Benchmarking GitHub Enterprise

githubengineering.com

62 points·dbussink··11 comments

I'd love an overview of how GitHub Enterprise is deployed and maintained as a product. How are releases handled? Are all installs on-premise, or any hosted at on GitHub infrastructure?

The documentation for GitHub Enterprise is public: https://help.github.com/enterprise/2.2/.

Prior to Enterprise2 upgrades are done with a package that updates the appliance 'in-place' and restarts services (or a reboot? It's been a long time).

Enterprise2 writes a new root file system on upgrade and starts from there. Files for this are distributed from either to enterprise homepage or the releases page, https://enterprise.github.com/releases/. In either page you'll find a link to a download page where a selection can be made for installation or upgrade.

Ah, I completely missed that. They have absolutely fantastic documentation for the entire product. Thanks for the links and info!

I find the drop with the peak before and afterward in the “Git Clone Over SSH Against GitHub.com” graph interesting. I would have expected a parabola. Why is that?

Doesn't GitHub Enterprise use JRuby?

Nope, it is a deployment of the ruby version we use on .com.

Got it. What's the overhead of using RubyEncoder or some sort of code obfuscation/loader protection technology? I'm interested to know how much it costs compared to straight up MRI/JRuby.

It used to use RubyEncoder, now it uses something custom. The overhead is very minimal, and it's only when the source is read off disk.

Has it used JRuby in prior versions of Enterprise (or FI)?

FI was JRuby; as part of the naming switchover and product overhaul to the first release of GitHub Enterprise, we switched it over to MRI in late 2011.

Off topic: any chance you could support the invocation of web-hooks via http proxy?

Reply on news.ycombinator.com