Skip to content

ops-scripts

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

ops-scripts

README.md

ops-scripts

Operational helpers you run from your workstation: building the Lambda layer, fetching the dataset, testing the API, checking monitoring and cleaning up. The layer build is also called by stack-inference at apply time when the zip is missing; nothing else here is invoked by SageMaker, Lambda or CI/CD.

What is in here

Script Purpose
build_lambda_layer.sh Build the Pillow and NumPy layer zip for the inference Lambda (make layer)
data_download_breakhis.sh Download BreakHis (about 4 GB) into breast_benign/ and breast_malignant/, keeping the file names the patient split and fairness gate read
test_api.py Send base64 images to POST /predict with the x-api-key header and compare predictions with the folder labels (needs requests and Pillow)
ops_generate_endpoint_traffic.py Send labelled images straight to the endpoint to seed data capture for the drift job
verify_monitoring.sh Check that the CloudWatch alarms, SNS topics and EventBridge rules exist
cleanup_all.sh Destructive. Delete the endpoint, endpoint configs, models, model packages and job log groups, and empty the project buckets (never the state bucket). With --destroy it then runs terraform destroy in stack-cicd, stack-inference and stack-training (make destroy)

Typical usage

# Build the Lambda layer
./ops-scripts/build_lambda_layer.sh

# Fetch BreakHis, then upload it (the marker starts the pipeline)
./ops-scripts/data_download_breakhis.sh data/breakhis
./scripts/data_uploader.sh data/breakhis

# Call the deployed API (URL and key default to the terraform outputs)
python3 ops-scripts/test_api.py \
  --api-url "$(terraform -chdir=stack-inference output -raw api_gateway_url)" \
  --api-key "$(terraform -chdir=stack-inference output -raw api_key_value)" \
  --image-path data/breakhis/breast_benign/<file>.png

# Seed data capture for the drift job
python3 ops-scripts/ops_generate_endpoint_traffic.py --profile <profile> \
  --endpoint medical-image-classification-endpoint --requests 200

# Check monitoring after a deploy
./ops-scripts/verify_monitoring.sh

# Clean up and destroy (asks you to type the project name)
./ops-scripts/cleanup_all.sh --profile <profile> --destroy

Run ./ops-scripts/cleanup_all.sh --help for --project, --region, --state-bucket and --yes.

Naming convention

  • Shell: snake_case.sh (for example build_lambda_layer.sh)
  • Python: snake_case.py (for example test_api.py)
  • Verb-first for action scripts: build_*, cleanup_*, verify_*, test_*