Skip to content

Observability

Cloud: Groundcover first. Errors also in Sentry. HPA still needs Prometheus / metrics-server.

SignalLocalCloud
Errors / tracesGlitchTip SENTRY=1 :8000Sentry + Groundcover
MetricsOTel OTEL=1 :8889Groundcover + kube-prometheus-stack
Logsmake logs-localGroundcover sensor, not app-OTLP

Where to look

Develop overlay → UI cluster Development. Prod → Production. Perf has no extra install — you watch the develop cluster while load hits ns performance.

OTLP logs from apps are dropped (nop). Logs in Groundcover are from the node sensor. Sampling is low (~5% OTEL).

Grafana sits behind oauth2-proxy: grafana.bigbash.life / grafana.bigbash.site.

Alerts (what the repos actually show)

PathIn the repos
GroundcoverPrimary cloud APM. Terraform IAM rfe-infra/groundcover. GitOps DaemonSet apps/groundcover
SentryProduct errors from app SDKs
Grafana / Prometheuskube-prometheus-stack. HPA CPU/memory. Grafana HTTPRoute patched per env
AlertmanagerChart values include Google Chat routes, but alertmanager.enabled: false in kube-prometheus-stack/base/values.yaml. Do not treat Google Chat as a live pager from that chart until someone enables it
GitHub ActionsTrivy / ship can notify Chat. alerts-deploy.yml deploys an Apps Script — separate from in-cluster Alertmanager
SLOs / on-call rosterNot defined in these four repos. Chaos Mesh README mentions PagerDuty; that is not confirmed as the prod path

Do not paste webhook URLs from Helm values into this site.

Kubecost + Chaos Mesh: develop only.

Gotchas: local OTel is metrics-oriented (traces belong to GlitchTip/Sentry). A zero in GlitchTip is not a pass — CrossSlot on shared Redis looks like “quiet”. Scale on db_pool_* / redis_pool_*.

CI release traces: ground-cover-push-traces.yaml. Inventory: rfetech-github-actions/docs/RFE-Metrics-Inventory.md.

Troubleshooting · Visual map

BigBash architecture SSOT — hosted for the team; edit via pull requests.