Skip to content

Local UI dev setup

The below are (high level) steps for doing local ui development of the static version of mlflow.

Clone repos

cd /some-work-dir/

# the composable-logs repo is needed to
#  - download build artifacts/run logs from demo pipeline
#  - convert these into format that can be compiled into static website
git clone git@github.com:composable-logs/composable-logs.git

# The mlflow clone repo contains a fork of the official mlflow project.
# This fork supports building static websites.
git clone git@github.com:composable-logs/mlflow.git --branch static_mlflow

The below steps will also use test pipeline run logs produced from the mnist-demo-pipeline repo, but this repo does not need to be cloned locally.

Download test data

Here we download test data using a PAT (Personal Access Token), see docs if these are not familar. A (safer) approach (that avoids using a PAT) would be to manually download artifacts from the Github UI.

1. Generate PAT in Github UI

  • Note that the PAT will have access to everything the logged in user has access to (for selected scopes). So assign minimum priviledges to the PAT:
  • only scope needed is public repos.
  • set minimum expirence limit, eg. 1 or 7 days.
  • delete the PAT when no longer needed (step 2b below).

2. Start and set up Docker container with dependencies

Eg. start VS Code dev container in the composable-logs repo's otel_output_parse directory.

pip install -e .
export GITHUB_TOKEN="<see above>"

# verify that generate_static_data cli is installed
generate_static_data --help

2a. Download workflow (zip) artifacts into local cache directory

The below will download available build artifacts from past workflow runs into a ./cache directory. 

rm -rf ./cache    # <-- !!
generate_static_data \
   --github_repository composable-logs/mnist-digits-demo-pipeline \
   --zip_cache_dir ./cache

2b. Delete the token created above in the Github UI

2c. Parse zip file content into form suitable for for static ML Flow website

  • this step does not require network/API access and uses only files in the ./cache directory.
  • this will delete any previous content in ./static_output directory.
rm -rf static_output    # <-- !!

generate_static_data \
   --zip_cache_dir ./cache \
   --output_www_root_directory ./static_output/www_root

3. Copy data into ML Flow repo

Copy the outputs into correct directories in the (forked) ML Flow repo:

# See step above where repos where cloned:
cd /some-work-dir/

rm -rf mlflow/mlflow/server/js/public/pipeline-artifacts    # <-- !!

# TODO: review details for the below commands.
cp -r \
   composable-logs/otel_output_parser/workspace/static_output/static-data.json \
   mlflow/mlflow/server/js/public/

cp -r \
   composable-logs/otel_output_parser/workspace/static_output/pipeline-artifacts \
   mlflow/mlflow/server/js/public/

4. Build static mlflow/run in watch mode

See makefile in the ./mlflow/server/js/ directory of the static_mlflow branch of the cloned mlflow-repo