Github seems to be down.

Edit: After I made this, their status page finally updated to indicate an issue.

Update - We are experiencing interruptions in multiple public GitHub services. We suspect the impact is due to a database infrastructure related change that we are working on rolling back.

  • thesmokingman@programming.dev
    link
    fedilink
    arrow-up
    72
    arrow-down
    1
    ·
    3 months ago

    This is a common problem. Same thing happens with AWS outages too. Business people get to manually flip the switches here. It’s completely divorced from proper monitoring. An internal alert triggers, engineers start looking at it, and only when someone approves publishing the outage does it actually appear on the status page. Outages for places like GitHub and AWS are tied to SLAs that are tied to payouts or discounts for huge customers so there’s an immense incentive to not declare an outage even though everything is on fire. I have yelled at AWS, GitHub, Azure, and a few smaller vendors for this exact bullshit. One time we had a Textract outage for over six hours before AWS finally decided to declare one. We were fucking screaming at our TAM by the end because no one in our collective networks could use it but they refused to declare an outage.

      • thesmokingman@programming.dev
        link
        fedilink
        arrow-up
        20
        ·
        3 months ago

        To be clear, usually there’s an approval gate. Something is generated automatically but a product or business person has to actually approve the alert going out. Behind the scenes everyone internal knows shit is on fire (unless they have shitty monitoring, metrics, and alerting which is true for a lot of places but not major cloud or SaaS providers).

    • RegalPotoo@lemmy.world
      link
      fedilink
      English
      arrow-up
      15
      arrow-down
      1
      ·
      3 months ago

      Or, alternatively, coms management is important and formally declaring an incident is an important part of outage response - going from “hey Bob something isn’t looking right can you check when you get a sec” to “ok, shits broken, everyone put down what you are working on and help with this. Jim is in charge of coordinating the technical people so we don’t make things worse, and should feed updates to Mike who is going to handle comms to non-technical internal people and to externals” takes management input

      • azertyfun@sh.itjust.works
        link
        fedilink
        arrow-up
        19
        ·
        3 months ago

        You’re describing proper incident response but I fail to see what that has to do with the status page. They have core metrics that they could display on that status page without a human being involved.

        IMO a customer-friendly status page would automatically display elevated error rates as “suspected outage” or whatever. Then management can add more detail and/or say “confirmed outage”. In fact that’s how the reddit status page works (or at least used to work), it even shows little graphs with error rates and processing backlogs.

        There are reasons why these automated systems don’t exist, but none of these reasons align with user interests.

      • thesmokingman@programming.dev
        link
        fedilink
        arrow-up
        9
        arrow-down
        1
        ·
        3 months ago

        Speaking from 10+ YoE developing metrics, dashboards, uptime, all that shit and another 5+ on top of that at an exec level managing all that, this is bullshit. There is a disconnect between the automated systems that tell us something is down and the people that want to tell the outside world something is down. If you are a small company, there’s a decent chance you’ve launched your product without proper alerting and monitoring so you have to manually manage outages. If you are GitHub or AWS size, you know exactly when shit hits the fan because you have contracts that depend on that and you’re going to need some justification for downtime. Assuming a healthy environment, you’re doing a blameless postmortem but you’ve done millions of those at that scale and part of resolving them is ensuring you know before it happens again. Internally you know when there is an outage; exposing that externally is always about making yourself look good not customer experience.

        What you’re describing is the incident management process. That also doesn’t require management input because you’re not going to wait for some fucking suit to respond to a Slack message. Your alarms have severities that give you agency. Again, small businesses sure you might not, but at large scale, especially with anyone holding anything like a SOC2, you have procedures in place and you’re stopping the bleeding. You will have some level of leadership that steps in and translates what the individual contributors are doing to business speak; that doesn’t prevent you from telling your customers shit is fucked up.

        The only time a company actually needs to properly evaluate what’s going on before announcing is a security incident. There’s a huge difference between “my honeypot blew up” and “the database in this region is fucked so customers can’t write anything to it; they probably can’t use our product.” My honeypot blowing up might be an indication I’m fucked or that the attackers blew up the honeypot instead of anything else. Can’t send traffic to a region? Literally no reason the customer would be able to so why am I not telling them?

        I read your response as either someone who knows nothing about the field or someone on the business side who doesn’t actually understand how single panes of glass work. If that’s not the case, I apologize. This is a huge pet peeve for basically anyone in the SRE/DevOps space who consumes these shitty status pages.

  • zzx@lemmy.world
    link
    fedilink
    arrow-up
    8
    ·
    3 months ago

    I’m always on-call whenever shit blows up… Nothing I can do GitHub is literally hard down