At this point, GitHub outages feel closer to cloud provider outages than a SaaS blip. Curious how many people here still run self-hosted Git (GitLab / Gitea) vs fully outsourcing version control.
My previous two startups used GitLab successfully. The smaller startup used paid-tier hosted by gitlab.com. The bigger startup (with strategic cutting-edge IP, and multinational security sensitivity) used the expensive on-prem enterprise GitLab.
(The latter startup, I spent some principal engineer political capital to move us to GitLab, after our software team was crippled by the Microsoft Azure-branded thing that non-software people had purchased by default. It helped that GitLab had a testimonial from Nvidia, since we were also in the AI hardware space.)
If you prefer to use fully open source, or have $0 budget, there's also Forgejo (forked from Gitea). I'm using it for my current one-person side-startup, and it's mostly as good as GitLab for Git, issues, boards, and wiki. The "scoped" issue labels, which I use heavily, are standard in Foregejo, but paid-tier in GitLab. I haven't yet exercised the CI features.
Can someone in GitHub senior leadership please start paying attention and reprioritise towards actually delivering a product that's at least relatively reliable?
I moved my company over to GH enterprise last year (from AzDO) and I'm considering moving us away to another vendor altogether as a result of the constant partial outages. Things that used to "just work" now are slow in the UI, and GH actions fail to schedule in a reasonable timeframe way more than they ever used to. I enjoy GH copilot as much as the next person, but ultimately I came to GH because I needed a git forge, and I will leave GH if the git forge doesn't work.
Not going to happen. This is terminal decline. Next step is to kill off free repos, and then they'll start ratcheting up the price to the point that they have one small dedicated engineering team supporting each customer they have. They will have exactly one customer. At some point they'll end up owned by Broadcom, OpenText, Rocket, or Progress.
The ultimate irony is that Linus Thorvalds designed git with the Linux kernel codebase in mind to work without any form of infrastructure centralisation. No repo trumps any other.
Surely some of your crazy kids can rummage up a CI pipeline on their laptop? 8)
Anyway, I only use GH as something to sync interesting stuff from, so it doesn't get lost.
I wonder if GitHub is feeling the crush of fully automated development workflows? Must be a crazy number of commits now to personal repos that will never convert to paid orgs.
I consider moving away from Github, but I need a solid CI solution, and ideally a container registry as well. Would totally pay for a solution that just works. Any good recommendations?
We can run a Forgejo instance for you with Firecracker VM runners on bare metal. We can also support it and provide an SLA. We're running it internally and it is very solid. We're running the runners on bare metal, with a whole lot of large CI/CD jobs (mostly Rust compilation).
The down side is that the starting price is kinda high, so the math probably only works out if you also have a number of other workloads to run on the same cluster. Or if you need to run a really huge Forgejo server!
I suspect my comment history will provide the best details and overview of what we do. We'll be offering the Firecracker runner back to the Forgejo community very soon in any case.
You should reach the same conclusion by trying to use it for this purpose, but also indeed for any purpose at all. Incidents that make you unable to deploy making all your CD efforts pointless are only the cherry on top.
Someone needs to make an mcp server for my claude so it can check if services are down, it goes stir crazy when github is down and adds heaps of work around code =D
This is the predictable outcome of subordinating the GitHub product to the overarching "AI must be part of everything whether it makes sense or not" mandate coming down from the top. It was only a year ago that GitHub was moved under the "CoreAI" group at Microsoft, and there's been plenty of stories of massive cost-cutting and forcing teams to focus on AI workflows instead of their actual product priorities. To the extent they are drinking their own Kool-Aid, this sort of ops failure is also an entirely predictable outcome of too much reliance on LLM-generated code and workflows rather than human expertise, something we see happening at an alarming scale in a number of public MS repos.
Hopefully it will get bad enough fast enough that they'll recognize they need to drastically change how they are operating. But I fear we're just witnessing a slow slide into complacency and settling for being a substandard product with monopoly-power name recognition.
It would be interesting to have a graph showing AI adoption in coding against the number of weekly outages across different companies. I am sure they are quite correlated.
> It would be interesting to have a graph showing AI adoption in coding against the number of weekly outages across different companies. I am sure they are quite correlated.
Probably a stronger correlation to the fact that vibe-coding has resulted in millions of new repos being created, with automatic CIs being triggered by agents continuously sending PRs for those projects.
Remember the other day when a bunch of yous were making fun of zig moving away from GitHub?
Now suddenly you all say this is not the future you wanted.
Everyday you opt in to get wrecked by Microsoft.
You all do realize you all could, for a change, learn something and never again touch anything Microsoft related?
I still say that mixing CI/CD with code/version control hosting is a mistake.
At it's absolute best, everything just works silently, and you now have vendor lock-in with whichever proprietary system you chose.
Switching git hosting providers should be as easy as changing your remotes and pushing. Though now a days that requires finding solutions for the MR/PR process, and the wiki, and all the extra things your team might have grown to rely on. As always, the bundle is a trap.
I'm starting to wonder if people doing what were previously unconventional workflows (which may not be performance optimized) are affecting things.
For example, today, I had claude basically prune all merged branches from a repo that's had 8 years of commits in it. It found and deleted 420 branches that were merged but not deleted.
Deleting 420 branches at once is probably the kind of long tail workflow that was not worth optimizing in the past, right? But I'm sure devs are doing this sort of housekeeping often now, whereas in the past, we just never would've made the time to do so.
I moved everything on github to a self hosted foregjo instanse some days ago. I really did not do anything. Created some tokens so that CC could access github and forgejo and my dns API. Self hosting is so much simpler and easier with AI. Expect more people to self host small to medium stuff.
Yeah, once you start self-hosting your code, it’s kind of nice having control over everything. Makes you think about moving other stuff, like analytics, to something self-hosted too.
67 comments
[ 2.9 ms ] story [ 71.8 ms ] threadMy previous two startups used GitLab successfully. The smaller startup used paid-tier hosted by gitlab.com. The bigger startup (with strategic cutting-edge IP, and multinational security sensitivity) used the expensive on-prem enterprise GitLab.
(The latter startup, I spent some principal engineer political capital to move us to GitLab, after our software team was crippled by the Microsoft Azure-branded thing that non-software people had purchased by default. It helped that GitLab had a testimonial from Nvidia, since we were also in the AI hardware space.)
If you prefer to use fully open source, or have $0 budget, there's also Forgejo (forked from Gitea). I'm using it for my current one-person side-startup, and it's mostly as good as GitLab for Git, issues, boards, and wiki. The "scoped" issue labels, which I use heavily, are standard in Foregejo, but paid-tier in GitLab. I haven't yet exercised the CI features.
I moved my company over to GH enterprise last year (from AzDO) and I'm considering moving us away to another vendor altogether as a result of the constant partial outages. Things that used to "just work" now are slow in the UI, and GH actions fail to schedule in a reasonable timeframe way more than they ever used to. I enjoy GH copilot as much as the next person, but ultimately I came to GH because I needed a git forge, and I will leave GH if the git forge doesn't work.
Surely some of your crazy kids can rummage up a CI pipeline on their laptop? 8)
Anyway, I only use GH as something to sync interesting stuff from, so it doesn't get lost.
Anyone else having issues? It is blocking any kind of release
The down side is that the starting price is kinda high, so the math probably only works out if you also have a number of other workloads to run on the same cluster. Or if you need to run a really huge Forgejo server!
I suspect my comment history will provide the best details and overview of what we do. We'll be offering the Firecracker runner back to the Forgejo community very soon in any case.
https://lithus.eu
(although admittedly less load and redundancy)
Do you allow me to run the following command?
https://thenewstack.io/github-will-prioritize-migrating-to-a...
And coincidentally, an early CircleCI engineer wrote an article about GitHub Action (TLDR: don't use GitHub Action for CI/CD!)
https://www.iankduncan.com/engineering/2026-02-05-github-act...
https://news.ycombinator.com/item?id=46908491
You should reach the same conclusion by trying to use it for this purpose, but also indeed for any purpose at all. Incidents that make you unable to deploy making all your CD efforts pointless are only the cherry on top.
It's just "yet another day of business as usual" as this point.
A prophecy that was predicted half a decade ago [0] which is now more important then as it is now today.
[0] https://news.ycombinator.com/item?id=22867803
Hopefully it will get bad enough fast enough that they'll recognize they need to drastically change how they are operating. But I fear we're just witnessing a slow slide into complacency and settling for being a substandard product with monopoly-power name recognition.
Probably a stronger correlation to the fact that vibe-coding has resulted in millions of new repos being created, with automatic CIs being triggered by agents continuously sending PRs for those projects.
The history for today is a bit of a mess really: https://www.githubstatus.com/history
Everyday you opt in to get wrecked by Microsoft.
You all do realize you all could, for a change, learn something and never again touch anything Microsoft related?
Fool me once...
At it's absolute best, everything just works silently, and you now have vendor lock-in with whichever proprietary system you chose.
Switching git hosting providers should be as easy as changing your remotes and pushing. Though now a days that requires finding solutions for the MR/PR process, and the wiki, and all the extra things your team might have grown to rely on. As always, the bundle is a trap.
For example, today, I had claude basically prune all merged branches from a repo that's had 8 years of commits in it. It found and deleted 420 branches that were merged but not deleted.
Deleting 420 branches at once is probably the kind of long tail workflow that was not worth optimizing in the past, right? But I'm sure devs are doing this sort of housekeeping often now, whereas in the past, we just never would've made the time to do so.