Over the next several weeks Opscode Engineering and Ops teams will be upgrading a number of systems in Hosted Chef. These upgrades are the result of months of behind the scenes hard work. We’ve been quiet, not because we don’t care, but because upgrading large complex systems like Hosted Chef requires lots of planning to get it right. On the contrary, we care deeply about our users and their experiences using all the many flavors of Chef. I’d like to share some details about upcoming upgrades to demonstrate how much we care and how hard we’re working to improve Hosted Chef.
One of Hosted Chef’s benefits is its support for fine-grained permissions. The original permission manager had performed well from the day we turned on Hosted Chef until recently. We analyzed the permission manager’s performance and concluded the current design was nearing its natural scaling limits.
We decided the best course of action for Hosted Chef and our users was to replace the current system with a new design. This design would address our current performance needs and more easily scale as Hosted Chef grows. The end result, after months of work, is a new permission manager twice as fast as the old design, stable under load, and completely backwards compatible. The new system, codenamed ‘bifrost’, will be deployed during the outage we recently announced for tomorrow 6:00 PM – 7:00 PM PDT.
We’re also finalizing plans to bring the Chef 11’s new API server, erchef, to Hosted Chef in the coming weeks. erchef has been proven in demanding environments such as Facebook and Cycle Computing‘s on-demand compute clusters.
Bringing erchef to Hosted Chef is a huge project involving a cast of virtual thousands touching almost every Hosted Chef system and component. We can’t wait to share erchef with our Hosted Chef users but we want to do it right. This means taking the extra time to attend to details and minimize any user-facing impacts. We’re making great progress and will have more to share soon.