haproxy-slim is just an experiment in creating a slim haproxy image which consists of a base image on alpine linux, containing haproxy as well as rsyslog, and a final image that contains just the config file. More on why below.
docker pull markbnj/haproxy:1.6.2
Having deployed haproxy several times in containerized environments on both AWS and Google Cloud, I've been giving some thought to the way it handles configuration, which comes into play in service discovery contexts where you want to update the routing to reflect the addition or removal of some service.
Typically the approach is some variation of: update the config and then get haproxy to do a fast reload by forking itself. However this violates the principle of keeping containers immutable.
In an orchestrated environment like kubernetes, for example, what you would want is something like this:
- Push updated config to a repo
- Build system kicks off and builds an image with the new config
- Do a graceful update
There's a lot in (3) but I don't want to go down that hole here. The main thing is that you're rebuilding your image and redeploying it. The new image gets the new config, and you can use various techniques to start moving traffic to it.
In this scenario the time to build, push, and subsequently pull the image will be the main factor in how long it takes to update haproxy and get the new rules into effect.
So these two images are the beginning of some experiments around that. The base image has haproxy 1.6.2 (currently the alpine package version), and rsyslog. The final image depends on the base and just adds the config file.
This should allow those base image layers to get cached and make pushes/pulls where only the config changes quite fast. The whole image is at this point only 12 MB.