Hacker News Reader

Comment by jeppester | Hacker News Reader

How stupid are we all to rely so heavily on this single point of failure which is so unstable?

I cannot count the times it's been down or had temporary instability. It's easy to draw the conclusion that it happened after MS took over, but I honestly no longer remember how it was before.

View on HN

It was worse before MS took over. But their status page displayed live metrics, it was more transparent, so people were less frustrated. As opposed to now, where the status dashboard is updated by a human.

IMHO, the issue is not the unstability. I'm really concerned that so much of our software infrastructure relies on one single service owned by Microsoft.

I remember it very differently. I remember remarkably good reliability, to the point that HN would regularly laud GitHub for “doing it right” by investing heavily in infrastructure and reliability and being “boring”. I remember some GitHub leader or founder affirming that sentiment, that they’ve always felt their value proposition depends greatly on reliability. I remember it didn’t change as soon as MS took over, but a year or so later, and that for the first few years of diminishing reliability people would always point fingers at the MS takeover.

You seem to have had a different pre ms github experience than I. Github didn't really embrace the Suck til it foisted Actions on the world. Gitlab had a lot of downtime for a few years (shockingly mostly to do with CI related mechs).

It was worse but I'm not sure people were less frustrated. Some outages back in 2012-2014 periods were so long that our dev team went for few pints to nearest pub.

Is reliability always a good thing? I could do with a few myself.

There are many github alternatives. Are you saying you'd prefer your ci/cd pipeline to be scattered across services, or that you'd like a github backup if it goes down? You could probably mirror your code on gitlab and host a Jenkins server on aws instead of relying a single point of failure. Personally I'll take the single point of failure downtime in exchange for lower maintenence

There are many github alternatives, sure. But sofware is still centralized on github.

I double dare you to "pip install", "cargo build", or "go get" without hitting github.

Your project can be hosted on repo.or.cz, gitlab or codeberg, you will still need github to build it locally, or to build it on your self hosted CI.

The solution is P2P git hosting. Dare I say magnet-links git repo urls?

Ok this makes sense, thanks for the clarification.

It is every month, something in Github (especially GitHub Actions) goes down. The last time this happened was less than a month ago. [0] [1]

At this point, It is so unreliable that if I was to bet anyone that something in GitHub would go down next month, I will win each time. Guaranteed.

[0] https://news.ycombinator.com/item?id=35359212

[1] https://news.ycombinator.com/item?id=35325709

We're a primarily GitLab shop so I have a biased view, but I haven't noticed that much disruption from GitHub's instability. We do get a lot of packages from GitHub too and I haven't noticed a ton of CI jobs failing or anything.

> We do get a lot of packages from GitHub too

Do you get them directly from GH or from a local mirror of the GH repo? The latter makes you immune to GH hiccups at the cost of some disk space (assuming you run GL yourself).

Should be directly from GitHub except with some caching involved (like our Docker build cache).

(We don't run GitLab ourselves.)

Because it is free.

It has a free tier but GitHub has plenty of paying customers and is very much a monetized service.

And thankfully not monetised like SourceForge did.

Reply on news.ycombinator.com