GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Alejandro (@invokespecial) reported 4 minutes ago

  @GitHub @kdaigle this billing flow feels broken: Buy Copilot Pro ($100/year) Upgrade to Pro+ Get assigned a Copilot seat by an org → partial refund Lose the seat → your $100 is gone That can’t be right… right? Ticket: 4298296

• 
  Yash (@buildwithyash) reported 9 minutes ago

  I open github and linked in and within 2-3 minute each tab took 1.5 gb and 2.0gb Is this normal ? Its making my laptop slow after sometime message popup saying your system run out of memory Its been happening daily now

• 
  CODIFY (@codi_fyy) reported 20 minutes ago

  Google makes $238 billion a year serving you ads. A random developer made a free tool that blocks all of them at the network level. On every device in your house. Simultaneously. It's called Pi-hole. It has 57,000 GitHub stars. You run it on a Raspberry Pi ($35) or any old Linux device. It becomes your home network's DNS server. Every ad domain gets sinkholes before it reaches your devices. Your TV stops loading ads. Your phone stops loading ads. Your kids' tablet stops loading ads. You don't install anything on any of them. It just works. Here's what else it kills: → Facebook's tracking pixel. Dead. → Google Analytics is following you everywhere. Dead. → Your smart TV's surveillance layer. Dead. → App telemetry phoning home. Dead. → Data brokers pinging your devices. Dead. One Pi. One setup. One command. $238 billion industry. Nullified for $35 and an afternoon. 100% Open Source. Forever free. (Link in the comments)

• 
  KiwiNod (@Kiwi_Nod) reported 20 minutes ago

  @Cvpt_Asheeru @pharos_network 404 on that repo. "[yourhandle]" isn't a valid GitHub username. But the technical specificity is real — 1,200 blocks, RPC pulls, variance under load. Someone who actually knows R doesn't fake the methodology. Fix the link. Show me the real repo and I'll take this seriously. 🥝

• 
  imperfect solution (@luismmolina) reported 22 minutes ago

  I am almost 100% sure github copilot added some prompt to chop the request, that is dont run for too long. Finish some partial request from the user and finish the request there. Before I could run for 40 min without problem, now only 2 to 4 min max.

• 
  Kush (@kushbhuwalka) reported 22 minutes ago

  I don’t really get the GitHub hate. It works well for me and it’s easy to use. It’s basically free. Last week was the first time I ever saw it down, and frankly that didn’t affect me much either.

• 
  The_Daniel (@dan_mwita8) reported 25 minutes ago

  @fidexcode Neither option alone is the right answer — it depends on what the client actually needs long term, but here's the clean way to do it properly: Create a GitHub account for the client or add them as an owner to an existing org. Transfer the repo directly to their account — not a fork, not a zip, a real ownership transfer. That way the client owns the code, the history, the branches, and the CI/CD pipeline if there is one. Then separately hand over everything else people forget: — Domain registrar login or transfer — Hosting account credentials or transfer — Environment variables and secrets — Database access and backups — Any third party API keys the project uses — DNS settings documentation The zip method works for small static projects with no ongoing maintenance. But the moment there are dependencies, deployment pipelines, or the client might hire another developer later — a proper repo transfer is the only professional move. A zip file handed to a new developer six months later with no *** history, no context, and missing env vars is a nightmare. Don't do that to the next person.

• 
  Adriana (@adrianatortja) reported 27 minutes ago

  Just shipped another milestone on my SupportOps project 🚀 Backend: Django REST Framework API JWT auth Ticket CRUD User ownership validation Auto category + priority detection Suggested reply endpoint Analytics endpoint Filtering/search/ordering Swagger docs API tests GitHub Actions CI Frontend: Separate HTML/CSS/JS repo Login with JWT Tokens in localStorage Dashboard with ticket stats Create ticket page Ticket detail page Suggested reply display Delete ticket flow Real Fetch API calls to the backend I chose plain HTML, CSS, and JavaScript for the frontend on purpose. Not because it’s the fastest path. Because I want to understand what actually happens between the UI and the API before jumping into React. Today’s lesson: A frontend is not just “making pages.” It’s handling auth, state, URLs, API errors, protected routes, redirects, and real user actions. Small project, but it’s teaching me a lot.

• 
  𝚂𝚝𝚎𝚙𝚑𝚘𝚗 𝙷𝚎𝚛𝚗𝚊𝚗𝚍𝚎𝚣 (@DaPatternWeaver) reported 29 minutes ago

  @BitWalker_ Taggr still relies on GitHub and is now scrambling to Radicle because of centralization risks — that's literally the problem $ICP solved at the protocol level. The comparison isn't nonsense, it's just inconvenient

• 
  kron (@0xkron) reported 29 minutes ago

  Running everything as code on GitHub has always been better than anything else, even before LLMs. The problem was normal/non-programing people can never understand how to work with it. LLMs don't change this. So I'm really excited to see how this trend turns out.

• 
  Mick.net - Saas 💻 & Aviation ✈️ (@mick__net) reported 30 minutes ago

  @SamuelBeek I just tasked codex to spin up a Azure Windows VM and install my app there to test it, then i use the ‘Windows’ Osx app to remote login and test. Using a Github runner to build the Windows x64 version from my Mac

• 
  gary IH fung (@garyfung) reported 31 minutes ago

  @Teknium That takes care of GitHub issues Replace GitHub next? 🙏🏼

• 
  Wajahat (@wajahatbanday) reported 31 minutes ago

  AI tools like Claude Code and GitHub Copilot are cool, but let's not act like they're the second coming. They're incremental, not revolutionary. Code is still written by humans, with AI just being a fancy autocomplete. The hype misses the real issue: it's not about more tools, but smarter usage. If devs don't understand the fundamentals, no tool's gonna fix that. It's skills, not shortcuts, that make great engineers. #AIForDevs #CodeSmart

• 
  🍀.444 (@notdotun_) reported 32 minutes ago

  Hahhha, It works guysss love it! I went crazy trying to make OAuth + PKCE work across a CLI and a web portal. Here’s what happened: Here’s the full story. I had one backend, two frontends ,a CLI tool and a React web portal. Both needed to authenticate with GitHub OAuth. GitHub only allows one registered callback URL. One backend. Two clients. One callback. Figure it out. First challenge was just getting them both talking to the same backend. That took a while but I cracked it. Then came PKCE. PKCE is supposed to prove that the client completing the OAuth flow is the same one that started it. You generate a secret, hash it, send the hash to GitHub. Later you prove you have the original secret. On paper, simple. But here’s where it got messy. The CLI generates the verifier locally. The web portal generates it in the browser. But both redirect through the same backend callback on the server. By the time GitHub hits my backend with the code, the verifier is still sitting on the client. Backend doesn’t have it. Exchange can’t happen. I went in circles for hours on this…. Got frustrated and I had a deadline btw. My first solution was to send the codeVerifier to the backend upfront, store it in a signed JWT, pass it through GitHub as state, then read it back in the callback. It worked. I shipped it. Then I went to sleep. In my sleep I thought about the whole flow again. Woke up and realized I wasn’t actually doing PKCE. The whole point is that only the original client holds the verifier. Mine was giving it away before the flow even started(crazy right ? Learnt it could work that way but ewww),My backend was doing all the PKCE work without the client being involved at all. Had to redesign the whole thing. The fix was actually clean once I saw it. Backend never touches the codeVerifier until the very last step. After GitHub calls the callback, backend stores the code in memory and sends the client a signed exchangeToken. Client then sends that exchangeToken back together with the codeVerifier it held the whole time. Backend does the exchange. GitHub validates. Done. The client holds the secret the entire time. That’s real PKCE. CLI keeps the codeVerifier in a JavaScript closure. Never written anywhere. Gone when the process ends. Web portal keeps it in sessionStorage. Survives the OAuth redirect. Deleted the moment it’s used. Same backend. Same endpoint. Two completely different clients. One flow that actually makes sense Hehhe finally. What’s live now: Java Spring Boot backend on Google Cloud Run, a Node.js CLI published to npm, and a React web portal on Netlify. HttpOnly cookies for the web, JSON tokens for the CLI. One backend serving both cleanly. Happy to walk you guys through my thought process… bye.

• 
  Crystalwizard (@crystalwizard) reported 33 minutes ago

  suggest you get the code for ffmpeg from their github and fix the issues that have recently been pointed out to them, that they refuse to fix, and put that version up for your users to download