I have an 8-cpu server and I installed Centos 7 on it. I would like to dynamically and programmatically spin up and down VM nodes to do work, ex. Hadoop nodes.
Is the technology I use for this Vagrant or Puppet, or something else? I have played around with Vagrant, but it appears that every new node requires a new directory in the file system, I can't just spin up a new VM as far as I can tell with an API call, I think. And it doesn't look like there's even a real API for Vagrant, just machine-readable output. And if I understand it properly, Puppet deals with configuration management for pre-existing nodes.
Is either of these the correct technology to use or is there something else that is more fitting to what I want to do?
Yes, you can use vagrant to spin up a new vm. Configuration of that particular vm can be done using puppet. Take a look at: https://www.vagrantup.com/docs/provisioning/puppet_apply.html
And if you're problem is having separate directories for each vm, you're looking for a multimachine setup: https://www.vagrantup.com/docs/multi-machine/
For an example using the multiserver setup take a look at https://github.com/mlambrichs/graphite-vagrant/blob/master/Vagrantfile In the config directory you'll find a yaml file which defines an array that you can use to loop over different vm's.