Local development

During development you often need to quickly iterate and test your changes. If the step is pushed to the director at every iteration, this can add a lot of friction between writing the code and getting feedback by running the step. Local development mode can help reduce that friction.

Overview

Local development mode mounts a path on the local filesystem for an in-development step rather than building the step code into a published step. This allows the developer to push the step once, locally, then update code and run the step without having to push the changes first. This can speed up experimentation and iteration significantly. The downside to using this method is that it only works with a locally running director that is able to mount the provided path, so it is unsuitable for production deployments.

Comparing step types

	Local	Docker
Purpose	Fast local development	Production deployment
Deployment	Local director only	Local or cloud director
Updating	Push once, updated code automatically used	Push every time code is updated
Running	Local director must be available	Step can run on any available director

info

In order to use local development mode, it's necessary to have a director running locally on the developer's machine, preferably via Docker.

Pushing steps locally

Once you have a step that is ready to run for the first time, push it locally by using the sand steps push local command:

$ sand steps push local --name localStep \
--volumeMountPath /path/to/step

Note the highlighted line: the --volumeMountPath flag should point to the directory containing your code (if it is not in main.py, use the --entrypoint flag to indicate the correct entry point).

warning

If you are developing in Devcontainer or another containerized environment, don't forget that the path to your step will differ based on where the Devcontainer files are located on your host system.

Once the step is created, any updates to the code in the mounted directory will be automatically mounted and run on future runs of that step. This enables you to rapidly expand, test, and adjust your step code without the overhead of having to push it to Sandgarden with every change.

Running local steps is the same as running docker steps: sand runs start --step <name> --input <input>. Local steps can also be combined into workflows, but keep in mind that running those workflows also requires your local director to run its steps.

Graduating to deployment

Once you're satisfied with the performance and functionality of a step, you can push it as a regular step by using the normal sand steps push docker command. The only changes you'll need to make to the command line are:

• Replace local with docker
• Replace --volumeMountPath with --file

Once a step is deployed using the docker type, it can run on any available director, local or cloud. If it is pushed to a specific cluster, it will run on any director in that cluster.