aws-infrastructure-ml-streaming

This repository defines aws infrastructure required to operate the ml-streaming showcase.

Required tooling

• aws cli https://docs.aws.amazon.com/cli/latest/userguide/install-cliv2-mac.html
• aws cli session manager plugin https://docs.aws.amazon.com/systems-manager/latest/userguide/session-manager-working-with-install-plugin.html

brew tap gjbae1212/gossm
brew install gossm terragrunt tfenv jq awscli fluxcd/tap/flux
tfenv install 0.14.1
tfenv use 0.14.1

You might run into issues because both terragrunt and tfenv depend on terraform. Some solutions to this are outlined here: gruntwork-io/terragrunt#580

Configure AWS Access

Ensure you have an aws profile configured that is named datareply.

We do not yet use role assumption for the use case, thus the only thing declared in ~/.aws/config for the datareply profile ist:

[profile datareply]
region = eu-central-1

Make additionally sure to have set proper access credentials in ~/.aws/credentials :

[datareply]
aws_access_key_id = YOU_ACCCESS_KEY_ID
aws_secret_access_key = theRespectivEAccessKey

Apply state locally

The terraform will pickup the profile called datareply and will assume you have sufficient privleges to apply the infrastructure spec. This repository will soon have an automated CD pipeline which will make this partly redundant.

cd env/dev/k3s
terragrunt apply

kubectl locally

This section explains which steps need to be done to run kubectl towards the cluster locally.

One-time config

You need to set this up once to get a local copy of the kubeconfig.

1. Run the bundled bash bin/bootstrap.sh
2. (optionaly) Run bash bin/bootstrap.sh k3s-arm-cluster to setup the arm cluster.

Connecting to the cluster

You need to run this, if you are planning to connect to the cluster via kubectl.

1. Run the bundled bash bin/connect.sh or bash bin/connect.sh k3s-arm-cluster
2. Use kubectl as usual

Connecting to kafka cluster

1. Make sure to have initialized kubernetes connectivity
2. Make sure you are connected to the cluster
3. Run bash bin/extract_kafka_clientconfig.sh k3s-arm-cluster which yields the kafka client config.

Run sample cifar10 workload

Please review the the cifar10 workload section.

Inspect model performance (disabled atm)

The installed grafana dashboard provides an indication about the model perfomance,

kubectl port-forward svc/seldon-core-analytics-grafana 3000:80 -n seldon-system

Open localhost:3000 in your webbrowser to get an indication of model performance. Login admin:password

bootstrap flux

This only needs to be done once by the cluster admin in order to bootstrap flux on the cluster and connect the app-configuration-layer-ml-streaming with it.

export GITHUB_TOKEN="TODO REPLACE BY TOKEN"
export GITHUB_USER="TODO REPLACE BY USER"
flux bootstrap github \
    --owner=DataReply \
    --repository=app-configuration-layer-ml-streaming \
    --branch=master \
    --kubeconfig k3s-cluster.yaml \
    --context=default \
    --path=clusters/dev

flux bootstrap github \
    --owner=DataReply \
    --repository=app-configuration-layer-ml-streaming \
    --branch=master \
    --kubeconfig k3s-arm-cluster.yaml \
    --context=default \
    --path=clusters/dev-arm

Tools/Frameworks planned to used

• https://github.com/dflook/terraform-github-actions
• https://github.com/fluxcd/flux2-kustomize-helm-example/tree/main/apps
• https://github.com/sagittaros/terraform-k3s-private-cloud