GitHub Incident
109 points
9 hours ago
| 12 comments
| githubstatus.com
| HN
bakje
9 hours ago
[-]
Perhaps the gemini-cli bot arguing with itself is taking its toll

https://github.com/google-gemini/gemini-cli/issues/16750

reply
MattIPv4
8 hours ago
[-]
https://github.com/google-gemini/gemini-cli/issues/16723 is even worse, GitHub shows `5195 remaining items` in the collapsed timeline.
reply
DeepYogurt
7 hours ago
[-]
Wow. If you look at all the issues this seems pretty common

https://github.com/google-gemini/gemini-cli/issues?q=is%3Ais...

reply
unaut
8 hours ago
[-]
Wow that's whole a lot of yapping
reply
lol768
8 hours ago
[-]
Jeez, what a mess. Some of those issues have over 5000 events on them.

I really hope that didn't send emails out to people.

reply
pdimitar
8 hours ago
[-]
I could not resist to put my sarcastic comment about RAM price increases serving a good cause in there.
reply
dgxyz
8 hours ago
[-]
Having just had to buy 4TB of RAM, I appreciate this.
reply
MisterTea
8 hours ago
[-]
That's like 100,000 USD. I keep thinking about making a rap video wearing a 10 TB gold chain surrounded by big booty girls with their naughty bits covered in m.2 SSD's while dissing the AI industry. Though I cant afford the RAM :-/
reply
zxcvasd
8 hours ago
[-]
like most rap videos do with cars/jets/mansions, just rent the ram sticks for a few hours!
reply
TheJoeMan
8 hours ago
[-]
It’s sad that I can’t interpret if you mean to actually shoot your rap video on film, or have an AI generate it lol. Either way you’re going to need RAM.
reply
kps
6 hours ago
[-]
Shooting on film doesn't need any RAM. Unfortunately the price of silver is also through the roof.
reply
dgxyz
8 hours ago
[-]
Yep that much. 64Gb DDR5 ECC sticks (128Gb don't exist at the moment apparently). They declined the PO 6 months ago. That'll teach 'em.

I was pissed that there weren't any sticks heading to the recycling out of the nodes otherwise I would make myself that chain :)

reply
dpacmittal
6 hours ago
[-]
Use Veo
reply
omoikane
8 hours ago
[-]
Maybe the bots need rule of ko.

https://en.wikipedia.org/wiki/Rules_of_Go#Ko

reply
embedding-shape
8 hours ago
[-]
Haha, reminds me off bringing down office mail servers by accidentally creating loops of emails back in the day... What is old is new again, but this time with probabilities :)
reply
johnisgood
9 hours ago
[-]
Wonderful, lmao.
reply
nullfish
9 hours ago
[-]
I suspect the migration to Azure is continuing to go well
reply
ascendantlogic
9 hours ago
[-]
This feels more like Copilot-as-platform-engineer to me
reply
DeepYogurt
8 hours ago
[-]
Github's been running on vibe code for a while now and it's starting to show
reply
rvz
9 hours ago
[-]
Yes indeed. 6 years of non-stop outages across the platform every month.

Even self-hosting would have been more stable than sitting on GitHub as predicted more than half a decade ago. [0]

Now there is no 'CEO of GitHub' to contact this time (Satya does not care).

[0] https://news.ycombinator.com/item?id=22867803

reply
someguyiguess
9 hours ago
[-]
I did not come to hacker news expecting comedy gold but you have done it my friend!
reply
corvad
9 hours ago
[-]
Github's recent reliability has honestly been abysmal. Not surprised.
reply
ferguess_k
9 hours ago
[-]
Unless some major customers are moving away, I don't think they are going to seriously care about it.
reply
corvad
8 hours ago
[-]
I suspect some companies may already be considering it. Especially with the wealth of alternatives today.
reply
g947o
8 hours ago
[-]
Companies are already using on-premise GitHub server, if they are using GitHub in the first place. There are many other self hosted solutions which are quite common in enterprise environment.
reply
supriyo-biswas
8 hours ago
[-]
In my experience companies are moving into GitHub for Copilot and GHA.
reply
appplication
8 hours ago
[-]
GHA maybe, but copilot is just another mid tier player in a congested space.
reply
ChromaticPanic
1 hour ago
[-]
It's a cheap mid tier player. You get more tokens per dollar.
reply
NewJazz
6 hours ago
[-]
Doesn't stop folks from wanting to buy the MS brand. Execs are really out of touch these days.
reply
nine_k
8 hours ago
[-]
What kind of alternatives do you see as viable for large(ish) commercial users?
reply
toephu2
8 hours ago
[-]
GitHub on-prem. Officially called GitHub Enterprise Server. You can have GitHub, but hosted on your own servers.
reply
NewJazz
8 hours ago
[-]
So you still pay them, you do the hosting work, and you get a product with worse features than gitlab?
reply
zxcvasd
8 hours ago
[-]
but you can be smug when theres a github incident, and thats hard to put a price on
reply
NewJazz
6 hours ago
[-]
You can do that with gitlab.
reply
pxc
7 hours ago
[-]
What is the quality-first, high uptime alternative to GitHub? My employer uses both GitHub and GitLab, and while I think GitLab is better, its quality also frankly sucks. It's riddled with bugs that have just been marinating on the issue tracker for years, and the most common "fix" for gnarly bugs in the CI platform is "revise the documentation to reflect the existing (broken) behavior".
reply
ferguess_k
4 hours ago
[-]
*Stupid question*: What is so hard about self hosting one's own repo? I get it must be difficult for a mega corporation, but for companies like us, who have hundreds of repos but only 20 of them are regularly used, and concurrent read/write is relatively light -- considering our largest team is less than 20 persons, so even if all of them are reading/writing from the repo, it doesn't seem to be a huge issue.

Even for a bigger company, say 5x developers (we have about 100+ SWEs and maybe 10-20 other titles who use GitHub), is it really a big thing to self host their own repos? External applications are definitely on another level because you could have hundreds of concurrent visits easily.

What did I miss?

reply
pxc
4 hours ago
[-]
> What is so hard about self hosting one's own repo?

Maybe nothing! I was genuinely asking. I still don't know what Actually Good™ forges are out there these days, generally suitable for corporate use in place of the likes of GitHub or GitLab. Forgejo? Something not based on Git?

reply
stefan_
6 hours ago
[-]
It's amazing, before we even had ChatGPT, GitLab was building so much endless slop halfbaked crap in their pursuit of ever more "enterprise checkboxes". Now they have slowed right down, no doubt collapsing under the escalating maintenance weight of all the nonsense that was created, like the canaries in the vibe coding mines telling us of impending doom.

Now you go to their blog, theres a banner at the top announcing "GitLab Agentic AI whatever is GA (GENERAL AVAILABILITY)" and you try to click it its literally a fucking 404 not found. That's the level of their stability and quality. Try it for yourself:

https://about.gitlab.com/blog/

reply
fhd2
6 hours ago
[-]
Maybe it's GU already.
reply
jbverschoor
8 hours ago
[-]
Good thing git is a distributed system
reply
dgxyz
8 hours ago
[-]
Virtually no one knows how to do anything with it outside of github.
reply
nine_k
8 hours ago
[-]
Your favorite search engine or LLM will show you in a second, it's really easy.

The problem is that it's not enough. The fact that Github uses Git specifically is a technical detail; it could use mercurial equally easily, as Bitbucket used to. Github Actions, OWNERS files, PRs and review tools, issue tracker, wiki are all not Git features.

reply
dgxyz
8 hours ago
[-]
Not a chance. I think you need to spend some time in low ball corporate IT. It's just monkeys throwing faeces at the wall. We only just levered them off subversion...

(I use Fossil 100% offline for personal projects for ref)

reply
TZubiri
8 hours ago
[-]
You might be surprised, but that's not true at all.

I once read someone commenting "Nobody writes code by hand without looking syntax up".

Man, you are just outing yourself as a complete beginner, the field is way deeper than you imagine and it's not even close.

reply
dgxyz
8 hours ago
[-]
Not really. I've been around a while. Git for about 15 years. Subversion before that. Perforce before that. rcs before that (back down to sun3 machines). Mostly Fossil now for personal things.

What I am saying is that people learn as much as they need to. They generally don't need to know any more git than is required to interact with github. If anything problematic comes up, they go in with a wrecking ball because they don't truly understand what they are doing. And git has a lot of wrecking balls available.

If you threw them at raw git and asked them to collaborate with someone they'd be up shit creek. They have no idea how SSH or email works for example.

reply
Joe_Cool
8 hours ago
[-]
That's a them problem.
reply
tonymet
8 hours ago
[-]
i still find insightful ways to use git every day. amazing tool. it's a shame for those who only see it as "how to sync my repo with my coworkers"
reply
nine_k
8 hours ago
[-]
Git is!

PRs and code review are not. CI/CD is not.

I mean, there are solutions, but none of them seems to have a large enough mindshare and efficiency. (Even though Github's code review tools are pretty spartan.)

reply
globular-toast
8 hours ago
[-]
> PRs and code review are not. CI/CD is not.

They can be. A PR can be made and code review conducted by submitting a patch to a mailing list. That's how the kernel and, I think, git itself is developed.

CI/CD is really a methodology. It just means integrating/deploying stuff as soon as its ready. So you just need maintainers to be able to run the test suite and deploy, which seems like a really basic thing.

reply
TZubiri
8 hours ago
[-]
True, workers can still commit to their local git.

I've been looking into having a separate git server that we can commit to and add plain ole git hooks to, and just having it be synced with github as a clone.

reply
sirmoveon
6 hours ago
[-]
Check out Gitea. Its kind of a clone of github but you can self host.
reply
howToTestFE
8 hours ago
[-]
If GH has an issue, it seems to always be around 4pm or 5pm GMT. I'm starting to think that i should avoid any planned production releases around this time.
reply
tapoxi
9 hours ago
[-]
helm repo add gitlab https://charts.gitlab.io/ && helm upgrade --install gitlab gitlab/gitlab

I did this in 2019, it avoided so many headaches. CI is better too since there's a nice clean mapping of build -> pod for everything and I can just exec in if something's borked.

reply
odie5533
9 hours ago
[-]
Things would have to get really bad before I considered managing my own repositories. Trading someone else's headaches for my own.
reply
tapoxi
8 hours ago
[-]
It's not as bad as you think, I run the helm upgrade when patches come out, the backing store is S3 or managed SQL, it runs a nightly k8s cron called gitlab-backup which tarballs the whole thing into an s3 bucket with a single command restore should disaster strike. (This is part of the product, not a thing I wrote.)

I probably only babysit it for 30 minutes per year, including all the upgrades.

reply
nine_k
8 hours ago
[-]
It depends how high you value your headaches, and how high, your org's downtime. Github not working accrues over the hourly rate of every developer affected, which is likely $70-$100 a hour. 10 hours of outage in a year affecting a team of 10 would cost north of $70k, enough to hire a part-time SRE dedicated just to tend to your Gitlab installation.
reply
zxcvasd
8 hours ago
[-]
>10 hours of outage in a year affecting a team of 10 would cost north of $70k

10 hours x 10 developers x $70 per hour = $7000, not $70000.

reply
nine_k
5 hours ago
[-]
Thank you for the correction! This indeed completely changes the picture :-\
reply
0xbadcafebee
8 hours ago
[-]
^ this. the last thing i want is to add to my workload. take my money and make my life easier, even if it means that for one hour every couple months i can't do anything
reply
NewJazz
8 hours ago
[-]
Have you ever actually hosted gitlab?
reply
nottimbo
9 hours ago
[-]
Microsoft, it's time to hire some SREs.
reply
arm32
9 hours ago
[-]
We did hire some, boss! Soshie, Vizzy and Dexter. They're AI, but they're supposed to be way better than a human SRE. At least that's what the Sintra salesguy told us.
reply
rvz
9 hours ago
[-]
So that's what the Tay, and Zoe AI bots were doing all this time after they were cancelled and banned off of Twitter.

Working on the GitHub Azure migration and for years it's gone so well so far.

reply
VirusNewbie
8 hours ago
[-]
Microsoft doesn't pay well enough to attract good SRE talent.
reply
lenerdenator
9 hours ago
[-]
Why hire anyone to fix a problem when you can make an AI agent to "fix" it, tell investors about it to pump the price, and not fix anything knowing that you have a monopoly?
reply
aruggirello
8 hours ago
[-]
Clippy to the rescue! :-)
reply
ferguess_k
9 hours ago
[-]
Yes we did hire SREs, unfortunately they are in another continent and they only know how to pull others into the chat. We also have some AI too, do you want to try them? They are pretty good SREs, one of them wrote 100K lines of code in a week while another one reviews every line along the way. It was fantastic! Fantastic!! FANTASTIC!!!

OK I have no idea about MSFT SREs, just to be /s.

reply
andrewinardeer
7 hours ago
[-]
Days since last GitHub incident: 0.2
reply
imglorp
7 hours ago
[-]
14 incidents this month. So far.
reply
postexitus
9 hours ago
[-]
I believe it is an Azure outage or some type of MS service - everything on Azure is down.
reply
zxcvasd
9 hours ago
[-]
having no issues on azure here, seeing no azure incidents on the status page or any of my admin panels
reply
deathanatos
8 hours ago
[-]
> seeing no azure incidents on the status page

… in all seriousness, that is hardly proof that Azure isn't having an outage.

reply
zxcvasd
8 hours ago
[-]
if i thought it alone was proof enough, i wouldnt have also included the bit about how i was actively using azure.

its one signal, among others. and in any case, i wasn't trying to prove the parent commenter wrong. i was offering my own signal to the crowd.

reply
verst
9 hours ago
[-]
I second this. Not experiencing any Azure issues at this time.
reply
ctxc
8 hours ago
[-]
My az services seem to be up.
reply
toephu2
8 hours ago
[-]
This is why companies should host their own source code on-prem.
reply
MadameMinty
9 hours ago
[-]
Angry unicorns seem to be over.
reply
phtrivier
8 hours ago
[-]
Fixed in about 30m to an hour.

Definitely annoying, but I'll try the hot take that, contrary to popular belief, GH is not critical infrastructure - or so I hope.

Please tell me no part of the Ukrainian air defense system depends on a gh action hook.

reply
eddd-ddde
8 hours ago
[-]
You've heard of infrastructure as code, now presenting air strikes as code!

Need a new secret offensive operation? Create a new JSON file with the coordinates, make a merge request and get Commander approval, merge it, and our new proprietary GitHub action runner will deploy a drone in seconds!

reply
philipallstar
8 hours ago
[-]
This is far too simple. The correct way is to generate an NFT that's a screenshot from Google Maps of where you'd like to hit, and a blockchain-watching AI will spot it, figure out where you probably mean and send the coordinates to the fire control system.
reply
vaylian
8 hours ago
[-]
It's not critical, but there's still a lot of reliance on it.

It's also the only reason why I still need IPv4.

reply
ares623
8 hours ago
[-]
When millions of man-hours are lost waiting for your service to be back up, I think that deserves a bit of resiliency.
reply
NewJazz
8 hours ago
[-]
The status page says things are still not fixed.
reply