prawilny / distributed-linters

Cluster setup

In order to set up the project execute the following steps:

• log into gcloud
• set project id: gcloud config set project irio-test-338916
• set docker environment: gcloud auth configure-docker europe-central2-docker.pkg.dev
• make sure you can list images in the registry: gcloud artifacts docker images list europe-central2-docker.pkg.dev/irio-test-338916/irio-test
• build container images: make -Bj
• push container images to artifact registry: make image_upload -j
• make sure you can see the images in the registry: gcloud artifacts docker images list europe-central2-docker.pkg.dev/irio-test-338916/irio-test
• deploy the application to kubernetes: kubectl create -f yaml
• deploy the monitoring to kubernetes: kubectl create -f prometheus

Tests

In order to run the tests execute the following steps:

• Check that the cluster is ready.
• Set the following environment variables with external IPs of appropriate just created services:
  • LOAD_BALANCER=IP:20000
  • MANAGER=IP:10000
  • PROMETHEUS=IP:9090
  • REGISTRY=europe-central2-docker.pkg.dev/irio-test-338916/irio-test
• build admin client: cd admin && go build
• enter test directory: cd ../test
• run the tests:
  • simple end to end test of machine manager managing linter deployments: go run test_manager.go
  • simple test showing that load balancer is working as expected:
    • prepare the environment: create linter deployments and set proportions:
      • ./admin $MANAGER add_version -image_url=$REGISTRY/python_linter -language=python -version=1.0
      • ./admin $MANAGER add_version -image_url=$REGISTRY/python_linter -language=python -version=2.0
      • ./admin $MANAGER set_proportions python 1.0 2 python 2.0 1
    • wait for the pods to start
    • run the test: go run test_loadbalancing.go
    • check that the values reported as routed to versions 1.0 and 2.0 have ratio close to 2.
    • if you want to remove the created linters:
      • set their weights to 0: ./admin $MANAGER set_proportions python 1.0 0 python 2.0 0
      • delete the deployments:
        • ./admin $MANAGER remove_version -language=python -version=1.0
        • ./admin $MANAGER remove_version -language=python -version=2.0

Witnessing autoscaling

In order to check the autoscaling out execute the following steps:

• Set up cluster (with monitoring!)
• Deploy load generating program: kubectl create -f load/load.yaml
• Use kubectl to watch pods being created
• Use Prometheus web interface (service "prometheus", port 9090) to view metric load_lints to assess the system's performance.
• It's suggested to delete the resources afterwards: kubectl delete -f load/load.yaml

Admin CLI tool usage

Arguments for the admin tool:

admin MANAGER_ADDRESS SUBCOMMAND [ARG]...

• admin MANAGER_ADDRESS list_versions
• admin MANAGER_ADDRESS add_version -image_url=IMAGE -language=LANGUAGE -version=VERSION
• admin MANAGER_ADDRESS remove_version -language=LANGUAGE -version=VERSION
• admin MANAGER_ADDRESS set_proportions [LANGUAGE VERSION PROPORTION]...