Streamlit¶
There are 4 main ways we recommend to spin up Streamlit applications for quick dashboarding:
The Streamlit demo created in this guide will accept a string as an input, and the dashboard will provide an output as to whether the sentiment is "positive" or "negative", following the guide's problem statement.
This guide will be similar to that of "Deployment" and "Batch Inferencing", with the difference mainly being the use of Streamlit as an interface to get your inputs and show your outputs from.
While it is possible for Streamlit to interact with the FastAPI deployment backend as a frontend engine/interface, for simplicities' sake, we will only dealing with the use case where Streamlit application directly loads the predictive model downloaded from GCS. For small scale infrastructure or prototyping, this would be sufficient in terms of simplicity and efficiency.
This template provides:
- a Python script (
src/streamlit.py) - a Dockerfile for containerised executions (
docker/{{cookiecutter.repo_name}}-streamlit.Dockerfile) - a Polyaxon config file for spinning up a Streamlit service (
aisg-context/polyaxon/polyaxonfiles/streamlit.yml)
Local Execution¶
To run the Streamlit app locally, one of course has to download a predictive model into the local machine:
$ export PRED_MODEL_UUID="<MLFLOW_EXPERIMENT_UUID>"
$ export PRED_MODEL_GCS_URI="gs://{{cookiecutter.repo_name}}-artifacts/mlflow-tracking-server/$PRED_MODEL_UUID"
$ gsutil cp -r $PRED_MODEL_GCS_URI ./models
$ $Env:PRED_MODEL_UUID='<MLFLOW_EXPERIMENT_UUID>'
$ $PRED_MODEL_GCS_URI="gs://{{cookiecutter.repo_name}}-artifacts/mlflow-tracking-server/$Env:PRED_MODEL_UUID"
$ gsutil cp -r $PRED_MODEL_GCS_URI .\models
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ export PRED_MODEL_UUID="<MLFLOW_EXPERIMENT_UUID>"
$ export PRED_MODEL_GCS_URI="gs://{{cookiecutter.repo_name}}-artifacts/mlflow-tracking-server/{{cookiecutter.author_name}}/$PRED_MODEL_UUID"
$ gsutil cp -r $PRED_MODEL_GCS_URI ./models
$ $Env:PRED_MODEL_UUID='<MLFLOW_EXPERIMENT_UUID>'
$ $PRED_MODEL_GCS_URI="gs://{{cookiecutter.repo_name}}-artifacts/mlflow-tracking-server/{{cookiecutter.author_name}}/$Env:PRED_MODEL_UUID"
$ gsutil cp -r $PRED_MODEL_GCS_URI .\models
{% endif %} PRED_MODEL_UUID is the unique ID associated with the MLFLow run that generated the predictive model to be used for dashboarding.
Spin up the Streamlit application locally:
$ export PRED_MODEL_PATH="$PWD/models/$PRED_MODEL_UUID/artifacts/model/data/model"
$ streamlit run src/streamlit.py -- \
hydra.run.dir=. hydra.output_subdir=null hydra/job_logging=disabled \
inference.model_path=$PRED_MODEL_PATH
$ $Env:PRED_MODEL_PATH="$(Get-Location)\models\$Env:PRED_MODEL_UUID\artifacts\model\data\model"
$ streamlit run src/streamlit.py -- `
hydra.run.dir=. hydra.output_subdir=null hydra/job_logging=disabled `
inference.model_path=$Env:PRED_MODEL_PATH
The application would look like the screenshot below:
Reference(s):
Docker Container¶
To use the Docker image, first build it:
$ docker build \
-t asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0 \
--build-arg PRED_MODEL_UUID="$PRED_MODEL_UUID" \
-f docker/{{cookiecutter.repo_name}}-streamlit.Dockerfile \
--platform linux/amd64 .
$ docker build `
-t asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0 `
--build-arg PRED_MODEL_UUID="$Env:PRED_MODEL_UUID" `
-f docker/{{cookiecutter.repo_name}}-streamlit.Dockerfile `
--platform linux/amd64 .
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ docker build \
-t asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0 \
--build-arg PRED_MODEL_UUID="$PRED_MODEL_UUID" \
-f docker/{{cookiecutter.repo_name}}-streamlit.Dockerfile \
--platform linux/amd64 .
$ docker build `
-t asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0 `
--build-arg PRED_MODEL_UUID="$Env:PRED_MODEL_UUID" `
-f docker/{{cookiecutter.repo_name}}-streamlit.Dockerfile `
--platform linux/amd64 .
{% endif %} After building the image, you can run the container like so:
$ sudo chgrp -R 2222 outputs
$ docker run --rm -p 8501:8501 \
--name streamlit-app \
--env GOOGLE_APPLICATION_CREDENTIALS=/var/secret/cloud.google.com/gcp-service-account.json \
-v <PATH_TO_SA_JSON_FILE>:/var/secret/cloud.google.com/gcp-service-account.json \
-v $PWD/models:/home/aisg/from-gcs \
asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0
$ docker run --rm -p 8501:8501 `
--name streamlit-app `
--env GOOGLE_APPLICATION_CREDENTIALS=/var/secret/cloud.google.com/gcp-service-account.json `
-v "<PATH_TO_SA_JSON_FILE>:/var/secret/cloud.google.com/gcp-service-account.json" `
-v "$(Get-Location)\models:/home/aisg/from-gcs" `
asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ sudo chgrp -R 2222 outputs
$ docker run --rm -p 8501:8501 \
--name streamlit-app \
--env GOOGLE_APPLICATION_CREDENTIALS=/var/secret/cloud.google.com/gcp-service-account.json \
-v <PATH_TO_SA_JSON_FILE>:/var/secret/cloud.google.com/gcp-service-account.json \
-v $PWD/models:/home/aisg/from-gcs \
asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0
$ docker run --rm -p 8501:8501 `
--name streamlit-app `
--env GOOGLE_APPLICATION_CREDENTIALS=/var/secret/cloud.google.com/gcp-service-account.json `
-v "<PATH_TO_SA_JSON_FILE>:/var/secret/cloud.google.com/gcp-service-account.json" `
-v "$(Get-Location)\models:/home/aisg/from-gcs" `
asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0
{% endif %} - GOOGLE_APPLICATION_CREDENTIALS allows the container's entrypoint to download the predictive model specified during build time from GCS. - -v <PATH_TO_SA_JSON_FILE>:/var/secret/cloud.google.com/gcp-service-account.json attaches the JSON file for the service account credentials to the Docker container. - -v $PWD/models:/home/aisg/from-gcs allows the models downloaded to the host machine to be used by the container after being mounted to /home/aisg/from-gcs.
To stop the container:
$ docker container stop streamlit-app
Integration with Polyaxon¶
Attention
As this mode of deployment would take up resources in a long-running manner, please tear the service down through the dashboard once you've gone through this part of the guide.
From the Docker build section, push the Docker image to GCR: {% if cookiecutter.gcr_personal_subdir == 'No' %}
$ docker push asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0
$ docker push asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0
$ polyaxon run \
-f aisg-context/polyaxon/polyaxonfiles/streamlit-service.yml \
-P DOCKER_IMAGE="asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0" \
-p {{cookiecutter.repo_name}}-<YOUR_NAME>
$ polyaxon run `
-f aisg-context/polyaxon/polyaxonfiles/streamlit-service.yml `
-P DOCKER_IMAGE="asia.gcr.io/{{cookiecutter.gcp_project_id}}/streamlit:0.1.0" `
-p {{cookiecutter.repo_name}}-<YOUR_NAME>
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ polyaxon run \
-f aisg-context/polyaxon/polyaxonfiles/streamlit-service.yml \
-P DOCKER_IMAGE="asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0" \
-p {{cookiecutter.repo_name}}-<YOUR_NAME>
$ polyaxon run `
-f aisg-context/polyaxon/polyaxonfiles/streamlit-service.yml `
-P DOCKER_IMAGE="asia.gcr.io/{{cookiecutter.gcp_project_id}}/{{cookiecutter.author_name}}/streamlit:0.1.0" `
-p {{cookiecutter.repo_name}}-<YOUR_NAME>
{% endif %} Just like with the VSCode or JupyterLab services, you can access the Streamlit service you've just spun up through the Polyaxon dashboard:
Reference(s):
Native Kubernetes Deployment (GKE)¶
Attention
As this mode of deployment would take up resources in a long-running manner, please tear it down once you've gone through this part of the guide. If you do not have the right permissions, please request assistance from your team lead or the administrators.
Similar to deploying the FastAPI server, to deploy the Streamlit dashboard on GKE, you can make use of the sample Kubernetes manifest files provided with this template:
$ kubectl apply -f aisg-context/k8s/dashboard/streamlit-deployment.yml --namespace=polyaxon-v1
$ kubectl apply -f aisg-context/k8s/dashboard/streamlit-service.yml --namespace=polyaxon-v1
To access the server, you can port-forward the service to a local port like such:
$ kubectl port-forward service/streamlit-svc 8501:8501 --namespace=polyaxon-v1
Forwarding from 127.0.0.1:8501 -> 8501
Forwarding from [::1]:8501 -> 8501
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ kubectl port-forward service/streamlit-{{cookiecutter.author_name.replace('_', '-')}}-svc 8501:8501 --namespace=polyaxon-v1
Forwarding from 127.0.0.1:8501 -> 8501
Forwarding from [::1]:8501 -> 8501
{% endif %}
Attention
Please tear down the deployment and service objects once they are not required.
$ kubectl delete streamlit-deployment --namespace=polyaxon-v1
$ kubectl delete streamlit-svc --namespace=polyaxon-v1
{% elif cookiecutter.gcr_personal_subdir == 'Yes' %}
$ kubectl delete streamlit-{{cookiecutter.author_name.replace('_', '-')}}-deployment --namespace=polyaxon-v1
$ kubectl delete streamlit-{{cookiecutter.author_name.replace('_', '-')}}-svc --namespace=polyaxon-v1
{% endif %} If you do not have the right permissions, please request assistance from your team lead or the administrators.