kagent Runtime Benchmark: Go vs Python

Benchmark scripts comparing the Go and Python agent runtimes in kagent.

Measures three key metrics:

• Container image size — pulled from the Kind node's image store
• Pod startup time — time from kubectl apply to pod Ready condition (averaged over N runs)
• Runtime memory — working set bytes at idle (via metrics-server or cgroup fallback)

Prerequisites

• Kind cluster with kagent deployed
• Go and Python agent images available in the cluster
• kubectl, docker, python3 on your PATH
• (Optional) metrics-server for memory measurement

Quick Start

# Install kagent (v0.8.0-beta5+)
helm install kagent-crds oci://ghcr.io/kagent-dev/kagent/helm/kagent-crds --version 0.8.0-beta5 --namespace kagent --create-namespace
helm install kagent oci://ghcr.io/kagent-dev/kagent/helm/kagent --version 0.8.0-beta5 --namespace kagent

# Run the benchmark (default: 3 runs)
./bench.sh

# More runs for better averages
./bench.sh --runs 5

How It Works

The script creates minimal declarative agents — one with runtime: go, one with runtime: python — using a dummy ModelConfig. No real LLM calls are made; the agents just need to start up and pass their readiness probe (/.well-known/agent-card.json).

See manifests.yaml for the Kubernetes resources used.

Sample Output

===========================================
         Go vs Python Runtime Summary
===========================================
  Metric              Go            Python
  --------------------------------------------------
  Image Size          29.7MB        335.4MB
  Startup Time        2707ms        18194ms
  Memory (idle)       7Mi           253Mi
===========================================