This setup assumes a shared filesystem rooted in
/tmp.
IMPORTANT we don't have a development environment for this yet, so it isn't finished. The aws setup without a shared filesystem is testing and working, with more complete documentation in the README there.
You'll want to first create the appropriate perilog and epilog scripts that will setup and teardown userspace
Kubernetes for a job. Here is an example for how to do that (and you only need to do this once) for a system
that has flux system configuration in /etc/flux. First, let's copy the script included here to
a file named "prolog" in /etc/flux/system and create a subdirectory for our prolog scripts.
mkdir -p /etc/flux/system/prolog.d /etc/flux/system/epilog.dWe assume you have cloned usernetes-python and are sitting at the root.
These top level scripts will be run, and execute those in (epilog|prolog).d
There is a script provided to install usernetes to where your system (root) will see it, along
with prolog/epilog scripts to your flux system configuration root, which defaults
to /etc/flux/system.
/bin/bash ./scripts/install-scripts.sh /etc/flux/systemFor batch jobs that want to deploy and teardown. The last step is to configure your flux instance to allow prolog and epilog. Follow steps two and three here to edit the broker config.
Anything that deploys a job with an epilog or prolog should work, and with one or more instances.
flux batch -N4 --setattr=attributes.user.usernetes=yes kubectl get pods
flux submit -N4 --setattr=attributes.user.usernetes=yes kubectl get pods
flux run -N4 --setattr=attributes.user.usernetes=yes kubectl get pods