own characterization of the requests[0], this is wildly off in all aspects. Most repos are hundreds of KB. Large repos are a dozen MB. Per the other guy's characterization[1], his repo was less than 8MB, right in line with what I see in other repos. He saw 4GB of traffic split up across "more than 500" requests. And as Drew said, these bouts of activity aren't sustained even across an entire hour. To top it off, the cost of AWS bandwidth is about the least representative example you could have picked. You've wandered off the path at nearly every available opportunity.
