Field Notes #3 · TL;DR — You deployed. Errors spiked. Someone opens kubectl, someone opens Grafana, someone blames the last commit. PerfSage SignalPilot runs an observe → correlate → explain → recommend → verify loop across K8s API, metrics-server, logs, cAdvisor, Prometheus, and optional git — then ranks findings with copy-paste
kubectl fixes. Open source. MIT. Landing page.
Field Notes #4 — For the why behind SignalPilot (war rooms, MTTR, expensive-tool gap), read I Got Tired of 3-Hour Post-Deploy War Rooms.
The question every deploy review should answer
“Why are errors and performance degradation happening after my last deployment?”
That question is simple. Getting a defensible answer in under five minutes is not.
kubectl describe shows one pod. Grafana shows a metric spike. Git shows a commit. None of them cite each other.
What SignalPilot does differently
SignalPilot fuses cross-source evidence into deterministic RCA rules:
| Rule | What it correlates | Typical fix |
|---|---|---|
oom_killed | OOMKilled + memory near limit | Raise memory limit |
cpu_throttled | CFS throttle + latency regression | Raise CPU request/limit |
crash_loop | CrashLoopBackOff + logs + config diff | Rollback or fix env |
code_regression | New log fingerprints + git suspect commit | Investigate commit |
Each finding cites multiple signal types — not a single chart anomaly.
Quick start
git clone https://github.com/perfsage/signalpilot
cd signalpilot && pip install -e .
kubectl apply -f deploy/signalpilot-rbac.yaml
signalpilot analyze my-namespace --deployment my-app --output report.htmlCI gate (exit 1 on HIGH+ findings):
signalpilot gate my-namespace --deployment my-app --junit-xml results.xmlFull docs on the SignalPilot landing page and GitHub README.
The PerfSage ladder: test → gate → RCA
- Reveal — JMeter JTL analysis in the lab (/reveal/)
- SLO Reporter — CI gates on load tests (/slo-plugin/)
- SignalPilot — post-deploy RCA in production (/signalpilot/)
Field Notes #3 · By Aashish Bajpai