Wow, yesterday was fun!

I had a rather buggy piece of code, it was bad when I first wrote it, and then I fixed it up, and it was still bad. Now I rewrote almost all of it, and it's much better.

Bad? How? Well, it was in Go, and it's basically an agent meant to execute tasks one at a time, and report the results back to home (live). Now while it worked, it was really flimsy, race conditions, way to much blocking, bad logic, and some very bad bugs.

So I had to rewrite it. Time for a quick primer on the design of this: you have a queue, a task gets add to the queue, the task manager runs the task. In the mean time, the agent is polling the host with the latest output from the task, and also receives new tasks to run (if there are any).

Seems like something that's for a messaging queue, you ask? Well, that would be true if each task was able to run on any random agent, but each task is only meant to run the agent it's tasked to (the tasks are of administrative nature al la apt-get), so having a whole separate service is a tad overkill.

So rewriting required rethinking how the tasks are executed by the task manager. I spent a day on this, it was fun, I ended up copying go contexts (very simple model, very useful). Why copy and not reuse? Because this is meant to be low memory code, so any extra parts are problematic, and I didn't really see a use for having a whole context, I just needed a way to announce that a task is done.

Anyways, if you're interested to see how the implementation worked out: https://github.com/chabad360/covey/...

Add Comment