Rebuild Path

Pick the lightest tier that answers your question.

Local checks need no cluster. Offline simulation needs Python dependencies. Live experiments need Kubernetes, Helm, kubectl, and Prometheus access.

Local quality gates

Run Go tests, Helm checks, Python unit tests, shell syntax checks, dashboard JSON validation.

make check

Offline MPC simulation

Generate a synthetic trace or start from the bundled trace examples, simulate the controller, inspect outputs under ignored analysis paths.

python3 -m pip install -e analysis
mpc-generate-synthetic-trace --scenario spike --out analysis/out/spike.csv
mpc-offline-sim --trace-csv analysis/out/spike.csv --out-dir analysis/out/offline/spike

Saved evidence summaries

Use curated evidence roots when local raw artifacts are present.

PYTHONPATH=analysis python3 -m mpc_autoscaler_analysis.cli.summarize_costs \
  --hpa-root experiments/thesis-evidence/main/hpa60-cpu-hpa-max70 \
  --hybrid-root experiments/thesis-evidence/main/hybrid-sa-max70-tuned \
  --out-csv /tmp/mpc-cost-summary.csv \
  --out-aggregate-csv /tmp/mpc-cost-aggregate.csv

Live cluster experiments

Deploy toy-load with Helm, then run HPA or MPC scripts against the service.

helm upgrade --install toy-load toy-load/deploy/helm/toy-load --namespace default --create-namespace
bash loadgen/scripts/run_hpa_experiment_incluster.sh spike
bash loadgen/scripts/run_mpc_experiment_incluster.sh spike

traces

Offline trace examples

The repository ships baseline step, spike, and seasonality CSV traces for local validation and offline simulation.

Use them with mpc-validate-trace before running mpc-offline-sim to check schema and timing assumptions.

Open trace examples

commands

Validate, then simulate

mpc-validate-trace \
  --trace-csv analysis/mpc_autoscaler_analysis/data/traces/baseline_spike_profile_dt15.csv

mpc-offline-sim \
  --trace-csv analysis/mpc_autoscaler_analysis/data/traces/baseline_spike_profile_dt15.csv \
  --out-dir analysis/out/offline/spike