Argo Rollouts Analysis using AI

Example Argo Rollouts Analysis that uses AI services to do the canary analysis.

This job can be executed as a Argo Rollouts canary analysis job. It will fetch the logs from pods with the label role=stable and role=canary and use an AI model to compare both logs to decide whether the canary promotion should continue or not.

The result is returned from the model as a json file like:

{
    "text": "The canary version initially had issues pulling the required image, leading to Bad Requests. However, the issue seems to have been resolved and the canary started successfully later on.  Further monitoring is recommended to ensure stability.",
    "promote": true,
    "confidence": 80
}

or

{
    "text": "Canary deployment encountered Bad Request errors (HTTP 400) due to image pulling failures, indicating a problem with the canary version's image or configuration.  The stable version did not exhibit these errors.",
    "promote": false,
    "confidence": 95
}

Tested with:

• Google gemini-1.5-pro-latest
• Google gemini-1.5-flash-latest

Usage

Argo Rollouts allows different sources for analysis, one of them being the execution of a job.

In your Rollout object set the spec.metrics.provider.job to something like the following. Full example.

        metadata:
          annotations:
            service: "{{args.service}}"
          labels:
            service: "{{args.service}}"
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 1200
          template:
            metadata:
              labels:
                app: success-rate-job-ai
            spec:
              restartPolicy: Never
              containers:
              - name: ai
                image: us-central1-docker.pkg.dev/api-project-642841493686/github/argo-rollouts-ai:580503203f0460f299e1f4de569b5876c597a37b
                imagePullPolicy: IfNotPresent
                env:
                - name: GOOGLE_API_KEY
                  valueFrom:
                    secretKeyRef:
                      name: success-rate-job-ai
                      key: google_api_key
                volumeMounts:
                - name: empty
                  mountPath: /logs
                # - name: code
                #   mountPath: /app
                resources:
                  requests:
                    memory: "32Mi"
                    cpu: "250m"
                  limits:
                    memory: "64Mi"
                    cpu: "500m"
              volumes:
              - name: empty
                emptyDir: {}
              # - name: code
              #   configMap:
              #     name: canary-demo-6mm6k4ffck

Different models can be passed to the app, ie.

python app.py gemini-1.5-flash-latest

Development Setup

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

export GOOGLE_API_KEY=...

Adding dependencies

pip install langchain-google-genai
python3 -m pip freeze > requirements.txt

Langchain

tutorial

Running

python app.py https://example.com

Running in Docker

make build
docker run -ti --rm \
    -e GOOGLE_API_KEY \
    csanchez/argo-rollouts-ai

Building docker images

make build

Notes

Google Cloud credits are provided for this project.

#AISprint