#mcollective

Volcane: i want to write a playbook for a few of our cluster. Mostly consul/elk/vault. Basic pattern is always: stop the agent apply system updates. Reboot. Puppet run. Continue with next node if cluster is green again

Do you have a recommendation for patching a node? I got facts that report all packages with available updates. I could iterate at them and update each package

Which would be 3 lines in puppet dsl

But that would mark all of them as explicitly installed, which ruins the dependency tree

I got archlinux, debian and centos nodes

Or maybe enhance https://github.com/puppetlabs/mcollective-packa...

https://github.com/puppetlabs/mcollective-packa... that looks easy to extend. But i am not sure if that is a good approach

All those bolt/plans/tasks/playbooks combinations are a bit confusing

yeah, you could make a agent to take care of the patching - or soon a task - and drive it with the new playbooks, it has loops and all that stuff

the package agent pretty much want to do 1 package at a time its safest and easiest to do in the most generic basis

but for your own agent where you understand your system and constraints I'd add a basic agent that just speaks to your package manager and takes maybe lists of packages

compared to trying to wedge that into existing agent - and i am reluctant to take multi package patch actions (be hard to get right with the puppet providers) your own agent is best

I thought it could fit into the existing package agent. Something like Shell.new(yum update -qy)

sure you could do that if you're happy with the shell agent :)

for yum updates though doing so via mcollective - a system wide update - is risky

update puppet-agent and that restarts mcollective - not apparently in latest version though

so its risky, with the tasks feature i will ship soon it will work better since a task execute entirely dissociated from mcollective process

i *think* the Shell agent does the same

I installed puppet-agent 5.4 and it restarted mcollective

Which was bad because i wanted to update the puppet agent via mco :D

ah right the thing i am thinking of is this https://github.com/puppetlabs/marionette-collec...

still restarts, doesnt kill children

Happened on centos. It works on arch because arch has no aio package. Puppet and mco are seperate

this cycle of many steps per node in the cluster - new playbooks make this easy :)

pre release docs http://dev.choria.io/docs/playbooks/basics/ ^F in there for "Loop the Puppet Agents in groups of 10"

Thanks. Will check it after lunch