ci/woodpecker/push/woodpecker Pipeline failed

Details

fix: use library/ Harbor project, add skills, fix pipeline secrets

- .woodpecker.yaml: image paths -> library/autojanet-{agent,dispatcher}
- .woodpecker.yaml: secret names RS_HARBOR_USER / RS_HARBOR_PASS (global)
- container/Dockerfile: restore COPY skills/, skills/ populated from opencode config
- skills/: 84 opencode skills bundled into image
- k8s/manifests: update image refs to library/

2026-05-30 15:43:14 -07:00

4.5 KiB

Raw Blame History

name	description
creating-grafana-dashboard	Use when adding a dashboard to Zoe's Grafana monitoring stack — whether importing from grafana.com or creating from scratch — including datasource UID patching, GitOps deployment via the grafana-dashboards repo, and verification.

Creating a Grafana Dashboard

Overview

Dashboards are delivered via GitOps from git@git.ctz.fyi:zoe/grafana-dashboards.git. Push to main → Woodpecker CI auto-deploys to Grafana at grafana.monitoring.ctz.fyi. The critical gotcha: any downloaded dashboard will have wrong datasource UIDs and must be patched before committing.

Stack Reference

Service	URL / Context
Grafana	grafana.monitoring.ctz.fyi (v11.6.1, Postgres backend)
Cluster	k3s `monitoring` context
Mimir (metrics)	datasource UID: `mimir`, type: `prometheus`
Loki (logs)	datasource UID: `loki`, type: `loki`
Tempo (traces)	datasource UID: `tempo`, type: `tempo`
Pyroscope (profiling)	datasource UID: `pyroscope`, type: `grafana-pyroscope-datasource`
Grafana API key	`secret/production/grafana/api-key` in OpenBao

Datasource UID Mapping (ALWAYS CHECK THIS)

What the dashboard JSON says	What to set
`type: prometheus`, any UID	`uid: "mimir"`
`type: loki`, any UID	`uid: "loki"`
`type: tempo`, any UID	`uid: "tempo"`
`type: grafana-pyroscope-datasource`, any UID	`uid: "pyroscope"`
`${DS_PROMETHEUS}` template variable	set default to `mimir`

Repo Structure

grafana-dashboards/
  dashboards/
    cilium/    # Cilium CNI dashboards
    lgtm/      # Mimir, Loki, Tempo, Pyroscope dashboards
    infra/     # Node, k8s cluster dashboards
    apps/      # Application-specific dashboards
  scripts/
    sources.sh              # upstream dashboard sources list
    update-dashboards.sh    # pull from upstream + patch UIDs
    push-to-grafana.sh      # push to live Grafana via API
  .woodpecker.yml

Path A: Import from grafana.com

# 1. Download
curl -o dashboards/<folder>/<name>.json \
  "https://grafana.com/api/dashboards/<id>/revisions/latest/download"

# 2. Patch datasource UIDs (REQUIRED — dashboard will show "No data" otherwise)
jq '
  (.templating.list[] | select(.type == "datasource") | .query) = "prometheus" |
  (.panels[].datasource | select(.type == "prometheus") | .uid) = "mimir" |
  (.panels[].targets[]? | .datasource | select(.type == "prometheus") | .uid) = "mimir"
' dashboard.json > dashboard-patched.json
mv dashboard-patched.json dashboard.json

# Repeat for loki/tempo/pyroscope as needed

# 3. Set a unique explicit UID
jq '.uid = "descriptive-slug-here"' dashboard.json > tmp.json && mv tmp.json dashboard.json

# 4. Check for UID collisions before committing
jq -r '.uid' dashboards/**/*.json | sort | uniq -d   # should output nothing

# 5. Add to sources.sh for future updates, then commit + push

Path B: Create from scratch in UI

Build panels at grafana.monitoring.ctz.fyi
Export: Dashboard → Share → Export → Save to file
Save to dashboards/<folder>/<name>.json
Verify .uid is set to a unique descriptive slug
Commit and push

For new app dashboards: check what metrics are exposed first.

# See what labels Alloy exposes for a service
kubectl --context monitoring exec -n monitoring ds/alloy -- alloy targets

# Or port-forward to the app's /metrics endpoint
kubectl port-forward svc/<app> 9090:9090
curl localhost:9090/metrics | grep -v '^#' | head -50

Deployment

Push to main triggers Woodpecker automatically. To deploy manually:

cd grafana-dashboards
GRAFANA_API_KEY=$(bao kv get -field=api-key secret/production/grafana/api-key)
./scripts/push-to-grafana.sh

Check pipeline status at ci.ctz.fyi → grafana-dashboards repo.

Verification

Go to grafana.monitoring.ctz.fyi → Dashboards → find the dashboard
All panels should show data (no "No data" panels)
If "No data": datasource UIDs weren't patched — re-run jq patch

Common Issues

Symptom	Cause	Fix
"No data" on panels	Datasource UID not patched	Re-run jq patch for that datasource type
Dashboard import fails	Duplicate UID	`jq -r '.uid' dashboards/*/.json \| sort \| uniq -d` then rename
Wrong data in panels	Wrong label matchers	Check `alloy targets` for actual label names
UID collision silently replaces existing dashboard	Forgot to set explicit UID	Always set `.uid` to unique slug before commit

4.5 KiB Raw Blame History