Looks like Azure as a platform just killed the ability for VM scale operations, due to a change on a storage account ACL that hosted VM extensions. Wow... We noticed when github actions went down, then our self hosted runners because we can't scale anymore.
Information
Active - Virtual Machines and dependent services - Service management issues in multiple regions
Impact statement: As early as 19:46 UTC on 2 February 2026, we are aware of an ongoing issue causing customers to receive error notifications when performing service management operations - such as create, delete, update, scaling, start, stop - for Virtual Machines (VMs) across multiple regions. These issues are also causing impact to services with dependencies on these service management operations - including Azure Arc Enabled Servers, Azure Batch, Azure DevOps, Azure Load Testing, and GitHub. For details on the latter, please see https://www.githubstatus.com.
Current status: We have determined that these issues were caused by a recent configuration change that affected public access to certain Microsoft‑managed storage accounts, used to host extension packages. We are actively working on mitigation, including updating configuration to restore relevant access permissions. We have applied this update in one region so far, and are assessing the extent to which this mitigates customer issues. Our next update will be provided by 22:30 UTC, approximately 60 minutes from now.
They've always been terrible at VM ops. I never get weird quota limits and errors in other places. It's almost as if Amazon wants me to be a customer and Microsoft does not.
Amazon isn't much better there. Wait until you hit an EC2 quota limit and can't get anyone to look at it quickly (even under paid enterprise support) or they say no.
Also had a few instance types which won't spin up in some regions/AZs recently. I assume this is capacity issues.
Perception changes with every next generation. For the last 8 years, I've been teaching at the Faculty of Mathematics and Informatics at our University. One of the courses that I lead is IoT, where students get to program on bare metal (embedded) systems. I noticed that newer (cloud) generations have a harder time accepting the constraints of embedded hardware and living with them.
I was surprised hitting one of these limits once, but it wasn't as if they were 100% out of servers, just had to pick a different node type. I don't think they would ever post their numbers, but some of the more exotic types definitely have less in the pool.
If you work at AWS in a technical role you can check the capacity of each pool in each AZ using an internal tool. Previously the main reason for pool exhaustion was automated jobs at the start of each working day as well as instance slotting issues (releasing a 4xl but only re-allocating a l means you now cannot slot another 4xl).
Really prefer Hetzner in this sense because they actually talk about limits. I recently got myself a hetzner account (after shilling it for so much, hearing positivity, I felt like it was time for me to discover it)
I wanted to try out the most cheapest option out of frugality & that was actually limited (but kudos to them that they mentioned that these servers have limits) so no worries I went and picked the 5.99 euro instead of the 3.99 euro option instead.
They also have limits option itself as a settings iirc and it shows you all the limits that are imposed in a transparent manner and my account's young so I can't request for limit increases but after some time, one definitely can.
Essentially I love this idea because essentially Cloud is just someone's else's hardware and there is no infinitium. But I feel as if it can come pretty close with hetzner (and I have heard some great things about OVH and have a good personal experience with netcup vps but netcup's payments were really PITA to setup]
Hetzner is a dedicated server (meaning monthly contract, 1 month setup fee and up to 1 week delivery time) company that branched out into cloud, so it's not that surprising they treat cloud a bit like that. While Amazon wants you to think they have an infinite capacity pool, and any failure to get a server is an unexpected error, Hetzner seems to not hide they have a finite number of servers in a finite number of racks, since that's how their main business works.
Nobody really shows their global limits including Hetzner. Hetzner doesn't, like, call it a secret internal error when they run out of capacity of a type.
Agreed...I've been waiting for months now to increase my quota for a specific Azure VM type by 20 cores. I get an email every two weeks saying my request is still backlogged because they don't have the physical hardware available. I haven't seen an issue like this with AWS before...
We've ran into that issue as well, ended up having to move regions entirely because nothing was changing in the current region. I believe it was westus1 at the time. It's a ton of fun to migrate everything over!
That’s was years ago, wild to see they have the same issues.
Can someone explain the point of cloud like I'm a 60 year old grumpy Unix admin because you could just get a real server from another company by now. If the whole point is unlimited capacity but you don't have unlimited capacity and you're paying through the nose then why? Compliance?
Compliance and tooling are a big part of it, but the places where the big public cloud providers shine is the PaaS offerings that you don't need to write yourself.
In Azure, for example, it's possible to use Entra as your Active Directory, along with the fine grained RBAC built in to the platform. On a host that just gives you VPS/DS, you have to run your own AD (and secondary backups). Likewise with things like webservers (IIS) and SQL Server, which both have PaaS offerings with SLAs and all the infra management tasks handled for you in an easily auditable way.
If you just need a few servers at the IaaS level, the big cloud platforms don't look like a great value. But, if you do a SOC2, for example, you're going to have to build all the documentation and observability/controls yourself.
It's awful. Any other service in Azure that relies on the core systems seems to have issues trying to depend on it, I feel for those internal teams.
Ran into an issue upgrading an AKS cluster last week. It completely stalled and broke the entire cluster in a way where our hands were tied as we can't see the control plane at all...
I submit a severity A ticket and 5 hours later I get told there was a known issue with the latest VM image that would create issues with the control plane leaving any cluster that was updated in that window to essentially kill itself and require manual intervention. Did they notify anyone? Nope, did they stop anyone from killing their own clusters. Nope.
It seems like every time I'm forced to touch the Azure environment I'm basically playing Russian roulette hoping that something's not broken on the backend.
All 3 hyperscalers have vulnerabilities in their control planes: they're either single point of failure like AWS with us-east-1, or global meaning that a faulty release can take it down entirely; and take AZ resilience to mean that existing compute will continue to work as before, but allocation of new resources might fail in multi-AZ or multi-region ways.
It means that any service designed to survive a control plane outage must statically allocate its compute resources and have enough slack that it never relies on auto scaling. True for AWS/GCP/Azure.
> It means that any service designed to survive a control plane outage must statically allocate its compute resources and have enough slack that it never relies on auto scaling. True for AWS/GCP/Azure.
In a way. It means that you can get new capacity most often, but the transition windows where a service gets resized (or mutated in general) has to be minimised and carefully controlled by ops.
Yeah I remember one maybe four years ago? Existing workloads were fine but I had to go and tell my marketing department to not do anything until it was sorted because auto-scaling was busted.
It's notable that they blame "our upstream provider" when it's quite literally the same company. I can't imagine GitHub engineers are very happy about the forced migration to Azure.
Having worked there around 2020-2021 there were many folks not happy with being forced to use azure and being forced to build GitHub actions based on azure devops. Lots of AWS usage still existed at that time but these days u bet it’s mostly gone.
True enough. The world is never as predictable as the computers we program, and the computers we program are never as predictable as we feel they should be.
Nobody is happy with Oracle anything! It has some users because it is free. It has paid users because Larry Ellison bribed the government. Nobody would choose it voluntarily.
A few years ago I talked to an developer advocate for Azure. I wanted to know why it took for ever when you wanted a new public IP. My take was that it felt like they went out on the internet to look for an IP to purchase from a 3rd. party. The answer I got was that do to the silos within Microsoft it might as well be a 3rd party supplier. The slowness is exactly because IPs are/were a managed by another Microsoft entity, who views any interaction, even within the company, as hostile.
I get your point, but it just sounds a bit funny when it's an artefact of corporate structure that it's true.
Like imagine if AWS was composed of separate companies for different services - Fargate was an Heroku acquisition say - and then they all went down and blamed their 'upstream provider' because they can't work without say VPC or EC2 availability.
I think that's all GP meant, it just reads a bit funny, not that it's wrong.
As an isolated event, this is not great, but when you see the stagnation (if not downwards trajectory) of GitHub as a whole, it‘s even worse in my opinion.
edit: Before someone says something. I do understand that the underlying issue is some issue with Azure.
Doesn't seem like Microsoft managers care - it's not their core business, so any time anyone complains about issues with GitHub they probably think something along the line of "peasants whining again".
Must be nice to be a monopoly that has most of the businesses in the world as their hostages.
At one point Gitlab seemed like it wanted to compete, but then they killed all the personal and SMB plans, and now they’re just out of the picture for a lot of people. Their team plan is more expensive that GH’s enterprise plan.
Gitlab was generous first, to rise as a valid alternative to GitHub. They never got the comminity aspect right, perhaps aiming for profitability with a focus on the runners instances which is how they make money.
With profitability, the IPO made sense.
GitHub probably had a different strategy..keep it generous, get the entire open source community, keep raising money and one day someone will buys us out for billions. We we are, Microsoft goal is to capture the community, it works. It's sticky.
Yes, but this also means that countless open-source projects are in what appears to be a precarious position. What if MS one day decides all this free hosting isn't worth it, and just cuts it off? There aren't really any alternatives I know of, except bad ol' Sourceforge I guess.
Sadly Github moving more into Azure will expose the fragility of the cloud platform as a whole. We've been working around these rough edges for years. Maybe it will make someone wake up, but I don't think they have any motivation to.
I really like codeberg if your project is licensed in an Open license.
One of the reasons I still use github is that I have starred quite a lot of projects and had to make an account initially to star a project. (I used to have bookmarks beforehand but I wanted to support author in a minor way :] and also github being de-facto & I wanted to talk to some projects which had issues which I wanted to create/discuss)
Another minor point is that Github actions are more generous than Codeberg's actions equivalent.
I believe hosting own Codeberg ie. Forejo (which is a gitea fork)/ gitea is actually easy. I once hosted them on my android phone using termux and on servers. Really liked the idea of having essentially github at my pockets.
For Gists [which is something that I like using a lot personally]. I found the idea of opengists really interesting as well. one minor complaint with opengists is that I love the comment part of gists which is an open issue in opengists but its not implemented yet. Wish it could be implemented.
Regarding losing bookmarks, I actually have a custom tampermonkey script in a private gist which shows a star button which essentially moves my bookmarks to some gist in a json format so as to not lose them ever again essentially.
Personally, I run my own Forgejo instance for the private repos I actually care about. But it's basically impossible to not have a GitHub account right now. I use "Refined GitHub" to make the UI somewhat usable.
In the Bad Old Days before Github (before Sourceforge even) building and package sucked because of the hundred source tarballs you had to fetch, on any given day 3 would be down (this is why Debian does the "_orig" tarballs the way they do). Now it sucks because on any given day either all of them are available or none of them are.
On what OS have you noticed this? Very in character for microsoft to artificially slow non-windows downloads. Then again, my apt upgrades on Debian have been dog slow lately...
Tay.ai and Zoe AI Agents probably running infra operations at GitHub and still arguing about how to deploy to production without hallucinating a config file and deploying a broken fix to address the issue.
Since there is no GitHub CEO, (Satya is not bothered anymore) and human employees not looking, Tay and Zoe are at the helm ruining GitHub with their broken AI generated fixes.
"Impact statement: As early as 19:46 UTC on 2 February 2026, we are aware of an ongoing issue causing customers to receive error notifications when performing service management operations - such as create, delete, update, scaling, start, stop - for Virtual Machines (VMs) across multiple regions. These issues are also causing impact to services with dependencies on these service management operations - including Azure Arc Enabled Servers, Azure Batch, Azure DevOps, Azure Load Testing, and GitHub. For details on the latter, please see https://www.githubstatus.com."
Information
Active - Virtual Machines and dependent services - Service management issues in multiple regions
Impact statement: As early as 19:46 UTC on 2 February 2026, we are aware of an ongoing issue causing customers to receive error notifications when performing service management operations - such as create, delete, update, scaling, start, stop - for Virtual Machines (VMs) across multiple regions. These issues are also causing impact to services with dependencies on these service management operations - including Azure Arc Enabled Servers, Azure Batch, Azure DevOps, Azure Load Testing, and GitHub. For details on the latter, please see https://www.githubstatus.com.
Current status: We have determined that these issues were caused by a recent configuration change that affected public access to certain Microsoft‑managed storage accounts, used to host extension packages. We are actively working on mitigation, including updating configuration to restore relevant access permissions. We have applied this update in one region so far, and are assessing the extent to which this mitigates customer issues. Our next update will be provided by 22:30 UTC, approximately 60 minutes from now.
https://azure.status.microsoft/en-us/status
reply