When using Helios and the SkyDNS plugin, over time etcd will have a bunch of junk in it. This is especially true when it is used in a continuous integration/deployment environment with the Helios Testing Framework. This is because the framework makes randomized subdomains so that more than one person can run the same set of tests using service discovery and not have them step on each other. In this case, when Helios tears down the registry entries and/or they expire (if the Helios agent was killed), the directories leading to the files it creates for SkyDNS do not go away when their associated files' TTL expires.

I've thought about fixing the SkyDNS plugin to do this cleanup, but it would suffer from a few drawbacks:

• Leaving some artifacts in etcd is helpful for debugging when tests fail.
• It would significantly complicate what is now a very simple plugin.
• In the case of agent restarts, these cleanups still wouldn't happen.
• In production environments, leaving directories around like this is unlikely to be truly problematic since the structure shouldn't change as much as it does in test environments (especially in the face of the testing framework), as the raw numbers will be significantly smaller.
• Even with tons of garbage nodes around, etcd seems to perform just fine -- it's really just a way to make the recursive fetch easier to parse as humans when debugging things.

And so we have SkyGC.

It's really braindead right now. All it does is try to remove directories on the specified etcd server that are part of the skydns tree. Directories that still have active nodes in them will not be removed.

skygc etcd_url skydns_domain

If your etcd server is at https://myetcd.example.com:4001 and the domain that SkyDNS manages is services.dev.example.com, your commandline would be:

skygc https://myetcd.example.com:4001 services.dev.example.com

It generates fairly copious amounts of output, and most of the failures it outputs aren't really failures in the normal sense, and since this is just a garbage collector, failure is generally ok.