Exporters and security CLI tools
Dockprom-AI uses open-source exporters and CLI tools. Enable only the tools needed for the host and keep optional scrape targets disabled or expected to be down until their service is installed.
Exporter reference
| Domain | Open-source integration | Default endpoint | Project |
|---|---|---|---|
| Host | Node Exporter | :9100/metrics | prometheus/node_exporter |
| PostgreSQL | PostgreSQL Exporter | :9187/metrics | prometheus-community/postgres_exporter |
| Ollama | Ollama Exporter | Implementation-specific | maravexa/ollama-exporter |
| NVIDIA GPU | DCGM Exporter | :9400/metrics | NVIDIA/dcgm-exporter |
| AMD GPU | AMD SMI Exporter | Implementation-specific | amd/amd_smi_exporter |
| Intel GPU | Device Plugins / GPU telemetry | Implementation-specific | intel/intel-device-plugins-for-kubernetes |
| OpenTelemetry Collector | Collector internal telemetry | :8888/metrics | open-telemetry/opentelemetry-collector-contrib |
| Caddy | Built-in Prometheus metrics | :2019/metrics when enabled | caddyserver/caddy |
| Falco | Falco metrics / Falcosidekick | :8765/metrics when configured | falcosecurity/falco |
Exporter metric names can change between releases. Pin exporter versions and
validate the dashboard queries against /metrics before production rollout.
NVIDIA DCGM Exporter
The nvidia Compose profile provides a placeholder service:
docker compose --profile nvidia up -d dcgm-exporter
curl --fail http://localhost:9400/metrics
This requires a supported NVIDIA driver and NVIDIA Container Toolkit. Do not enable the profile on non-NVIDIA hosts.
AMD and Intel GPU alternatives
Do not run DCGM Exporter for AMD or Intel hardware. Use the vendor's open-source telemetry tool and either expose Prometheus metrics directly or write bounded metrics into the Node Exporter textfile collector.
Install Falco
Follow the official Falco package-repository instructions for the host distribution. After adding the official repository, a Debian/Ubuntu host can install the CLI and service with:
sudo apt-get update
sudo apt-get install --yes falco
sudo systemctl enable --now falco
sudo falco --version
Configure Falco JSON output for structured Loki ingestion. Enable its metrics endpoint or deploy the open-source Falcosidekick component when Prometheus event counters are required. Keep custom rules in version control and validate them:
sudo falco --validate /etc/falco/falco.yaml
sudo falco --list
Falco requires elevated kernel visibility. Review the driver choice and host security policy before installation.
Install Trivy CLI
Use the official Aqua Security repository or download a checksummed release. After configuring the official repository on Debian/Ubuntu:
sudo apt-get update
sudo apt-get install --yes trivy
trivy --version
Run CLI scans without committing reports containing secrets:
trivy repository --scanners vuln,secret,misconfig .
trivy config compose.yaml
trivy image --severity HIGH,CRITICAL --exit-code 1 grafana/grafana:13.3.0
trivy fs --format json --output /tmp/trivy-results.json .
Trivy does not expose a Prometheus endpoint by itself. For dashboards, convert a small non-sensitive scan summary to Node Exporter textfile metrics, or ship a redacted JSON summary to Loki. Never publish discovered secret values.
Operational safety
- Use read-only database roles for Grafana.
- Bind local dashboards and exporters to loopback or the internal Docker network unless an authenticated proxy is present.
- Keep source IPs and user identifiers out of metric labels.
- Pin container images by digest for production.
- Validate downloads with upstream checksums or package signatures.
- Review exporter licenses and metric compatibility during upgrades.