Question: Helm with k8s_yaml not refreshing resources? #5932
Comments
|
if you change the version.yaml, does that change the kubernetes yaml that helm() generates? (you can see this yaml if you do a if the yaml doesn't change, then kubernetes will not restart the app. |
|
I think it changes only the initial request, which at the end includes the fresh timestamp: I don't think it changes anything else, except the container image nothing else is changing. The container is different though, so this should probably trigger the update? Anyways, do you see any workaround for me, to force redeployment in such case? Here is the helm yaml for the reference: |
|
Anyways, I added extra tag to helm chart which contains timestamp value, this made the yaml different and now the re-deployment is triggered. This seems however like some dirty trick, it would be great if Tilt could handle it natively. I'm kind of missing two features, to add additional dependency outside the helm chart and force the update even if the chart produces the same yaml: |
|
A couple things to note here: All of Tilt's image-building primitives will use an immutable tag instead of a mutable one. In your example, you're using For more on this, see: https://docs.tilt.dev/custom_build.html#why-tilt-uses-immutable-tags re: "This seems however like some dirty trick" -- Most modern orchestrators work this way by default -- you declare the infrastructure you want, and if nothing has changed, than the orchestrator will not make any changes. Here's some docs on Kubernetes on this: https://kubernetes.io/docs/concepts/architecture/controller/. i.e., if you deploy the same yaml, but you've mutated the image that the YAML points to, then Kubernetes will not restart the pod. |
|
(closing this issue, because this is basically working as I would expect, though happy to answer further questions about how you're building/deploying the image -- which i think is the underlying misunderstanding here) |
|
Yeah, perhaps I'm not using Tilt as it was initially designed. The system which I'm building locally just for development tests is also built for production via full blown CI/CD pipeline, so I don't want to duplicate all these building details in the Tiltfile. Therefore it is convenient for me to build the container image and deliver it to local image registry via maven plugins, as it will work the same way for production builds, just the registry address will be different. Also Helm chart is the same as for production, with different variable values. That's why I'm not using any tilt-specific building primitives. |
|
Interesting, it seems that with my |
|
ya, that makes sense. tilt doesn't care what you use to build container images; custom_build lets you use non-docker builders. |
I'm not sure if this is bug or I'm doing something in a wrong way. Here it goes:
I have simple test Spring Boot app, for which Jib plugin is creating the container and pushes it to local registry during each software build. Then I wanted Tilt to deploy it automatically to local K8s cluster, using my Helm setup. Here is my
Tiltfilecontents:So after intial start with
tilt upthe App is deployed by Tilt and I can see in the console all the logs etc, everything works nice. But for some reason it is not updating my App after new build will happen.I disabled in
.tiltignorethetarget-helmfolder as it is recreated during each build and this seems to confuse Tilt. However each successful build (which is also pushing new image to the local registry) is refreshing dummyversion.yamlfile which is containing just a timestamp. And I see that Tilt is watching properly on that file.After each build I can see in the console that Tilt discovers changes to
version.yaml, however it is not triggerring updates to the App running in the cluster. And when I click manual updates, it works properly, helm is executed and the App is redeployed.Also output from filewatch:
Any hints greatly appreciated!
The text was updated successfully, but these errors were encountered: