Canary Deployments with Analysis

An Analysis step allows you to perform various checks and tests on your application during or after deployment. This essentially allows you to implement progressive delivery while ensuring stability and performance.
Getting Started
Let's assume you have already configured an application for a Canary Rollout using the Canary Deployments Documentation.
Create an AnalysisTemplate, which contains the analysis you want to carry out. In this sample, we'll query OpsVerse MetricServer to run a PromQL query
YAML
apiVersion: argoproj.io/v1alpha1
kind: AnalysisTemplate
metadata:
  name: api-probe-threshold
spec:
  args:
  - name: job-name
  metrics:
  - name: api-probe-threshold
    # The successCondition can be modified as required
    successCondition: result[0] < 0.5
    failureLimit: 3
    count: 5
    interval: 1m
    provider:
      prometheus:
        headers:
          - key: Authorization
            value: Basic dXNlcm5hbWU6cGFzc3dvcmQ= # Base64 of `username:password` for OpsVerse Metrics
        address: <opsverseMetricsHostURL>
        timeout: 20
        # This query can be modified as required
        query: |
          avg_over_time(probe_duration_seconds{job="{{args.job-name}}"}[5m])

apiVersion: argoproj.io/v1alpha1
kind: AnalysisTemplate
metadata:
  name: api-probe-threshold
spec:
  args:
  - name: job-name
  metrics:
  - name: api-probe-threshold
    # The successCondition can be modified as required
    successCondition: result[0] < 0.5
    failureLimit: 3
    count: 5
    interval: 1m
    provider:
      prometheus:
        headers:
          - key: Authorization
            value: Basic dXNlcm5hbWU6cGFzc3dvcmQ= # Base64 of `username:password` for OpsVerse Metrics
        address: <opsverseMetricsHostURL>
        timeout: 20
        # This query can be modified as required
        query: |
          avg_over_time(probe_duration_seconds{job="{{args.job-name}}"}[5m])
﻿
Note the configuration we are defining for the analysis. We will be making the avg_over_time query every 1m for 5 mins. Our success condition is for the result to be < 0.5, and if the analysis fails 3 times or more, Argo Rollout will fail the analysis.
Then add an Analysis step in your Canary Rollout pipeline.
YAML
canary:
      steps:
      - setWeight: 10
      .
      .
      - analysis:
          templates:
          - templateName: api-probe-threshold
          args:
          # Set variable values needed for the Analysis
          - name: job-name
            value: my-job
      .
      .
      - setWeight: 80
canary:
      steps:
      - setWeight: 10
      .
      .
      - analysis:
          templates:
          - templateName: api-probe-threshold
          args:
          # Set variable values needed for the Analysis
          - name: job-name
            value: my-job
      .
      .
      - setWeight: 80
﻿
Now, as the Argo Rollout is executing and reaches the analysis step, it will execute the analysis step listed in previous step and will proceed with the rollout only if the analysis step is passed.
If the analysis fails, the rollout will be aborted, and the canary weight will be set back to 0, and the rollout will end up in a Degraded state.
Argo Rollouts supports a variety of Analysis sources that can leverage datasources ranging from proprietary tools like Datadog and Cloudwatch, to OSS tools like Graphite and Prometheus, and even customisable jobs and Web APIs. For more details, check out the Argo Rollouts Progressive Delivery & Rollouts documentation
﻿
Did this page help you?
Canary Deployments
Sync Windows