https://dagster.io/ logo
#announcements
Title
# announcements
a

Adrian

03/11/2021, 8:45 PM
Is there a suggested developer workflow for deploying workspaces to dagster k8s?
y

yuhan

03/11/2021, 8:47 PM
Hi @Adrian please check out our k8s deployment docs: https://docs.dagster.io/deploying/kubernetes
a

Adrian

03/11/2021, 8:50 PM
Hi @yuhan I have everything deployed. The helm chart values let you select user code and repos but once that is done how would iterate on those repos?
y

yuhan

03/11/2021, 8:56 PM
Ah got it! Did you mean deploying multiple repos / user code, i.e. multiple User Code Deployments? or the dev workflow of iterating the user code after you’ve deploy it?
cc @rex
a

Adrian

03/11/2021, 8:58 PM
exactly. looking to write a bunch of new stuff and "deploy" those workloads
r

rex

03/11/2021, 8:58 PM
hey Adrian - our workflow around this is that we created a CI workflow to
helm upgrade
our release whenever we push changes to our repository holding our dagster pipeline code
a

Adrian

03/11/2021, 8:58 PM
I have microk8s w/ the basic k8s deployment going but I want to ship solids and pipelines as workloads
@rex ah I see. I'm using kustomize to generate my manifests for different environments. It looks like the only way to do this would be to re apply those manifests? Hard part is I keep my infra in a different project then my workloads
If I'm understanding this correctly, if my repo are shipped to my docker registry and the deployment is set to pull
tag: latest
as long as the deployment config is not changing when run it should pickup the changes?
do you have the source for this example? https://hub.docker.com/r/dagster/k8s-example
r

rex

03/11/2021, 9:10 PM
I’m not familiar with kustomize, but I’m assuming it works like Helm. In Helm, upgrades are triggered are diffing each manifests - for the manifests that have changed, Helm will redeploy those
if you set
tag:latest
and nothing else has changed, by itself it will not pickup the changes
you’ll need to add some dynamic part in the manifest: similar to https://helm.sh/docs/howto/charts_tips_and_tricks/#automatically-roll-deployments
a

Adrian

03/11/2021, 9:14 PM
ah, the values.yml allows you to specify annotation too
p

Peter Stratton

03/11/2021, 9:45 PM
we’ve used an annotation with the current timestamp of the deployment to force a re-deployment
r

rex

03/11/2021, 9:47 PM
yeah - in our case, we don’t use
tag:latest
, we tag images by their git hash so each deployment manifest will be different (if indeed the pipeline code changed)
a

Adrian

03/12/2021, 2:11 PM
okay great, I'm going to try this out
I'll have to pull the helm chart and separate the configmaps to the other repo that has the Dockerfile and solids and use Skaffold