- Switch vmalert from blackhole mode to sending alerts to local
Alertmanager
- Import alerttonotify service so alerts route to NATS notifications
- Move alertmanager and grafana CNAMEs from http-proxy to monitoring02
- Add monitoring CNAME to monitoring02
- Add Caddy reverse proxy entries for alertmanager and grafana
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Loki bound to localhost, Caddy reverse proxy with basic_auth
- Vault secret (shared/loki/push-auth) for password, bcrypt hash
generated at boot for Caddy environment
- Promtail dual-ships to monitoring01 (direct) and loki.home.2rjus.net
(with basic auth), conditional on vault.enable
- Terraform: new shared loki-push policy added to all AppRoles
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add standalone Loki service module (services/loki/) with same config as
monitoring01 and import it on monitoring02. Update Grafana Loki datasource
to localhost. Defer Tempo and Pyroscope migration (not actively used).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add metrics.home.2rjus.net and vmalert.home.2rjus.net CNAMEs with
Caddy TLS termination via internal ACME CA.
Refactors Grafana's Caddy config from configFile to globalConfig +
virtualHosts so both modules can contribute routes to the same
Caddy instance.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Set up the core metrics stack on monitoring02 as Phase 2 of the
monitoring migration. VictoriaMetrics replaces Prometheus with
identical scrape configs (22 jobs including auto-generated targets).
- VictoriaMetrics with 3-month retention and all scrape configs
- vmalert evaluating existing rules.yml (notifier disabled)
- Alertmanager with same routing config (no alerts during parallel op)
- Grafana datasources updated: local VictoriaMetrics as default
- Static user override for credential file access (OpenBao, Apiary)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Set interval=60s on rate() panels to match the actual Prometheus scrape
interval, so Grafana calculates $__rate_interval correctly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add gazetteer reference for country code lookup resolution.
Remove unnecessary reduce transformation. Make geomap panel
full-width (24 cols) and taller (h=10) on its own row.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Enable compactor-based retention with 30-day period to prevent
unbounded disk growth. Add basic rate limits and stream guards
to protect against runaway log generators.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use handle directive instead of path in site address for the metrics
endpoint, as the latter is deprecated in Caddy 2.10.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Configure Garage object storage on garage01 with S3 API, Vault secrets
for RPC secret and admin token, and Caddy reverse proxy for HTTPS access
at s3.home.2rjus.net via internal ACME CA. Includes flake entry, VM
definition, and Vault policy for the host.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a systemd timer that triggers builds for all hosts every 2 hours
via NATS, keeping the binary cache warm.
- Add scheduler.nix with timer (every 2h) and oneshot service
- Add scheduler NATS user to DEPLOY account
- Add Vault secret and variable for scheduler NKey
- Increase nix-cache02 memory from 16GB to 20GB
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Removed:
- hosts/nix-cache01/ directory
- services/nix-cache/build-flakes.{nix,sh} (replaced by NATS builder)
- Vault secret and AppRole for nix-cache01
- Old signing key variable from terraform
- Old trusted public key from system/nix.nix
Updated:
- flake.nix: removed nixosConfiguration
- README.md: nix-cache01 -> nix-cache02
- Monitoring rules: removed build-flakes alerts, updated harmonia to nix-cache02
- Simplified proxy.nix (no longer needs hostname conditional)
nix-cache02 is now the sole binary cache host.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Move nix-cache CNAME from nix-cache01 to nix-cache02
- Remove actions1 CNAME (service removed)
- Update proxy.nix to serve canonical domain on nix-cache02
- Promote nix-cache02 to prod tier with build-host role
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
nix-cache01 serves nix-cache.home.2rjus.net (canonical)
nix-cache02 serves nix-cache02.home.2rjus.net (for testing)
This allows testing nix-cache02 independently before DNS cutover.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Parameterize harmonia.nix to use hostname-based Vault paths
- Add nix-cache services to nix-cache02
- Add Vault secret and variable for nix-cache02 signing key
- Add nix-cache02 public key to trusted-public-keys on all hosts
- Update plan doc to remove actions runner references
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The actions runner on nix-cache01 was never actively used.
Removing it before migrating to nix-cache02.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Configure builder to build nixos-servers and nixos (gunter) repos
- Add builder NKey to Vault secrets
- Update NATS permissions for builder, test-deployer, and admin-deployer
- Grant nix-cache02 access to shared homelab-deploy secrets
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Radarr on TrueNAS jail is too old - exportarr fails on
/api/v3/wanted/cutoff endpoint (404). Keep sonarr which works.
Vault secret kept for when Radarr is updated.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add prometheus exportarr exporters for Radarr and Sonarr media
services. Runs on monitoring01, queries remote APIs.
- Radarr exporter on port 9708
- Sonarr exporter on port 9709
- API keys fetched from Vault
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Empty valid_status_codes defaults to 2xx only, not "any".
Explicitly list common status codes (2xx, 3xx, 4xx, 5xx) so
services returning 400/401 like ha and nzbget pass the probe.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Use joinByField transformation instead of merge to properly align
rows by instance. Also exclude duplicate Time/job columns from join.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard includes:
- Stat panels for endpoints monitored, probe failures, expiring certs
- Gauge showing minimum days until any cert expires
- Table of all endpoints sorted by expiry (color-coded)
- Probe status table with HTTP status and duration
- Time series graphs for expiry trends and probe success rate
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Only care about TLS handshake success for certificate monitoring.
Services like nzbget (401) and ha (400) return non-2xx but have
valid certificates.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The rules were already added to rules.yml but the blackbox.nix file
still had them, causing duplicate 'groups' key errors.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Move certificate alert rules to rules.yml instead of adding them as a
separate rules string in blackbox.nix. The previous approach caused a
YAML parse error due to duplicate 'groups' keys.
Also add policy to CLAUDE.md: never force push to master.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Enable Kanidm users to authenticate to OpenBao via OIDC for Web UI access.
Members of the admins group get full read/write access to secrets.
Changes:
- Add OIDC auth backend in Terraform (oidc.tf)
- Add oidc-admin and oidc-default policies
- Add openbao OAuth2 client to Kanidm
- Enable legacy crypto (RS256) for OpenBao compatibility
- Allow imperative group membership management in Kanidm
Limitations:
- CLI login not supported (Kanidm requires HTTPS for confidential client redirects)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The template used | min(100) | max(0) which is invalid Jinja2 syntax.
These filters expect iterables (lists), not scalar arguments. This
caused TypeError warnings on every MQTT message and left battery
sensors unavailable.
Fixed by using proper list-based min/max:
[[[value, 100] | min, 0] | max
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The 2-hour threshold was too aggressive for temperature sensors in
stable environments. Historical data shows gaps up to 2.75 hours when
temperature hasn't changed (Home Assistant only updates last_updated
when values change). Increasing to 4 hours avoids false positives
while still catching genuine failures.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When one host fetches the latest flake revision, it publishes to NATS
and all other hosts receive the update immediately. This reduces
redundant nix flake metadata calls across the fleet.
- Add nkeys to devshell for key generation
- Add nixos-exporter user to NATS HOMELAB account
- Add Vault secret for NKey storage
- Configure all hosts to use NATS for revision sharing
- Update nixos-exporter input to version with NATS support
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Loki-based dashboard for tracking NixOS operations including:
- Upgrade activity and success/failure stats
- Build activity during upgrades
- Bootstrap logs for new VM deployments
- ACME certificate renewal activity
Log panels use LogQL json parsing with | keep host to show
clean messages with host labels.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard for monitoring systemd across the fleet:
- Summary stats: failed/active/inactive units, restarts, timers
- Failed units table (shows any units in failed state)
- Service restarts table (top 15 services by restart count)
- Active units per host bar chart
- NixOS upgrade timer table with last trigger time
- Backup timers table (restic jobs)
- Service restarts over time chart
- Hostname filter to focus on specific hosts
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard for monitoring Proxmox VMs:
- Summary stats: VMs running/stopped, node CPU/memory, uptime
- VM status table with name, status, CPU%, memory%, uptime
- VM CPU usage over time
- VM memory usage over time
- Network traffic (RX/TX) per VM
- Disk I/O (read/write) per VM
- Storage usage gauges and capacity table
- VM filter to focus on specific VMs
Filters out template VMs, shows only actual guests.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard for monitoring NixOS deployments across the homelab:
- Hosts behind remote / needing reboot stat panels
- Fleet status table with revision, behind status, reboot needed, age
- Generation age bar chart (shows stale configs)
- Generations per host bar chart
- Deployment activity time series (see when hosts were updated)
- Flake input ages table
- Pie charts for hosts by revision and tier
- Tier filter variable
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard includes:
- Current temperatures per room (stat panel)
- Average home temperature (gauge)
- Current humidity (stat panel)
- 30-day temperature history with mean/min/max in legend
- Temperature trend (rate of change per hour)
- 24h min/max/avg table per room
- 30-day humidity history
Filters out device_temperature (internal sensor) metrics.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Deploy Grafana test instance on monitoring02 with:
- Kanidm OIDC authentication (admins -> Admin role, others -> Viewer)
- PKCE enabled for secure OAuth2 flow (required by Kanidm)
- Declarative datasources for Prometheus and Loki on monitoring01
- Local Caddy for TLS termination via internal ACME CA
- DNS CNAME grafana-test.home.2rjus.net
Terraform changes add OAuth2 client secret and AppRole policies for
kanidm01 and monitoring02.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Keep base groups (admins, users, ssh-users) provisioned declaratively
but manage regular users via the kanidm CLI. This allows setting POSIX
attributes and passwords in a single workflow.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Kanidm does not expose a Prometheus /metrics endpoint.
The scrape target was causing 404 errors after the TLS
certificate issue was fixed.
Also add SSH command restriction to CLAUDE.md.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Include both auth.home.2rjus.net (CNAME) and kanidm01.home.2rjus.net
(A record) as SANs in the TLS certificate. This fixes Prometheus
scraping which connects via the hostname, not the CNAME.
Fixes: x509: certificate is valid for auth.home.2rjus.net, not kanidm01.home.2rjus.net
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
