IPv6 support for cloning Git repositories

[aparcar]

Hi, I'm surprised I didn't find an existing discussion with this topic. Some services like github-releases.githubusercontent.com or user pages do support IPv6, however the webpage (github.com) itself, including cloning of git repositories, does not work.

Is there a public roadmap on enabling IPv6 for GitHubs very core business, distributing Git repositories? If I'm wrong and there is already IPv6 support, please guide me.

The same issues exists for api.github.com and thereby making CLIs unusable on IPv6 only connections.

EDIT: I left this topic on "Subscribe" to receive notifications if GitHub ever responds, after 3 years I lost hope and will unsubscribe now. I'm sure I'll get the news somewhere else - if IPv6 on Github it ever happens.

[ffrediani]

I am unable to understand what in 2022 still holds full IPv6 support for a platform like GitHub.
Even if there is already a work in progress for this it is more useful to disclose a proper ETA rather than a "we are working on it" with no ETA.
Thanks

[ghishadow]

https://www.githubstatus.com/incidents/5y8b8lsqbbyq

This incident was the result of an infrastructure change that was made to our load balancers to prepare us for IPv6 enablement of GitHub.com. This change was deployed to a subset of our global edge sites.

looks like they are preparing for IPV6

[pmarks-net]

We have so far identified three remediation items here:

• Update the IP Allow List functionality to handle IPv4-mapped addresses.
• Audit the rest of our stack to confirm there are no further places this IPv4-mapped IPv6 addresses flaw exists.

At OtherBigCompany, the networking libraries would convert IPv4-mapped addresses to plain IPv4 immediately after reading from the socket, to avoid this class of problem. Ideally nobody should see an ::ffff:0.0.0.0/96 address unless they're working on socket code.

[Rifahad]

yes of course

[davidmi4268]

Yes that's true

[Leseratte10]

No. Nobody needs assistance in understanding how IPv6 works. Please stop this spam, as I already said in my comment two weeks ago: https://github.com/orgs/community/discussions/10539#discussioncomment-11684249

[ayubio]

New ISPs in my country are IPv6-only because there is no new IPv4 space to be provided to them. They do have a over-shared IPv4 address by CGNAT but due to the oversharing, it is unstable and not rare to be offline. For these companies, the internet access is stable only in IPv6.

Thinking about the server-side, some cloud providers are making extra charges for IPv4 addresses (e.g.: Vultr.com) so most of the servers in my company are IPv6-only. Cloning github repositories is very cumbersome due to the lack of IPv6 support and this issue affects me and my team mates on a daily basis.

The math is simple: there are 4.88 billion internet users in the world but the IPv4 space only provides 4 billion addresses. It's over: IPv4 is obsolete and is provided in a legacy mode. Current applications and services must be IPv6 enabled otherwise it should be seen as obsolete. For that matter, Github.com is an obsolete service because it relies on obsolete technology as IPv4.

[treysis]

They'd need a NAT64 gateway to reach github not a tunnel as they already have native IPv6.

[divinity76]

what country is that?

[Malix-Labs]

@divinity76 Brazil

[14KBZFxd]

In a CGNAT scenario, it is easy to trigger a limit rate based on ip authentication.

[Leseratte10]

Can we please stop with these "what specific issue do you have due to lack of IPv6". There's no "help" that people operating IPv6-only servers need.

The issue is you need to pay for IPv4 addresses for all your servers to reach Github. Tons of backend-only servers don't need IPv6 except for accessing Github.

[DanielDamito]

I really tried to use github in a IPv6 Only Network and I was not successful.
We are not talking about a new feature request or a little detail, we are talking about the current standard protocol of the Internet defined by IANA.

Is there any plan on the github's roadmap to fully adopt IPv6?

[NiKiZe]

If GitHub can't get v6 on GitHub.com soon, maybe at least an ipv6.github.com proxy for SSH git cloning?

[andrediashexa]

IPv6 is the actual internet protocol, while IPv4 is a legacy protocol. Please, priorize this request.

[maxadamo]

[NiKiZe]

GitHub SSH is probably a custom implementation, and even if not the authentication and logging, as well as all tools that is reading and searching in those tools needs to be tested.

I think there is no excuse to not have IPv6, but at the same time I atleast understand that it is not just a "frontend reconfiguration"

[NiKiZe]

Please try it, I still think that they have issues with client IPs in logging that is causing the major delays.

[maxadamo]

@aparcar in the end I decided to remove my messages because I was not very keen to reply to the non-sense.
A service like this one proves that Github doesn't need to re-engineer its "smart load balancing" (as you call it): https://danwin1210.de/github-ipv6-proxy.php
And if the "smart load balancing" works for the script kiddies, then it works with all the trimmings with an enterprise-grade solution.
p.s.: I'm not saying that @DanWin is a "script kid", but he has created a very simple solution with 19 lines of code.

[bjmgeek]

I'm sure they're doing some smart load balancing which could make it harder to migrate, surely there is more involved than a single IPv6 checkbox.

A load balancer is at its core just an http proxy. There's no reason it can't listen on ipv6 and make requests to ipv4 backend hosts.

[michaelmior]

A load balancer is at its core just an http proxy.

Not all load balancers work at the HTTP(S) level. Although it is probably true that whatever load balancer used could be made to work with IPv6.

There's no reason it can't listen on ipv6 and make requests to ipv4 backend hosts.

There are solutions to this problem, but one problem is logging and abuse detection. I think it's likely that some methods currently in place rely on IPv4. Of course that doesn't mean it's not possible to adapt to IPv6 as well, but not necessarily trivial.

[angerman]

My private ci is forced to go full IPv6 only, and this requires me to have one IPv4 gateway to access github. This in turn means I keep running into rate limits all the time. For now I've worked around this with an access token, but that's not sustainable.

Any ipv6 support would be much appreciated.

[BroOtti]

Is GitHub deprecated or why there is still no IPv6 support? We are talking about a over 20 year old technology and the standard for about 5 years.

[MrLawrence]

No, no. It's apparently just more important to rename the "master" branch to "main".

[9mm]

@MrLawrence no truer words have been spoken. What an absolute joke

[Leseratte10]

Yes. Multiple people moved their projects to Gitlab because they want to access them from IPv6-only machines.

[ffrediani]

Probably because people that manage the technical team still didn't realize what this mean and didn't prioritize it enough.
Every time I get as an answer that something like "there are many complex things that must be taken into account and it is not easy to adust everything to make it happen. It is in the roadmap"

Well, then why hasn't all that been done beforehand and being carried out gradually that we can some progress on it ? Are there people working dedicated to make that happen ?
The lack of these answers makes it look like that is being treated as something minor or less important.

After companies like Google, Facebook, Netflix, Akamai, Cloudflare have made it 100% I don't see any other strong arguments for companies use complexity as a reason to delay it further.

Hope someone from GitHub's team is reading it.

[dwmw2]

It's kind of weird because RFC1883 was published in the final decade of the 1900s, over a quarter of a century ago. That was long before git was invented; let alone GitHub.

[NiKiZe]

Totally agree! Defined as current in RFC8200

[BroOtti]

Imho IPv6 is a lot easier than IPv4. Also it's not that big difference from IPv4. Big companies just lazy on setting it up. It's just a shame to see services still not available with IPv6.

[xtaran]

I vaguely remember that years ago (pre-MS, pre-pandemic) somewhere Github engineers held a talk (or maybe it was a blog posting?) which more less said that IPv6 is in the makes, but less trivial than one might think. I just can't find that talk or slides or so anymore. I also vaguely remember that they mentioned ipv6.github.com — which exists, but is just a (probably wildcard-) dummy page which just points to https://ipv6.github.io/ (likely also a DNS wildcard) which then just states that there is no content yet. Probably because it (nowadays?) belongs to the Github user @IPv6. And the Wayback Machine only has a record from 2022 of that site. Maybe someone else remembers where that talk or blog posting was published.

What I though have found is that there is an ipv6 label in Github's blog — it though only lists one posting so far, which talks about Github Pages now having IPv6 support.

There even once was github-ipv6.com which worked as a reverse proxy according to this Reddit comment, but since it no more exists, I guess they got a DCMA takedown or cease-and-desist noticed for using the Github trademark.

But yeah, another Github user here with (on purpose) trying to run hosts IPv6 only and the first (and so far only) hard stumbling block was not being able to clone Git repos from Github. 🤌

I wonder if I should use Gitlab.com for these repos, because they do have an AAAA record for their main site:

→ host -t AAAA gitlab.com
gitlab.com has IPv6 address 2606:4700:90:0:f22e:fbec:5bed:a9b9

[MeesJ]

It's 2022, World IPv6 Launch Day was 10 years ago. Yet, GitHub still doesn't have IPv6 support. The IPv4 address space is exhausted for years now, and ISPs are using techniques such as CGNAT to still be able to give their customers access to the legacy IPv4 internet, with the instability of these techniques as the cost.

Why doesn't GitHub provide native IPv6 support? And, more importantly, is IPv6 support for GitHub on the roadmap?

[danielehrhardt]

Biggest fail i ever saw.

[TheBlueMatt]

CC rust-lang/cargo#10711 this causes real issues for open source software users. This is an absurd conversation in 2022.

[pmarreck]

Try 2024... What's the over/under on Github having IPv6 support in 2026?

Also, WTF, github? I'm guessing all the greybeards who actually built the thing are long gone and the generation that is there can barely touch anything without something breaking.

Ever hear of "code smell"? Not supporting IPv6 in 2024 as the premier developer SCM website is a massive "org smell"

[lalelunet]

This is $MS. This company was and will be ever a enemy of open source or new technologies that comes not from $MS. They did nothing that helps the community. $MS is only interested in earning money and gives a shit on your needs

[sush-abc]

How can you not have ipv6? Some cloud providers charge extra for ipv4!

[pmarreck]

I ran into this with Hetzner last week and I'm still on the journey to get IPv6 working across all my internets. And then I run into this Github stupidity. Unbelievable.

[TeaDrinkingProgrammer]

Same here

[madalinignisca]

Would any Github/Microsoft representative tell us in which year, century or millenium will they support IPV6?

We have public cloud environments where we are with IPV6 only already...

Although, based on recent experience with Azure, I think their public cloud environment is also like with 10 years behind Google and Amazon's public cloud anyway. Why would they bother for Github?

[tonymet]

thanks chatgpt

[ffrediani]

Hopefully ? There is not hope.

The time it is taking to someone even give any proper attention to it shows the bad lack of technical management that is behind GitHub. We are not talking about a simple cosmetic issue.

[symgryph]

Sadness, shows MS lack of vision and slowness AI AI AI

[queenofcorgis]

@sufal54:

We’ve clarified our stance on using generative AI tools like ChatGPT within our Community via this announcement. Please review the guidelines to ensure your post meets them as failure to adhere to those rules can result in action taken by our moderator team.  You can read our updated Code of Conduct and the announcement for more details. Thank you for helping us maintain an authentic and beneficial space for everyone.

[mh4x0f]

2025, GitHub still doesn't support IPV6.

[angerman]

I wonder if we had issues with Finland getting cut off from GitHub as well if we had IPv6 connectiity to GH.

[chrlns]

Consider migration to Codeberg.org? ;-)

[fourhundredfour]

Consider migration to Codeberg.org? ;-)

Wrong place to advertise another platform. This doesn't solve the problem that Github doesn't support IPv6.

[fsa]

Consider migration to Codeberg.org? ;-)

You're missing the core issue. Even if we migrate our repository to another hosting platform, all the build dependencies required for the project will still remain on GitHub. We'll still have to implement hacky workarounds to make deployment work properly.

[mrmelon54]

Kind of funny how the most popular source code hosting platform doesn't support IPv6 when it is over 25 years old.

[hevisko]

;(

[alexvancasper]

Hey GitHub,
I have tunnel broker service and I can give you ipv6 prefix absolutely free with huge internet channel!

[31KM]

Still not supported :(

[altf4arnold]

Microsoft azure also rents IPv4. Ironically, Microsoft makes money of not implementing IPv6 on Github. This might be the reason why they don't. They get paid not to.

[altf4arnold]

According to Chandler Harris in 2011:

The acquisition by Microsoft last month of 666,624 IPv4 addresses from Nortel for $7.5 million, or $11.25 per address, highlighted what could be a reality for businesses looking to expand their IPv4 in the coming months or years. While IP addresses are currently free--for those companies that show a need, which Microsoft could not--they may be costly once the free supply expires by the end of the year.

In other words, we need to move from IPv4 to IPv6 to avoid the kind of business that benefit from resource scarcity.

Absolutely. But that's why microsoft is acting against us by not putting v6 on github. People won't move away. They win anyway. (or you can buy their solution in azure that hosts git for you in dual-stack). At this point, I won't be surprised if it's ready but just held back

[mrmelon54]

Interestingly raw.githubusercontent.com has IPv6 support. I assume they don't want (or are in the process of) converting the main GitHub software to support IPv6.

[rnhmjoj]

Github pages also has IPv6. They were also experimenting with IPv6 in the main website a couple of years ago (it was mentioned in an incident report).
I guess they tried migrating, something broke, so they rolled back and haven't done much since.

[4920441]

Github pages also has IPv6. They were also experimenting with IPv6 in the main website a couple of years ago (it was mentioned in an incident report). I guess they tried migrating, something broke, so they rolled back and haven't done much since.

Think about "experimenting" with the public depolyment of IPv4 on the main webseite...... I mean.... IPv6 ist noch the newest stuff any more, its proven, solid and working since nearly 30 years now...... I mean.... migrating it to ipv6 should only be a matter of time nothing else..... maybe manpower, but I think they still have enough.....

[mrmelon54]

Migrating to IPv6 should have been done 15+ years ago. With IPv4 backwards compatibility via reverse proxies or dual stack.

[maurya-07]

[Ernest314]

What do you get out of pasting a ChatGPT answer? Does it make you feel smarter? Or do you genuinely feel like you're helping out? Do you really feel like you've added to the conversation?

I keep seeing comments like these more and more and my personal curiosity must be satisfied

[maurya-07]

I deleted the comment, but just to clarify! I only paste a ChatGPT answer when I feel it’s actually useful and can save someone’s time. It’s not about looking smarter, it’s just about sharing something that might help. If I have my own thoughts to add, I include them too. Sharing a good explanation that’s already written doesn’t hurt anyone.Sorry if you find something wrong..

[NiKiZe]

"doesn't hurt anyone" you posted duplicates of what was already said, wasting everyone's time trying to figure out if it has anything new in it. During that time there could actually have been lives saved.

We need AI to remove tasks like mundane reading, NOT add more text to read, parse, and understand.

[pmarreck]

@Ernest314 To be fair, I’m sure there are non-native-English-speaking users who use it to sound more professional in English, so I wouldn’t jump to conclusions that someone is just being lazy.

I would also start to get used to seeing more of this, because learning new languages is very difficult and time-consuming and AI actually does an excellent job of it

[levsha]

If anyone would want to read an AI-generated summary they could just go to ChatGPT directly - no need to be a proxy.

[tonymet]

I've thought quite a lot about github's motivations. They do face a unique challenge with ipv6 as an infrastructure company. The moment they publish AAAA , they will receive a massive amount of ipv6 traffic -- 50% of the internet by everyone's estimates. I don't see how they can canary or AB test this config like most other rollouts.

What this means is that github will have to deploy ipv6 across their entire stack. They can't just roll a NAT64 , a proxy or other bodge like most of the n00bs in this discussion are proposing.

Given we have so many bright engineers here. Instead of shaming github -- why don't you propose how they can deploy ipv6 responsibly. That is , how could they gradually roll it out with minimal resources and risk to their primary github.com service.

It's not a rhetorical question. Think about it.

[arnowelzel]

"just publish a new endpoint like dualstack.github.com (what aws did)" -- this is a bodge and a liability. For one, you'll never canary more than 5% of your traffic, so you're only delaying the inevitable. and now you have 2 service endpoints to support forever.

No - not a new endpoint! The IPv6 traffic could go to github.com, but this is may be a proxy which will forward requests to the IPv4 only infrastructure behind. But for SSH this is not as simple, yes.

Otherwise I agree to the others here - there are no technical reasons why IPv6 should not be possible.

I assume Microsoft just does not want to spend the money and they fear to give up existing security measures which are also based on filters to block "bad" IPv4 adresses. That's also the reason, my Microsoft mail servers also only accept IPv4 like outlook-com.olc.protection.outlook.com which is the MX for many domains there. Blocking "bad" IPv4 addresses or whole address blocks of ASNs is much easier than dealing with IPv6 to block spammers based on IP adresses.

My personal solution for this: I moved away from GitHub for my own repositoroies and migrated to Codeberg (yes, including all existing issues, PRs and actions which is supported quite well) - problem solved for me.

[tonymet]

Otherwise I agree to the others here - there are no technical reasons why IPv6 should not be possible.

Yes it's possible but more costly than most of the proposals here because Github can't just use off the shelf LBs and Proxies for HTTP traffic. nat64 won't scale for them because they would need double the infra. They likely have to support ipv6 e2e before exposing to customers. Canarying IPv6 with a proxy wouldn't actually test the final system.

And the benefits to them are minimal. Honestly the benefits for most people are not that great. Hence ipv6's slow adoption over 30 years compared to nearly every other network upgrade--even hardware upgrades.

[regalialong]

Honestly the benefits for most people are not that great.

Tony I swear to god, unironically, how much is Github paying you for astroturfing?

There's testimonial over testimonial how users are migrating to competitors, and those that still bother (despite the service not giving a shit about letting them connect for +3 years) have to use sketchy third-party services that are asking for a supply chain attack to happen.

This doesn't stop being a concern in 1-5 years (more and more cloud services are not providing v4 addresses for free anymore), how do you settle the """cost""" question when you're actively hemorrhaging users?

[regalialong]

I assure you no executive questions themselves on how much the infra to power the latest LLM thing nobody wants costs, but the moment IPv6 gets brought up, the cost to value needs to be carefully weighed to make sure the delicate indie company (you know, one of the largest and most influential tech companies) doesn't collapse.

Setting up new infrastructure is not rocket science, IPv6 deploys aren't bleeding edge things (fuck, Github Enterprise Server can do it), there's people at Github who regularly get paychecks to plan this out.

Quite simply, nobody who can assign this priority at Github cares.

[arnowelzel]

This isn't shaming, it's customers

"customers" ?

Are you claiming the DNS workings is different for SSH from Https?

git+ssh traffic is different from https. it's a stateful socket service . doing nat64 will be double the resources and errors

JFTR: The company I work for is a paying customer (more than 1000 USD per month). And they also don't have IPv6 available.

As I told that I moved to Codeberg - this is only the case for my private repositories which are open source.

[tonymet]

I assure you no executive questions themselves on how much the infra to power the latest LLM thing nobody wants costs,

That's just business. IPV6 migration won't be on the quarterly report. No ones promo will depend on it. I hate the LLM products as well, but wall st likes them, so companies are dumping billions into them. To steel man the argument, there is potential tremendous upside on the scale of Google's entire market cap . In the best case , 10% of their customers might want ipv6 , and Github can just comp them for the ipv4 gateways for 1/10th the cost and none of the risk of the migration.

I'm not astroturfing. You can look at my github that I have pretty decent bonafides with ipv6 migrations. So I think I'm allowed to speak to its shortcomings -- of which there are many. Even some of ipv6 designers complain about it, and many developers, which explains the sloth-like adoption.

My position in this thread is that we should be offering help with the migration rather than shaming them with cliches like "it's been 25 years" and "spin up a nat64 proxy" .

[arnowelzel]

In an earlier comment you where talking about 50% of their traffic

paying customers needing ipv6=10% (closer to 3%) , "free" internet traffic = 50% once they turn it on. Please do try to pay attention

Ah - "customers" is by definition "paying" for you. If someone just uses GitHub as a user - even if he pays for sponsorships etc. - he is not a customer.

And your point is, that only paying customers need IPv6?

[tonymet]

In an earlier comment you where talking about 50% of their traffic
Ah - "customers" is by definition "paying" for you. If someone just uses GitHub as a user - even if he pays for sponsorships etc. - he is not a customer.

I'll explain "dollars" since you think MS can just dump billions of dollars on IPv6 because they have a large market cap.

1 customer generating $20m in ARR has 2e7-X the influence of someone donating $20 to their favorite open source developer (assuming github keeps $1 fee). And in practice, the difference is even greater than that, because the paying customer has an account manager they can call up and nag.

Ranking of Influence on the IPv6 Decision

1. someone in leadership who cares and can show $200m + revenue in 5 years

2. investors at big banks who think it will generate $1B+ revenue in 5 years

3. a senior eng at MS who needs a pet project and can show it will generate / save $200m+ in 5 years

4. a few hundred customers paying $1m or more ARR where ipv4 is costing them > $50k / yr

5. a golfing buddy or burning man acquaintance of group (1) who thinks ipv6 is cool
   ....

6. "open source" developers shaming MS on the internet because "IPv6 is 25 years old".

Remember , "ipv6 is 25 years old" just reminds them that it's not that important.

[mrmelon54]

If Microsoft/GitHub wasn't to still have no v6 in 10 years time, well that's on them. I expect them to be figuring out bugs and glitches in their v6 network when other companies have had fully working v6 networks for 20+ years. After seeing the various bugs in Microsoft's v6 stack in the Windows NT kernel, I expected them to have many issues running v6 networks.

The difference now is that according to some sources 40-50% of global internet traffic is IPv6, along with v6 being almost 30 years old. This shows that it is the new standard and lives up to its purpose.

IPv6 isn't about revenue anymore, it is about existing on the Internet for the foreseeable future.

[rnhmjoj]

I think you're mistaken: I doubt there's anyone else here that think publicly shaming github for not implementing IPv6 is a way to influence Microsoft decision making; no, it's simply to have a laugther at how bad they are at being a tech company.

[regalialong]

it's simply to have a laugther at how bad they are at being a tech company.

Yeah, I've started calling this discussion a group therapy session with how many people come here to vent the exact same (solvable, if given priority!) frustration. ^^'

[guipoletto]

Bumped into this today!
After a LOT of head-scratching, since ipv6-first has been standard in my mind for so long,
I hadn't even considered that a 1st-tier platform such as github would not support IPV6. just didn't even cross my troubleshooting mind

Now I have to either mentally reclassify github as a second-class service, or go back to thinking like it's 1996. (can i have my hair back?)
I want my hair back!

[altf4arnold]

Yup. At this point, github is the cute public front for stuff. For actual prod, we need other solutions

[depeo]

Seriously, just fix this, even if you don't support IPv6 internally, at least you should be able to set ipv6 addresses on your frontend/load balancers.
Just use split DNS internally if necessary.

[daniel-fahey]

This should be on the front page of Hacker News every day until Microsoft gets off its fat cat arse and fixes this!

[tonymet]

I tried posting to hackernews and it was flagged as advocacy or something against TOS / community guidelines

[daniel-fahey]

This got fairly big a few years ago https://news.ycombinator.com/item?id=32579914

[depeo]

Its a bit annoying, a lot of containers/VM:s I have would work great on IPv6-only, but it always fails because github requires IPv4.

I could of course mirror all repos I need, but that shouldn't really be neccesary.

[Malix-Labs]

I'm using https://gh-v6.com/ because GitHub and Microsoft seem incapable of addressing this

Insane stuff

[5NineN]

IPv6 should be a given. Github is no longer useful for me, as many of our systems are IPv6 only. Thanks for the amusing workaround.

[jgrg]

Maybe the announced move of GitHub infrastructure to Azure will give us IPv6 support as a side effect?

[miyurusankalpa]

Probably not, GitHub uses azure IPs even now for all regions except US and EU(they have own DC in those two).

If they could, they would have done it already.

host github.com
github.com has address 20.205.243.166