- Dockerfile: deterministic build from pinned CONTAINER_APP_VERSION + FEATURES
- Merges named feature branches at specific SHAs for reproducibility
- Switch CronJob to custom image with --clone-url-base and --all-organizations
- Add kingfisher to service-versions.yaml (version tracks upstream main SHA)
- Document spork container builds in new how-to card
- Document spork workflow in CLAUDE.md
- Update kingfisher service docs for custom image
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The pod-level RuntimeDefault seccomp profile (07e9c81) overrides the
DinD sidecar's privileged flag in newer Kubernetes versions, blocking
Docker daemon syscalls. Set Unconfined explicitly on the DinD container
while keeping RuntimeDefault on the runner container.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Deploys MongoDB Kingfisher as a weekly CronJob on minikube-indri
- Scans all Forgejo repos (eblume + all orgs) for leaked secrets with live validation
- Produces timestamped HTML and JSON reports on sifaka NFS (`/volume1/reports/kingfisher/`)
- Forgejo API token sourced from 1Password via ExternalSecret
- Uses official `ghcr.io/mongodb/kingfisher:1.91.0` container image
- Runs Sunday 4am (after Prowler's 3am k8s scan)
## Resources
- CronJob, PV/PVC (sifaka NFS), ExternalSecret
- ArgoCD Application with manual sync + CreateNamespace
## Test plan
- [x] Sync ArgoCD `apps` app to pick up new kingfisher Application
- [x] Set `--revision feature/kingfisher-cronjob` on kingfisher app
- [x] Verify ExternalSecret creates the `kingfisher-forgejo-token` Secret
- [x] Trigger manual job: `kubectl create job --from=cronjob/kingfisher kingfisher-manual -n kingfisher --context=minikube-indri`
- [ ] Verify reports appear on sifaka at `/volume1/reports/kingfisher/`
- [ ] After merge: set `--revision main` and re-sync
Reviewed-on: #317
Resources were under wrong Helm value keys (server.resources,
machine-learning.resources) and never applied to pods. Move to correct
bjw-s chart paths (*.controllers.main.containers.main.resources).
Increase liveness/readiness probe timeouts from 1s to 5s to prevent
kubelet from killing healthy-but-busy pods during ML inference load.
Remove CPU limits (keep requests only) to avoid throttling.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Adds a second borgmatic config (`photos.yaml`) that backs up `/Volumes/photos` (sifaka SMB mount, ~128 GB) to a dedicated BorgBase repo (`immich-photos`), running daily at 4 AM
- Separate launchd agent (`mcquack.eblume.borgmatic-photos`) so photo backups run independently from the main backup
- Refactors `borgmatic_metrics` script to support multiple repos with a `repo` Prometheus label
- Updates Grafana "Borg Backups" dashboard with a `repo` template variable so you can filter/compare repos
- Docs updated: `backups.md`, `borgmatic.md`
## Prerequisites (manual)
- [x] Create `immich-photos` repo on BorgBase with same SSH key
- [ ] Upgrade BorgBase plan to Small ($24/yr) if currently on free tier (128 GB exceeds 10 GB limit)
- [ ] After deploy: `borg init` the new repo (borgmatic does this automatically on first run)
## Test plan
- [ ] Dry run: `mise run provision-indri -- --check --diff --tags borgmatic,borgmatic_metrics`
- [ ] Deploy borgmatic role and verify both configs deployed
- [ ] Run `borgmatic --config ~/.config/borgmatic/photos.yaml create --verbosity 1` manually for first backup (will take hours)
- [ ] Verify metrics script collects from both repos: `~/.local/bin/borgmatic-metrics && cat /opt/homebrew/var/node_exporter/textfile/borgmatic.prom`
- [ ] Sync grafana-config in ArgoCD and verify dashboard repo selector works
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Reviewed-on: #315
## Summary
- Add `authentik` database (blumeops-pg cluster) to borgmatic pg_dump backups
- Add `immich` database (immich-pg cluster) to borgmatic pg_dump backups
- For immich-pg: new borgmatic managed role with `pg_read_all_data`, ExternalSecret, Tailscale LoadBalancer service, and Caddy L4 TCP proxy on port 5433
- Update backup docs to reflect all four CNPG databases + mealie SQLite
## Deploy plan
Deploy order matters — k8s resources must exist before ansible can route to them:
1. **ArgoCD (databases app):** sync to pick up immich-pg borgmatic role, ExternalSecret, and Tailscale service
```
argocd app set blumeops-pg --revision feature/borgmatic-all-pg-backups
argocd app sync blumeops-pg
```
2. **Wait** for `immich-pg-tailscale` service to get a Tailscale IP and `immich-pg.tail8d86e.ts.net` to resolve
3. **Ansible (caddy):** deploy Caddy L4 route for port 5433
```
mise run provision-indri -- --tags caddy
```
4. **Ansible (borgmatic):** deploy updated config and .pgpass
```
mise run provision-indri -- --tags borgmatic
```
5. **Verify:** trigger a manual borgmatic run and check all four pg_dump streams succeed
```
borgmatic --verbosity 1 2>&1 | grep -E '(Dumping|ERROR)'
```
## Test plan
- [x] `kubectl kustomize` builds cleanly
- [x] `ansible --check --diff` for borgmatic and caddy show expected changes
- [ ] ArgoCD sync succeeds for databases app
- [ ] `immich-pg.tail8d86e.ts.net` resolves
- [ ] `pg.ops.eblu.me:5433` accepts connections
- [ ] `borgmatic --verbosity 1` dumps all four databases without errors
Reviewed-on: #314
The 5-minute lookback window kept stale data from terminated pods
visible during rollouts, causing the alert to sit in Pending for
~5 minutes after every routine deployment. 60s still covers two
scrape cycles (30s interval) while clearing stale data much faster.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Reduced `for` from 30m to 5m and lookback window from 5m to 1m. The old
values caused alerts to linger long after apps returned to Synced state.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Upgrade External Secrets Operator from v1.3.2 (helm-chart-2.0.0) to v2.2.0
- Migrate from Helm chart deployment to static kustomize manifests, matching the repo's kustomize-first pattern
- Merge separate `-config` ArgoCD apps into the main operator apps (6 → 4 apps)
- Clean up Helm-specific labels (`helm.sh/chart`, `managed-by: Helm`)
- Update README example from v1beta1 to v1 API
## Breaking changes assessment
Low risk — v2.0.0 removed Alibaba and Device42 providers (we use neither). No templating changes affect us. All ExternalSecrets already use v1 API.
## Deployment steps
1. Sync CRDs first on both clusters (new CRD version)
2. Sync operator apps (now kustomize-based)
3. Verify ClusterSecretStore and all ExternalSecrets are healthy
4. Delete orphaned config apps: `argocd app delete external-secrets-config` and `-config-ringtail`
5. `mise run services-check`
Reviewed-on: #312
## Summary
- Add Snowflake proxy as a native systemd service on ringtail (NixOS)
- Uses `pkgs.snowflake` from nixpkgs (v2.11.0)
- Hardened systemd unit with DynamicUser, ProtectSystem=strict, 512MB memory limit
- Prometheus metrics enabled on localhost:9999
## What is Snowflake?
A Tor pluggable transport that helps censored users reach the Tor network via WebRTC. **This is NOT a Tor exit node** — traffic exits through Tor exit nodes operated by others. The proxy operator cannot see traffic content (double-encrypted) and destination servers never see the proxy's IP.
## Changes
- `nixos/ringtail/configuration.nix` — new systemd service definition
- `docs/reference/services/snowflake-proxy.md` — service reference card
- `docs/reference/infrastructure/ringtail.md` — updated systemd services section
- `service-versions.yaml` — added entry (type: nixos)
## Deploy plan
After review, deploy via `mise run provision-ringtail`. Service starts automatically.
## Test plan
- [ ] `mise run provision-ringtail` succeeds
- [ ] `ssh ringtail 'systemctl status snowflake-proxy'` shows active
- [ ] `ssh ringtail 'journalctl -u snowflake-proxy --no-pager -n 20'` shows broker connections
- [ ] `ssh ringtail 'curl -s localhost:9999/metrics'` returns Prometheus metrics
Reviewed-on: #311
The kubectl image lacks curl/python3. Use the prowler image
(which has Python) with a pure-Python urllib script instead.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prowler's --registry flag doesn't work (registry args not passed
to ImageProvider constructor, prowler-cloud/prowler PR #10128
regression). Use an init container to enumerate images from the
zot catalog API and generate an image list file instead.
See: https://github.com/eblume/prowler/tree/fix/image-provider-registry-args
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Image scan: add https:// scheme to registry URL.
IaC scan: use --scan-repository-url (Prowler clones the repo
itself), removing the need for an init container. The flag
is --scan-path for local dirs, --scan-repository-url for git.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Clone repo in init container, scan Dockerfiles and K8s manifests
with Prowler's IaC provider (Trivy). Reports written to
sifaka:/volume1/reports/prowler-iac/.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add Trivy to the Prowler container for image and IaC scanning.
New CronJob (Saturday 3am) scans all blumeops/* images in the
registry for CVEs, embedded secrets, and Dockerfile misconfigs.
Reports written to sifaka:/volume1/reports/prowler-images/.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C0 follow-up to #309: update kustomization newTag for all containers
rebuilt by the merge (authentik, authentik-redis, ntfy, alloy).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Replace upstream `docker.io/library/redis:7-alpine` (Redis 7.4.8) with a nix-built container using Redis 8.2.3 from nixpkgs
- Introduce **attached service pattern**: `parent` field in service-versions.yaml, `<parent>-<component>` naming convention, and `assert pkgs.redis.version == version` in default.nix to prevent silent version drift on `flake.lock` updates
- Document the pattern in [[review-services]] so future attached services slot in cleanly
- Backfill `parent: grafana` on existing `grafana-sidecar` entry
## Version drift protection
1. `flake.lock` update bumps nixpkgs redis → `assert` in `default.nix` breaks `nix-build`
2. Developer updates `version` in `default.nix` → prek's `container-version-check` demands matching `service-versions.yaml` update
3. Both must agree before commit succeeds
## Test plan
- [ ] Build container from branch on ringtail (`mise run container-build-and-release authentik-redis`)
- [ ] Update kustomization `newTag` to branch-built image tag
- [ ] Sync authentik ArgoCD app from branch (`argocd app set authentik --revision localize-redis && argocd app sync authentik`)
- [ ] Verify Authentik login, session persistence, and task queue still work
- [ ] After merge: C0 follow-up to update `newTag` to the main-built image tag
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Reviewed-on: #309
Bump from RC to latest stable (security fixes for config endpoint and
cross-camera auth). Add new 0.17 motion retention tier at 365 days,
reduce continuous from 180 to 30 days.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Uses the grafana_folder annotation to place the dashboard in the
existing folder created by alert rule provisioning.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two panels: currently firing alerts (firing/pending/noData/error) and
recent state changes. Refreshes every 30s. Uses Grafana's built-in
alertlist panel type — no datasource needed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Dramatiq defaults to one worker process per CPU core. On ringtail (16 cores)
this spawned 16 processes, each loading the full Django app, exceeding the
1Gi memory limit and causing a crash loop (228 restarts over 7 days).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove `group: ""` from ignoreDifferences in tailscale-operator and
tailscale-operator-ringtail — ArgoCD normalizes away the empty string
field, so the live state never matches git.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
v1.96.3 exists as a GitHub release but Docker Hub images for both
tailscale/tailscale and tailscale/k8s-operator haven't been published
yet (v1.94.2 is still latest). Revert the image tags; the fly/start.sh
`tailscale wait` improvement and review date stamps are retained.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Bump Tailscale operator, proxy containers, and init containers from v1.94.2 to v1.96.3 across both clusters (indri + ringtail via shared base kustomization)
- Replace hand-rolled `until tailscale status` polling loop in `fly/start.sh` with `tailscale wait --timeout 60s` (new in v1.96.2)
- Stamp kube-state-metrics review date (already current at v2.18.0)
## Notable upstream changes (v1.94.2 → v1.96.3)
- Go upgraded from 1.25 to 1.26
- `tailscale wait` command — blocks until daemon is running + interface has IP
- AuthKey policy now applies only when users are not logged in (behavioral change)
- Peer Relay improvements (metrics, EC2 IMDS, UDP socket scaling)
- UPnP stability fixes
## Deploy plan
1. Merge PR
2. Sync tailscale-operator on indri: `argocd app sync tailscale-operator`
3. Sync tailscale-operator on ringtail: `argocd app sync tailscale-operator-ringtail --server ringtail...`
4. Verify proxy pods roll with new image: `kubectl --context=minikube-indri -n tailscale get pods`
5. Verify ingress connectivity (spot-check a few `*.tail8d86e.ts.net` services)
6. Rebuild + deploy Fly proxy container (separate step, picks up `tailscale wait` change)
## Test plan
- [ ] ArgoCD diff looks clean for both apps before sync
- [ ] Proxy pods on indri come up healthy with v1.96.3 images
- [ ] Proxy pods on ringtail come up healthy with v1.96.3 images
- [ ] Tailscale ingress services remain reachable (e.g., grafana, prometheus)
- [ ] Fly proxy rebuild deploys successfully with `tailscale wait`
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Reviewed-on: #304
## Summary
Mikado chain to replace `mise run services-check` with Grafana Unified Alerting backed by ntfy push notifications.
**Design:**
- Grafana Unified Alerting evaluates rules against Prometheus/Loki
- ntfy webhook contact point delivers iOS notifications
- Anti-noise policy: page once per 24h per alert group
- Every alert links to a runbook in `docs/how-to/alerts/`
- services-check eventually queries the alerting API instead of doing its own probes
**Chain (bottom-up):**
1. `configure-grafana-alerting-pipeline` — enable alerting, ntfy contact point, notification policy
2. `first-alert-and-runbook` — end-to-end proof of concept with blackbox probe failure
3. `port-services-check-alerts` — migrate all services-check probes to alert rules + runbooks
4. `refactor-services-check-to-query-alerts` — rewrite services-check to query Grafana API
5. `deploy-infra-alerting` — goal card
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Reviewed-on: #303
Also catches kiwix's transmission sidecar up from v4.0.6-r4 to
v4.1.1-r1, matching the torrent service (upgraded in PR #282 but
the kiwix sidecar was missed). No breaking changes — old RPC
protocol is supported through 4.x.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Point all services at the 613f05d images which carry the new
consistent OCI labels. Skipped kiwix/transmission (old v4.0.6-r4
version, no matching build) and docs/quartz (no 613f05d build).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The top-level "uid" in Grafana dashboard JSON is at 2-space indent
near the end of the file, not the first occurrence. Match on ^ "uid"
to avoid clobbering nested datasource uid references.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The previous sed replaced ALL "uid" fields in dashboard JSON files,
including datasource references inside panels, causing dashboards to
go dark. Scope the replacement to only the first occurrence (the
top-level dashboard UID) using GNU sed 0,/pattern/ addressing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add init container to pre-populate ConfigMap dashboards before Grafana
starts, eliminating the race between the sidecar and the provisioner
that caused dashboard DB records to be deleted and re-created with new
IDs. Also stamp stable UIDs on TeslaMate and UnPoller dashboards
fetched from upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
StatefulSet volumeClaimTemplates are immutable and minikube's hostpath
provisioner doesn't enforce PVC size limits anyway. Add comments noting
the data grows freely on the 1.8TB backing disk.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
