Open jomeier opened 11 months ago
Just to rule it out, if you do actually re-initialize the code server, does it then work correctly?
No, it doesn't.
Re-initialization of the code server reloads anything (assets, jobs, ...) but it always complains about the instanciated variable "not found" ...
The User Code Container Must be started with an initialized "defs = Definitions(...)" in the target python_file, then further "reloaded" Updates seem to work.
@dpeng817 I found out that for some reason, if I mount a PVC (Kuberntes Persistent Volume Claim) into the Dagster user code location Pod in the Helm Chart, everything seems to work as expected. I can change code and even if initially there is no code, I can add code and after a reload in the UI everything works.
Without the PVC reload does not work.
That does not make any sense for me :)
Dagster version
1.5.11
What's the issue?
Hi, I use the "code-server" experimental feature on OpenShift/Kubernetes. My use case is, that I want to change code in the user code container on Kubernetes and "hot reload" it in Dagsters Web UI, without rebuilding the user code image. This works so far. Why I want to do that? Because later I want to connect to the running container with Visual Studio Code and comfortably change the code during experimentation phase.
I use the default user-code-example Docker image from DockerHub and point to the (empty) file: /example_project/example_repo/init.py
Dagster complains first, that init.py is empty and that it doesn't find any job, asset, ...
If I add this code here to init.py afterwards in the running user code container:
... reload the code location in Dagsters UI and materialize the Job, Dagster complains with this error message:
It seems as if Dagster does not "hot reload" the Definitions.
What did you expect to happen?
No errors.
How to reproduce?
Described above.
Deployment type
Dagster Helm chart
Deployment details
values.yaml of my Helm Chart:
Additional information
No response
Message from the maintainers
Impacted by this issue? Give it a 👍! We factor engagement into prioritization.