Mgmt – Next generation distributed, event-driven, parallel config management | Hacker News Reader

Mgmt – Next generation distributed, event-driven, parallel config management

github.com

92 pointspabs325 comments

I'm not op, but I am the author of this project. Always a bit surreal when you're browsing the internet and you notice something familiar like this! Happy to answer any questions you have.

Hi purpleidea, I've been following the project for years because of how much I loved the concept and how inspiring I found one of the talks you gave about how the configuration management space is all wrong. I remember when you left your job at RedHat to work on mgmt full-time and was excited for what would come of it.

I was just wondering if you can comment on the current status of the project and its future direction at this point?

Is there any documentation on how to deploy it at large scale?

You'd have to ask a more specific question, this is a very general broad one which I can't answer directly. I don't know of any org using it atm on 10,000+ machines, but if you have such a cluster and want to try it out there, please lmk and I'll be happy to give you some tips.

What's the path to production for mgmt? Is there a plan to get there? Is there a timeline?

Great question...

Firstly, as I said above:

> the amount of time I devote to it ebbs and flows since I need to eat and have job and life responsibilities... It's still ever present in my mind, and the more patches and donations I get always helps!

So sadly I don't have a real timeline. There are some people using it for real things, but that number is probably small as there are still some features I'd like to see for MVP.

Please get involved if you'd like to make it go faster!

Hey purpleidea, I was listening your presentation(s) about mgmt-config & graphs on Fosdem and these were very entertaining. Good work!

Some time later tried to use it and it seemed project was abandoned. Not sure if it was but there were no recent commits and I felt quite sad. Happy to see the project is back to life!

Hiya! It was never abandoned, but the amount of time I devote to it ebbs and flows since I need to eat and have job and life responsibilities... It's still ever present in my mind, and the more patches and donations I get always helps!

The idea of instant updates across the fleat is terrifying to me. But if you added some sort of rate limiter it would be very cool. For example it can only update one machine per data center every hour and only half of the data centers on any given day.

It should be terrifying. I’ve seen chef/puppet runs across an entire fleet cause multimillion dollar outages multiple times in my career.

> To Err is Human, It takes automation to really screw things up

You can model this if you want it that way, but most people generally don't want code that runs slower ;)

You can build staged roll outs too, although I always wanted to add this as a native resource but didn't yet.

I've been using rqlite for distributed config. No scripting, just put your config params in there. Then your apps/services read from SQLite no problem.

rqlite author here -- good to hear that rqlite is working for you. Are you using read-only nodes? Happy to answer any other questions.

https://github.com/rqlite/rqlite

There's no reason AFAICT that mgmt couldn't use rqlite as a backend assuming rqlite supports watching for events on certain key paths...

If adding such an alternate backend would be interesting for you, we'd love to have it.

Yea, a few nodes are the voting/origin and then everywhere else is ro-node. And then some app-code reads the DB directly.

What is it and what does it do? I tried reading the readme, but I have no idea of its use or utility.

Controls your stuff.

Take only what you need from it.

(do-do-do-do do-de-do-do, dooo-dooo)

Sorry, silly music/lyrics joke. But it seems to be a scripting-focussed alternative to Puppet and Ansible, at least from what I can see.

So, remote deployment of packages, timed/scheduled configuration changes etc., I think.

I wouldn't say "scripting focused". It seems to have a good amount of Puppet lang compatibility. It is definitely an alternative to both Puppet and Ansible that doesn't introduce dependency hell.

Is it fire-and-forget (like Ansible) or desired-state (like Salt)?

It is desired state, but it's also real-time. Something new. Try it out or watch a talk.

Previously: https://news.ycombinator.com/item?id=25359845

I guess i missed partaking in that discussion, but a lot of those comments probably could use some good replies to offset the knee-jerk reactions I saw in some of them. Ping if you'd like to discuss. I think your fears about etcd are unfounded btw, as it's actually really excellent software.

I'm not OP, but I think your FAQ (and some other places) comes off pretty strong. It's okay to be opinionated, but your wording comes off as "everyone is wrong, I'm not".

Do you have a real example please? It's meant to just represent what we believe. If you think I've gotten it wrong somewhere or I could rephrase, please let me know or send a patch!

Reply on news.ycombinator.com