Welcome to a public collection of monitoring and observability configurations for a production-grade 10+ node Proxmox environment. The dashboards are designed to help operators evaluate infrastructure patterns before they reach production.
The project follows Infrastructure as Code practices and treats self-hosted environments with the same rigor as an enterprise data center. It provides a baseline for others building complex self-hosted stacks involving high-speed networking, security meshes, and containerized services.
For security and privacy, these files have been processed through a sanitization pipeline before being published from an internal Gitea instance.
- Network Masking: Deployment addresses are replaced with documentation examples or placeholders. New exports use RFC 5737 example addresses.
- Domain Generalization: Private service domains are generalized to example domains.
- Portability: Deployment-specific data-source UIDs are replaced with import inputs such as
${DS_PROMETHEUS}. Match the query language and adapt site-specific selectors before use. - Secret Management: Live credentials and private telemetry samples are excluded from publication.
- Core Infrastructure: Real-time health of a 10+ node Proxmox cluster and Docker-GitOps workflows.
- Security & Traffic: Distributed CrowdSec bouncers, Cloudflare WAF metrics, and pfSense firewall throughput.
- Identity & Access: Authentication trends and security events from Authentik SSO.
- Media & Apps: Uptime and listener metrics for Lounge24 Radio, a self-hosted station built on AzuraCast.
- Networking: 2.5Gbps backbone metrics across UniFi and Netgate hardware.
These dashboards are exported in standard Grafana JSON format. To import them:
- Download the JSON file for the service you wish to monitor.
- In your Grafana instance, go to Dashboards -> Import.
- Upload the JSON and map the
${DS_...}variables to your local Prometheus or InfluxDB data sources.
See the dated review for measured results and limitations across all 32 dashboards. Immich now includes 74 data panels, six inventory trend graphs, account counts/sizes, recent additions, queues, containers, database, resources and backup/cloud coverage. Its read-only collector and native systemd timer are included.
The remaining-panel follow-up repairs or replaces unsupported views in 11 dashboards. Its published audit found 826 of 826 panels returning data, with no empty panel results or query errors in the checked views. Garmin now distinguishes watch uploads from actual health sample freshness; maintenance cards distinguish no recorded events from incomplete history.
InfluxDB dashboards need the correct query language: map the Flux input to a Flux connection and Garmin to InfluxQL. They are separate data sources. Example addresses, fixed host selectors, bucket names and collector paths must be adapted to the importing deployment. A missing optional metric is labelled as unavailable; it is not converted into a healthy zero.
Dashboard exports can be refreshed with the sanitizer, a private binding map and optional private literal redactions. Saved query selections are cleared. Inspect every generated diff: automated masking alone does not prove an export is safe to publish. Provisioning examples contain no production credentials.
Import Soju IRC Bouncer into the Services folder and map ${DS_PROMETHEUS} to your Prometheus data source. Enable Soju's Prometheus listener and scrape it with the job name soju; keep the metrics endpoint on a private network. The dashboard uses a 30-second minimum query step for a 30-second scrape interval. Adjust that step if you use a different scrape interval.
The dashboard covers users, connections, IRC protocol traffic, upstream errors, worker panics, process resources, uptime by instance, and scrape health. Its native metrics work with SQLite and PostgreSQL backends. The Instance selector includes targets whose scrapes are failing, and native queries stay scoped to the soju job. Scrape status reports exporter reachability; upstream IRC connection failures appear separately in the reliability panels. Graphs require newly collected samples and cannot reconstruct missing metrics history.
Configured network totals, paused and unexpectedly disconnected networks, saved channels, and currently joined channels use the optional read-only soju_admin_* textfile collector supplied with soju-tui's Grafana integration. Scrape its private exporter under soju-admin and set its soju_instance label to the matching native Soju target. These implementation-specific aggregates work with either database backend and count each user/network configuration separately. Saved channels can include parted channels; joined channels reflect live upstream joins. The dashboard excludes failed, unreachable, and older-than-120-second collections, and displays missing counts as N/A. Counts collector reports freshness. Native metrics alone continue to populate the remaining panels.
Reliability rates, Events in selected range, and the since start counters use different time spans. Repeated failed retries of one upstream each count as another attempt. A raw counter does not return to zero when a connection recovers; selected-range estimates use Prometheus increase() and are rounded for display. The lifetime totals use a neutral color because they describe historical events rather than current outages.
Networks down is the current count of enabled but disconnected configurations. The uptime card shows the shortest uptime among selected Soju processes; the uptime chart shows each process separately.
Maintained by the KDN Observability Stack contributors.