Comment by EpicEng

> The approach in taking to self hosted services is that if it's down for an hour or even a couple of days it's fine since I'm the main user.

Well, maybe, depends on what you're depending on that service for. However, we were talking about large scale centralization v decentralization. If you took everything currently on GitHub and moved it to a bunch of different self-hosted servers, you would have a problem. Many _do_ depend on what's there on GH.

Even if it's not critical, it really sucks to want something and not be able to get it for 2 days because Jim went on vacation and his server crashed.

> For disaster recovery, I put everything in ansible playbooks and roles. The role also takes care of setting up hourly backups (with Borg) and will import the latest backup when setting up the service. Because backups are separate, I can delete the last couple of backups if they've been compromised.

> It's obviously not as resilient as using the centralized equivalent, but it feels like a decent trade off.

The issue isn't that it cannot be done, it's that centralized services mean _I_ don't have to worry about it. I don't want to be a server admin; I just want to store my code in source control.