The Dagger engine's internal OTLP proxy returns 500 on /v1/metrics when
there's no real backend, causing ~9s retry warnings per pipeline step.
Point OTEL_EXPORTER_OTLP_ENDPOINT at Tempo to give it a real endpoint.
Also removes the stale os.environ workaround from main.py (the SDK
initializes telemetry before our module loads, so it had no effect).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Dagger engine shim sets OTEL_METRICS_EXPORTER before our module
loads, so os.environ.setdefault was a no-op. Switch to a hard override.
Remove the redundant workflow-level env var since the fix belongs in
the module.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Upgrade grafana-sidecar from 1.28.0 to 2.6.0 (the 2.x memory regression #462 is resolved; ~35MB static overhead is acceptable)
- Port build from Dockerfile to native Dagger container.py
- Add liveness/readiness probes using the new /healthz endpoint on port 8080
- Update docs to reflect container.py migration and remove stale pin note
## Test plan
- [ ] Build container: `mise run container-build-and-release grafana-sidecar`
- [ ] Update kustomization tag with new image tag
- [ ] Deploy from branch: `argocd app set grafana --revision grafana-sidecar-2.6.0 && argocd app sync grafana`
- [ ] Verify sidecar health endpoint: `kubectl exec -n monitoring <pod> -c grafana-sc-dashboard -- wget -qO- http://localhost:8080/healthz`
- [ ] Verify dashboards load in Grafana UI
- [ ] `mise run services-check`
Reviewed-on: #332
The control claimed all images came from the private registry, but 12+
services pull from external public registries. Updated description to
reflect reality and catalogued external-image categories in notes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Outdated leaf card removed; zot.md now links to new service-versions
reference card instead. Added reverse link from review-services.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace broken SSH+filesystem log retrieval with Forgejo web API
endpoint. Fix CLI to use run numbers (not task IDs), add --repo
for querying any forge repo (e.g. sporks), --limit/-n for listing
size. Document runner-logs as the way to verify build success in
CLAUDE.md and container build docs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Point miniflux kustomization at the main-built v2.2.19-138e23d image
(replacing the branch tag). Disable the OTLP metrics exporter at module
import time to prevent ~11s retry delays in CI — the env var must be set
inside the module, not the runner shell, because the SDK runs inside the
Dagger engine container.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Dagger Python SDK's OTLP metrics exporter hits a non-functional
local endpoint (500s), burning ~9s per retry cycle. Set
OTEL_METRICS_EXPORTER=none in the build-dagger CI job.
Also update navidrome kustomization to the main-SHA tag (c86b5d7).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Move Dagger module from `.dagger/` to repo root (`src/blumeops/`), rename `blumeops-ci` → `blumeops`
- Replace opaque `docker_build()` with native Dagger pipelines that surface full build errors per step
- Migrate navidrome as the first container (`containers/navidrome/container.py`)
- Upgrade navidrome from v0.60.3 to v0.61.1 (major artwork overhaul, SQLite FTS5 search, server-managed transcoding)
- Add `dagger call container-version` for CI version extraction without Dockerfile parsing
- All mise tasks (`container-list`, `container-version-check`, `container-build-and-release`) updated for hybrid mode
- Legacy `docker_build()` fallback preserved for all other containers
## Motivation
When navidrome v0.61.0 added a new Go build tag (`sqlite_fts5`), `docker_build()` showed only "exit code: 1". We had to run `docker build --progress=plain` manually to find `undefined: buildtags.SQLITE_FTS5`. Native Dagger pipelines show the full error inline.
## Container build dispatch needed
After merge, dispatch container build for navidrome:
```
mise run container-build-and-release navidrome --ref 470b4bd
```
## Deploy steps
1. Wait for container build to complete
2. Back up navidrome-data PVC (non-reversible DB migrations)
3. `argocd app set navidrome --revision main && argocd app sync navidrome`
4. Verify at https://dj.ops.eblu.me
## Future
Remaining containers migrate incrementally in follow-up PRs using the same pattern.
Reviewed-on: #330
Add flyio-tailscale (v1.94.1), flyio-nginx (1.29.6-alpine), and
flyio-alloy (v1.14.1) entries with new `fly` service type so future
upgrades go through the service-review workflow.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Tailscale :stable pulled v1.96.5 during last deploy, which returns
SERVFAIL for tailnet DNS names (no upstream resolvers set). This broke
all public routing (forge/docs/cv.eblu.me) through the Fly proxy.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
check_alert() used head -1 to display only the first firing instance,
silently swallowing additional alerts (e.g. frigate pod-not-ready was
hidden behind ollama).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The general rate limit zone used $binary_remote_addr (Fly's internal
proxy IP), causing all external clients to share one bucket. Switch to
$http_fly_client_ip to match forge_auth's correct behavior.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Pulumi code has had a forge.eblu.me CNAME since it was added, but the
doc's DNS table only listed docs and cv. Also fixed the __main__.py
description to mention CNAMEs alongside A records.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
preview.quality was at the top level (invalid); moved under record
with a valid preset (very_low). Also fix services-check to catch
Grafana "Alerting (NoData)" state which was silently passing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Bug-fix release with web UI fixes, LDAP page size, and SAML SLO
redirect. Also bumps client-go to v3.2026.2.1.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previews are ~4MB/hour at default quality (CRF 1), served over NFS from
sifaka. Reducing to CRF 8 shrinks preview files to improve review page
load times when scrubbing older footage.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Strip redundant "unifi-poller-" prefix from generated slugs, bringing
UIDs from 45-48 chars down to 32-35 chars.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C0 follow-up to #327: update from branch-SHA tags to main-SHA tags
after squash-merge rebuild.
indri: v2.18.0-f59f885
ringtail: v2.18.0-f59f885-nix
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Build kube-state-metrics v2.18.0 locally from forge mirror, replacing upstream `registry.k8s.io` image
- Dockerfile (two-stage Go build) for indri/minikube
- default.nix (buildGoModule + buildLayeredImage) for ringtail/k3s
- Both kustomization files updated with `newName` pointing to local registry
## Verification
- [x] Nix build succeeded on ringtail (`nix-build` → 10-layer image)
- [x] Dockerfile build succeeded locally (`dagger call build` → ~2min)
- [x] `container-version-check --all-files` passes (2.18.0 consistent across Dockerfile, nix, service-versions.yaml)
- [ ] CI builds container images from this branch
- [ ] Update kustomization `newTag` with SHA-tagged version from CI
- [ ] ArgoCD sync on both clusters
## Test plan
- Trigger CI build: `mise run container-build-and-release kube-state-metrics`
- Verify tags: `mise run container-list kube-state-metrics`
- Update newTag in kustomization files with CI-produced tag
- Sync ArgoCD on indri: `argocd app sync kube-state-metrics`
- Sync ArgoCD on ringtail: `argocd app sync kube-state-metrics --context=k3s-ringtail` (note: argocd uses its own auth, not kubectl context)
- Verify metrics still flowing to Prometheus
Reviewed-on: #327
Worker forks 4 Dramatiq processes each loading the full Django app
(~250MB each), hitting the 1Gi limit on startup. Ringtail has ample
RAM headroom.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
teslamate had superuser on the shared blumeops-pg cluster (which also
hosts miniflux and authentik). Downgraded to plain database owner with
extension ownership (cube, earthdistance) transferred manually so it
can still ALTER EXTENSION UPDATE. earthdistance is untrusted in PG so
DROP+CREATE would need temporary superuser escalation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Doc review found the authentik database, user, and external secret were
missing, along with the immich-pg borgmatic secret. Added Cluster column
to Users table for clarity. Set last-reviewed: 2026-04-07.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Patch upgrade with bug fixes (diff normalization, installation ID cache).
Pin the upstream manifest URL to commit SHA for supply chain integrity.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rich markup parser interprets [text] as style tags, stripping
wiki-links like [[review-compensating-controls]] to empty [].
Escape description lines with rich.markup.escape().
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Verified: tailscale serve status shows only svc:k8s, ACLs restrict
tag:flyio-target to port 443 with admin/operator ownership only,
indri has no flyio-target tag. All 10 muted findings remain valid.
Noted gap: no automated alerting on new flyio-target devices.
Tracked in Todoist as MC4 (Manual Compliance Control Check CronJob).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Resolves 4 unmuted Prowler core_seccomp_profile_docker_default
findings on alloy, immich-server, immich-machine-learning, and
immich-valkey.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New mise task fetches Prowler reports from sifaka, parses with proper
muted/unmuted distinction, shows week-over-week delta, and includes
a scaffold for Kingfisher once JSON/CSV output is available upstream.
Moved all legacy top-level reports on sifaka into date subdirectories
to match the current CronJob output structure. Updated
read-compliance-reports doc with task reference and links.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The `--exclude` flag added in #321 never existed in nix — it was
introduced broken and never tested. Replace with dynamic input
discovery: query `nix flake metadata --json` for all input names,
filter out skip_inputs (default: nixpkgs-services), pass the rest
as positional args. Also bump NIX_IMAGE 2.33.3 → 2.34.4.
Updated inputs: nixpkgs, home-manager, disko.
nixpkgs-services stays pinned.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
All Helm chart mirrors (grafana-helm-charts, connect-helm-charts,
cloudnative-pg-charts) have been deleted from forge.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The storage-provisioner is a bare Pod with no controller. If the node
restarts via Docker Desktop (rather than `minikube start`), kubelet
restores static pods but bare pods are lost. Detect this and re-run
`minikube start` to restore addons.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
