Hi all, just trying to setup a dev flyte deploymen...
# flyte-deployment
f
Hi all, just trying to setup a dev flyte deployment on our local (k3s backed) cluster and wondering how "integral" the example sandbox deps are... The dashboard I already have, object storage (minio) as well... main question is how integral contour is to the flyte deployment... Should I bite the bullet and try to make the charts work with our existing ingress (traefik) or rip that out and go with contour? And how hard is it to change the deployment to use our existing minio (where our valuable data/model live) or maybe just deploy minio again for "just" flyte workflow related storage?
k
We do not have an opinion about ingress
even dashboard and minio are to make sandbox work
f
so if I already have an ingress I should be able to just deploy the flyte-core helm chart? I have to admit I'm a bit overwhelmed with the options in the chart
k
I know
sorry about that, we are working on simplifying it further
is this a full deployment
or just a demo cluster
f
so any pointers on the minimal set I would need to deploy or how to approach this?
k
Cc @Yuvraj / @Eduardo Apolinario (eapolinario) can any of you give good pointers
Ping again cc @Eduardo Apolinario (eapolinario)
e
Sorry, I'll take a look at this today.
y
@Felix Ruess We have one deployment guide for sandbox/on prem. You can try that https://docs.flyte.org/en/latest/deployment/sandbox.html#deployment
f
yeah, that is what I was looking at and the question was about... since there it says to install flyte-deps which includes countor, etc... and there are contour labels etc in the flyte-core chart
y
you can use any type of ingress with flyte-core, We internally use nginx, few folks use contour/Google Ingress in open source, traefik should work. but i guess you need to do some work
k
@Felix Ruess hey! I'm coming in here a bit late but I have installed Flyte on our cluster and we use traefik. There's some things we had to do to enable GRPC but other than that, the ingress manifest files are minimal that we had to add. If you and other folks are interested in putting some heads together to write something up for a minimal production deployment for folks running their own k8s clusters, please let me know. It was a bit of work but we had to generate the manifests manually using helm (but not helm install) and removed the things we don't need, like dashboard, ingress, etc and update things that are different for us like node selectors etc
f
hey @Katrina P sorry, just had time to come back to this again. Would be great to pick your brain and put something together!
157 Views