etcdwatch

WARNING This is alpha quality software at this point. I have so far used it only in limited settings so it is possible that changes to command line options, behavior etc. may occur before I consider it stable enough to be non-alpha.

Etcdwatch is a simple program that does three things:

1. It tracks current state of a given etcd server/cluster.
2. It serializes each state into JSON, YAML or other format.
3. It gives each state as a standard input for a specified program.

In reality it's a bit more complex, but that's what you are buying.

Try this. First, start etcd if it's not yet running:

$ etcd

then start etcdwatch:

$ etcdwatch -- sh -c 'cat; echo'

and do some changes in the registry:

$ curl -s http://127.0.0.1:2379/v2/keys/test -XPUT -d value=something
$ curl -s http://127.0.0.1:2379/v2/keys/test2 -XPUT -d value=other
$ curl -s http://127.0.0.1:2379/v2/keys/test -XDELETE
$ curl -s http://127.0.0.1:2379/v2/keys/test2 -XDELETE

Watch what happens.

Setup

Simply:

$ python setup.py install

Usage

There are several options that can be used configure stuff like etcd host, port etc. --- see etcdwatch --help.

Examples

Here's a snippet from a project:

$ etcdwatch -- sh -c './regenerate.py && nginx -s reload'

The regenerate script reads the input, updates nginx configuration and reloads the configuration. In this case it is used to handle a devtest environment where each branch gets deployed to AWS ECS. During deployment the address and port information of each sub-service gets updated to etcd which in turn triggers the reverse proxy configuration to be dynamically updated --- and we have **branch**-**service**.dev.some.dns.name available instantly!

Caveats

Etcd is not a transactional data store. If you use multiple keys it is possible that some of them are set or updated while others are not. Thus you should always check whether the input actually makes sense before using it.

Etcdwatch runs the command only on stable registry state. This occurs when, after an update, no more updates have occurred for a short period of time (a few seconds, configuable via --stable-timeout option). If you really want every single detected update (as told by etcd, which again may not be exactly equal to each update done by a client) to cause the program to be run, use --no-stable command line option.

License

MIT License © Santeri Paavolainen