Add image pull 10 nodes scenario

[jasminetMSFT]

Summary

Adds containerd image pull metrics collection to the existing CRI module using a scrape_containerd toggle.

Approach

Instead of creating a new module, this PR extends the existing CRI infrastructure:

• Adds scrape_containerd parameter to enable containerd metrics collection
• Reuses cri-resource-consume topology
• Adds containerd-measurements.yaml config to CRI module

Metrics Collected

ContainerdCriImagePullingThroughput

Image pull throughput (MB/s) with the following aggregations:

Metric	Description
Avg	Weighted average throughput per image pull
AvgPerNode	Unweighted average - each node contributes equally
Count	Total number of successful image pulls
Perc50	50th percentile (median) throughput across nodes
Perc90	90th percentile throughput across nodes
Perc99	99th percentile throughput across nodes

Results

1. scrape_containerd=True
   Pipeline: #20260106.12 • Add 20s pod_startup_latency_threshold
   Json Result: perf-eval/image-pull-n10/image_pull_prototype/47996-233c2bdc-3a73-570f-5588-ad30a97c6402.json
2. scrape_containerd=False
   Pipeline: #20260106.27 • test with scrape_containerd=False
   Json Result: perf-eval/image-pull-n10/image_pull_prototype/48089-233c2bdc-3a73-570f-5588-ad30a97c6402.json

[Copilot]

Pull request overview

This PR adds a new ClusterLoader2-based performance test scenario to measure container image pull performance on AKS clusters. The scenario deploys 10 small deployments with 1 replica each (totaling 10 pods) using imagePullPolicy: Always to force image pulls, then collects detailed metrics from kubelet and containerd via Prometheus.

Key Changes

• New image-pull-n10 scenario with terraform configuration for 3-node default pool, 1-node Prometheus pool, and 10-node user pool
• ClusterLoader2 test configuration that measures image pulling throughput, kubelet runtime operations, and pod startup latency
• Python module with execute/collect functions following established patterns from cri and other modules
• Comprehensive unit tests for the new Python module

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
steps/topology/image-pull/*.yml	Topology step templates for validation, execution, and collection
steps/engine/clusterloader2/image_pull/*.yml	Engine-level execute and collect steps with environment configuration
modules/python/clusterloader2/image_pull/image_pull.py	Main Python module implementing CL2 config override, execution, and results collection
modules/python/tests/test_image_pull.py	Unit tests covering execute, collect, and CLI functions
modules/python/clusterloader2/image_pull/config/image-pull.yaml	Main CL2 test configuration with pod deployment and measurement steps
modules/python/clusterloader2/image_pull/config/deployment_template.yaml	Kubernetes deployment template with imagePullPolicy: Always
modules/python/clusterloader2/image_pull/config/kubelet-measurement.yaml	Kubelet runtime operation duration measurements
modules/python/clusterloader2/image_pull/config/containerd-measurements.yaml	Containerd CRI metrics for image pulling throughput and network operations
scenarios/perf-eval/image-pull-n10/terraform-inputs/azure.tfvars	Infrastructure configuration for Azure AKS cluster with 3 node pools
scenarios/perf-eval/image-pull-n10/terraform-test-inputs/azure.json	Test input parameters for scenario
scenarios/perf-eval/image-pull-n10/README.md	Documentation of scenario purpose, infrastructure, and usage

[vittoriasalim]

We have image pull benchmark at cri module, is it possible to use the same module(add the extra metrics there) instead of creating a new module

[wonderyl]

Do you have a test run result?

[johnsonshi]

Looks good so far. Please address comments.

[xgugeng]

LGTM, please request approval from the maintainers.

[vittoriasalim]

change vm size

Addressed all requested changes in the latest commits. Dismissing to unblock merge due to dependent work.