How We Built Airbnb Holiday Cards in Five Days (opens in new tab)

I'm curious about how exactly to go about extricating authentication, but I'm not sure where to look. Any pointers for best practices or cool patterns for this?

ZooKeeper is really the most accepted option here. It is used by many other important projects (Kafka, Storm, Hadoop) so adding it to your stack is likely to pay dividends later when you want to adopt some other cool technology that relies on it.

Doozer is the other option I've heard of that fills this role, but I don't personally know of anyone using it.

In terms of initial discoverability, services register in a designated place in ZooKeeper using an ephemeral node and then other services can find them there. You can set a "watch" on the ephemeral node to get notified if it goes away for any reason (ie. the other service or node goes down)

Doozer: https://github.com/ha/doozerd

We do use a Rails engine for one of our components, but this isn't something we've explored a lot (or at least, I haven't personally; maybe some of the other Airbnb nerds will step in with their own thoughts). Even so, I don't see this as a good long-term solution for our architecture, mostly because not everything we do is Rails -- we currently run JVM services and Node.js services, for instance, and we'd like to keep things flexible in the future so we can always pick the best tool for a particular use case. Our routing service should be able to handle all those needs.

When they say gems, I would have to assume they mean rails engines since they mentioned they are using it to package up assets as well.

I've been using the same pattern. It works ok, but still feels clunky. ∞ bundle update. dev mode reloading for certain classes can be a major pain as well.

Hey there, thanks for your interest!

Our service discovery system is indeed a layer that sits between ZooKeeper and the application. But it's also pluggable -- that means that you don't have to use ZooKeeper as the system that tracks which service nodes are available; you can use something else, and write a plugin for our service discovery system that will talk to that instead. So it's not necessarily a layer on top of ZooKeeper, although we certainly use it that way.

I don't want to say too much about the technical details yet since we're still rolling it out in production and may change some pieces before it's released. Also, I'm not one of the primary authors and don't want to steal their thunder. ;) But yes, we do plan to open source it once the code is fairly stable (and once we have some time to jump through all the usual hoops of open sourcing code), and then we'll do a detailed post about how it all works. It's really cool and stunningly simple.