4 ms·
If it were me, and I wasn't willing to just block the traffic, I might just set a 128 kbps limit on it and call it a day[1]. Eventually, the other side will fig
by toast0 4y ago
If it were me, and I wasn't willing to just block the traffic, I might just set a 128 kbps limit on it and call it a day[1]. Eventually, the other side will figure out that their fetchers are all backed up and work out how to do their job without burning so much bandwidth.
[1] Yeah, that can be a bit of a pain to setup depending on the server settings, but some people have to pay for bandwidth and server resources, so it's probably worthwhile.
- WhyNotHugo 4y agoKeeping the rate-limited connection alive eats up resources. In this case, the organisation executing the DDOS attack has virtually limitless resources.
- mpol 4y agoI guess this is the professional response ;) I would just blacklist the IP addressess, just to see what would happen.
- toast0 4y agoYeah, but that's not an option according to the article. > I can’t blackhole their IP addresses, because that would make all Go modules hosted on git.sr.ht stop working for default Go configurations (i.e. without GOPROXY=direct)
- CryptoBanker 4y agoThat’s more of a “won’t” than a “can’t”
- eatonphil 4y agoIt would mean that his host is not an option for any Go developers packaging libraries. So it's not really an option for their business, I'm guessing.
- biorach 4y agoIt would mean his service would be useless for a significant portion of end-users. Not really an option for a site that wants to stay relevant.
- sangnoir 4y agoA source code repo can't block access to code it hosts for a major language and maintain its user's goodwill. "cant'" sounds right.
- spicybright 4y agoYou could easily proxy around it. It's inconvenient to set up, but it can definitely be a set it and forget it thing.
- coldacid 4y agoYou can set GOPROXY=direct in your environment and not have to deal with Google's scumbag proxy service in the first place, but nobody seems interested in making this the default behaviour.
- ShowalkKama 4y agoOr you could mess with a random percentage of the requests: tarpit them, drop random packets, reply with malformed answers etc. If you keep the percentage low they might have a fun time debugging :)
- toast0 4y agoIf the rate limit is high enough, and pooled across all matching requests, you'll get different effects depending on the (global) concurrency, which might be tricky enough. And it might lead people into the right direction anyway.
- chitowneats 4y agoSeems like there's no need for this. HTTP spec covers "too many requests" already. Why further antagonize things rather than clearly state the problem?
- qwery 4y agoWhile amusing, responding in this way seems overly antagonistic/escalatory. i.e. the behaviour of a poor citizen of the web. It also would be likely to result in an increased number of requests? Drew's actions so far (as far as I can tell) are not only inoffensive/reasonable but overtly good-willed and presumably result from a considerable amount of effort.
- xigoi 4y agoAs if Google's response (that they, the largest tech company in the world, can do nothing about it) wasn't antagonistic…
- klodolph 4y agoThe rate limit is unlikely to cause a problem. Google has been crawling the web since its very start, and the internal services which fetch resources from external web servers are extremely resilient. Some request fails? Some request is slow? It's not going to slow down other requests. Maybe these services aren't being used for Go, but the expertise is on tap. (These kinds services are also supposed to rate-limit their requests.)
- toast0 4y ago> The rate limit is unlikely to cause a problem. If that's the case, then great. It solves the DDoS issue without impacting the Google side, and everyone can be happy.
- verdverm 4y agoThis is completely different architecture and services, largely open sourced. This is limited to Go and the dev team. Most likely these errors would manifest down to the users of Go and those who've a dependency which lives on sourcehut
- klodolph 4y agoA big chunk of the Go core maintainers are Google employees, and they run Go services on top of Google infrastructure.
- verdverm 4y agoBeing tied to Google has very little impact on the technical implementation of go modules fetching, proxy, & sumdb. It is intentionally isolated so that request data is not part of Google's data business. Go uses git under the hood for these requests, so there are more complications. How do you cache or respect robots.txt when you don't actually talk to the servers?
- stefan_ 4y agoYou would think so, but has someone made a zip bomb with Git packs yet? ;)
- tmp_anon_22 4y agoIn theory what could Google SREs do to get back at OP for messing with their services? Could they be banned? Blacklisted?
- mrighele 4y agoWhy not just return a 429 (Too Many Requests) if the specific repo has been requested by google not too long ago ? (e.g. 1 hour earlier or). It's a standard response code, and with a bit of luck google will scale the requests accordingly. If not, this will still allow the proxy to operate properly without burning too much server resources. (I understand that this may leave some customers unhappy since the proxy may be up to 1 hour stale. If that's the case, the server could also add a check to see if there have been pushes since last request, though this is quite more complex)
- quadrifoliate 4y ago> It's a standard response code, and with a bit of luck google will scale the requests accordingly. Google already says in the comments [1] that it would be "a fair bit of work" for them to read the standard robots.txt for "boring technical reasons". I would not necessarily rely on them to respect HTTP 429 either. In fact, given the facts at hand with the current situation, I would guess that they probably wouldn't handle it correctly, and would cite more "boring technical reasons" when asked. ---------------------------------------- [1] https://github.com/golang/go/issues/44577#issuecomment-1137883692 https://github.com/golang/go/issues/44577#issuecomment-11378...
- mro_name 4y ago> "a fair bit of work" for them to read the standard robots.txt for "boring technical reasons". that's embarrassing. What other reasons should there ever be? Stop taking their money, stop using their shit. Turn your attention to where it's beneficial.
- opmac 4y agoReturning a 429 would not just to hope that the Go proxy would obey the response, but rather it would significantly reduce the workload the server has to do when it gets a request it doesn't want to serve.