This has not been my experience - it's phenomenalally cheaper to shell out a few bucks to GH than deal with self hosted self control.

Even an hour a month of lost dev time would make it worth it.

Also, thanks to the stupid design decisions of most package managers, you will have trouble getting anything done anyway if github.com is down, even if you just get dependencies from there. If that problem affects you, then self-hosting your Git repo may just add a second point of failure.

You want to be setup so you can do a build completely offline of head or any tag. GitHub being down shouldn't make a difference. If the PM is getting in the way of that then dump it.

Rebuilding the entire toolchain of your language sounds like a colossal waste of time when you just wanted to avoid a few hours per year of developer downtime.

No need to rebuild everything. Simply having the "DLLs" or equivalent checked in or in a maven/NuGet folder backed up would do the trick.

Your not just saving developer downtime. Remember that non incident a few months ago? Could be a security issue too.

Edit: npm incident

I guess you are referring to the recent incidents with Github. At least when Github (or any non-self hosted service) goes down, _you_ don't have to stop doing the task that your user/clients care about and go fix it, you have a whole team dedicated to that, for free.

I really want you to produce evidence for that. Everyone I know who selfhosts gitlab has much lower uptime than github.

I don't know about GitHub or GitLab, but in my entire professional career, I think my total downtime due to failures in locally hosted Git repos is zero. Likewise I don't recall ever having a problem due to locally hosted bug trackers or code review systems. This stuff isn't hard, and we've been able to do it reliably for a very long time.

What certainly has wasted a horrible amount of my time in recent years is working around build systems and package managers that are so badly designed that not only do they have a dependency on some online repository in the first instance, they also make it difficult or impossible to download and cache those dependencies in a supported way so that you can have 100% reproducible builds with nothing but locally hosted resources. Surely this is just about the most basic requirement for a robust software development process?

Since you're looking to collect anecdata, I've had 100% uptime from my self-hosted GitLab instance, which has been online for about a year. GitHub may be on par with that, but it's hard to beat 100%.

Just curious : don't you apply updates ?

I self host my gitlab and when there's an update it goes down for several minutes (well actually I don't know if it's unusable, haven't tried, but with the backup and the updates, it's long)

I'm the only one working on it, so that's not a problem for me.

Still, it takes time.

I must have transposed "unexpected" downtime in my head. Yes, I update the box and installation periodically, and yeah, that is time that my GitLab environment is not available, so I guess it's not 100%, but like you I am the only person using it, so it's effectively 100%

Sometimes you need the ability to offload blame more than the statistics that you had one more Nine of uptime.

You should generate your own electricity too! Sarcasm aside, every client will accept "Amazon is down, millions of programmers are affected". And then I get to head to the pub while the only finger you can point is at yourself. And then cost. I don't know your hourly rate, but an hour of fixing something myself pays for a year of most services out there.