Can someone explain what exactly is so bad now that leaving it entirely to use some new platform, even spinning up your own servers, is a reasonable alternative?
So much so that they stopped posting uptime metrics for a while on their status page and an independent 3rd party created a website just for this:
https://mrshu.github.io/github-statuses/ (not my website)
According to that website, which unsurprisingly reports a lower number than what Github themselves claim, Github uptime is down to ~86%.
And if you work in the space, you know how terrible that is, but even more so for such a critical piece of infrastructure.
Yes, because it throws all partial outages into one bucket, which is a dumb idea because the bigger a platform becomes with more loosely coupled components the more untainable high uptime number become.
Looking through the incidents, a good portion of them are regarding Copilot and Codespaces, two products I couldn't care about less. I do also have my regular run-ins with Github outages, but that website is just hyperbolic.
The truth likely lies somewhere in the middle.
And not sure how your individual preferences invalidate others' experiences with the platform. To be frank Copilot is likely one of their most visible and in-demand products as it was a very cost-effective way to access frontier models (and are unsurprisingly nerfing usage limits in a month).
When they changed the PR view to not display all the changes at once, was the moment I said "I really need to find something else", not only is the platform very unreliable (at least from Spain), but most product changes they do are making the platform less efficient for me as a developer to use.
> Can someone explain what exactly is so bad now that leaving it entirely to use some new platform, even spinning up your own servers, is a reasonable alternative?
It always was, but network-effect of GitHub been large. But seemingly not infinite, at one point people start favoring "Being able to access platform" over "people can star my repository" it seems.
And the most recent bug is "we added random code to some PR and some PR became invisible" which is a wtf bug which should be impossible to exist.
Most random errors in Github Actions (e.g. jobs just randomly failing or getting stuck and requiring a manual restart, or just being plain slow) also never show up on the Github Status page. The Github Actions VMs are also so slow that I'm seriously pondering setting up a cheap throw-away laptop at home as runner, that would easily be 10x faster. But then we're at playing IT admin at home :/