The first run of soak scenario 2 (docs/soak-scenarios.md): a steady 60 requests/s of lookups for two hours from a machine on the cluster’s network while the deployment was changed between image tags 10132026-7a8c8ac (new) and 10112026-85feab8 (previous).
Analysis: docs/performance-report.md, section 4.
summary.md, windows.csv, meta.json: the soak’s own output (120 one-minute windows).events.txt: the marks made while changing the deployment (the build node’s clock); events.md: what soak_analyze.py events made of them.watch.csv: the pods seen from the cluster side every 15 s (perf_tests/pod_watch.py): pods, ready, terminating, restarts, memory, CPU used, throttling.pods.jsonl (per-pod detail, 2.7 MB) and canaries.jsonl (600 known-answer checks, all correct); the maintainers keep them.