Given that we have had difficulties with pinning specific Python dependencies that play nicely together in DPS, creating a Dockerfile based on the same Docker image that is specified in our algorithm_config.yaml file should help us identify and correct such problems early, prior to any attempt to register our algorithm, let alone attempt to run a job with it.
A relatively simple Dockerfile should suffice, and should likely need only to do the following:
- Use a
FROM statement that specifies the same Docker image as specified in algorithm_config.yaml
COPY the minimal set of files necessary to allow a subsequent RUN ./build.sh --dev to succeed
This can also serve as basis for later creating a devcontainer for use within VS Code, as well as for setting up a GitHub Action for running automated tests.
Given that we have had difficulties with pinning specific Python dependencies that play nicely together in DPS, creating a Dockerfile based on the same Docker image that is specified in our
algorithm_config.yamlfile should help us identify and correct such problems early, prior to any attempt to register our algorithm, let alone attempt to run a job with it.A relatively simple Dockerfile should suffice, and should likely need only to do the following:
FROMstatement that specifies the same Docker image as specified inalgorithm_config.yamlCOPYthe minimal set of files necessary to allow a subsequentRUN ./build.sh --devto succeedThis can also serve as basis for later creating a devcontainer for use within VS Code, as well as for setting up a GitHub Action for running automated tests.