Add pipeline for A/B testing stateless CNI

[alyssa1303]

This pull request introduces a new HOST_NETWORK configuration option to the clusterloader2 testing framework, allowing users to toggle between host network and pod network modes. Additionally, it adds new metrics collection for sandbox runtime performance, updates test cases, and introduces a new performance evaluation scenario for CNI benchmarking.

New HOST_NETWORK Configuration:

• Added a HOST_NETWORK parameter to control whether pods use the host's network namespace. This is configurable via CLI, YAML configurations, and Terraform inputs. (modules/python/clusterloader2/cri/README.md, modules/python/clusterloader2/cri/config/config.yaml, modules/python/clusterloader2/cri/config/deployment_template.yaml, modules/python/clusterloader2/cri/cri.py, scenarios/perf-eval/cni-ab-testing/terraform-inputs/aws.tfvars) [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]

Metrics Enhancements:

• Introduced a new metric, KubeletRunSandboxDuration, to measure pod sandbox runtime performance, including 99th, 90th, and 50th percentiles. (modules/python/clusterloader2/cri/config/kubelet-measurement.yaml, modules/python/clusterloader2/cri/cri.py) [1] [2]

Test Updates:

• Extended unit tests to validate the new HOST_NETWORK parameter, including scenarios where it is explicitly set to False. (modules/python/tests/test_cri.py) [1] [2] [3] [4]

New Performance Evaluation Scenario:

• Added a new CNI benchmarking pipeline and corresponding Terraform inputs for performance evaluation. The pipeline includes configurations for both Azure and AWS environments, with scenarios for different pod counts and network configurations. (pipelines/perf-eval/CNI Benchmark/cni-ab-testing.yml, scenarios/perf-eval/cni-ab-testing/terraform-inputs/aws.tfvars) [1] [2]

[Copilot]

Pull Request Overview

This pull request introduces a new feature to control whether pods use the host's network namespace in CNI A/B testing scenarios. The implementation adds a HOST_NETWORK parameter throughout the testing framework and includes infrastructure configurations for performance evaluation.

• Adds HOST_NETWORK parameter configuration support across CLI, YAML, and Terraform inputs
• Introduces new Prometheus metric KubeletRunSandboxDuration for sandbox operation monitoring
• Updates infrastructure configurations including Kubernetes version upgrades and AMI type changes

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
steps/engine/clusterloader2/cri/execute.yml	Adds HOST_NETWORK parameter to clusterloader2 execution step
modules/python/clusterloader2/cri/cri.py	Implements host_network parameter in configuration and CLI interface
modules/python/clusterloader2/cri/config/*.yaml	Updates templates and config to support HOST_NETWORK variable
modules/python/tests/test_cri.py	Adds comprehensive tests for host_network functionality
scenarios/perf-eval/cni-ab-testing/*	New scenario configuration files for CNI A/B testing
pipelines/perf-eval/CNI Benchmark/cni-ab-testing.yml	New pipeline for automated CNI performance testing
scenarios/perf-eval/cri-resource-consume/terraform-inputs/aws.tfvars	Updates AMI type and Kubernetes version for existing scenario

Comments suppressed due to low confidence (1)

pipelines/perf-eval/CNI Benchmark/cni-ab-testing.yml:59

• The stage name 'azure_swedencentral_stateless' suggests this stage is for stateless CNI testing, but the stage configuration appears identical to the 'azure_swedencentral_default' stage. The naming should clearly differentiate the purpose or the configurations should differ to justify separate stages.

  - stage: azure_swedencentral_stateless