Ember Production Deployment

This skill covers running Ember in production: headless daemon mode, Prometheus metrics export, and Docker deployment. For JSON output, scripting, CI integration, and ember diff/ember wait, see the ember-json skill.

Daemon Mode

The --daemon flag disables the TUI and runs Ember as a background metrics collector. It requires --expose to serve a Prometheus endpoint.

ember --expose :9191 --daemon

This starts an HTTP server with:

• /metrics: Prometheus text format (v0.0.4)
• /healthz: JSON health check

Combined with the TUI

You can also keep the TUI while exposing metrics:

ember --expose :9191

Structured logging

For log aggregation (ELK, Loki, etc.), use JSON log format:

ember --expose :9191 --daemon --log-format json

Logs go to stderr. When Ember loses Caddy connectivity, it logs one error immediately then suppresses duplicates for 30 seconds to avoid flooding. A "fetch recovered" message is logged when connectivity returns.

Signals (Unix only)

# Debug dump
kill -USR1 $(pgrep ember)

# Rotate certificates
kill -HUP $(pgrep ember)

Prometheus Metrics

Exported metrics

Caddy host metrics:

FrankenPHP metrics:

Process metrics:

Custom metric prefix

Add a prefix to all metric names to avoid collisions:

ember --expose :9191 --metrics-prefix myapp

This turns ember_host_rps into myapp_ember_host_rps, frankenphp_threads_total into myapp_frankenphp_threads_total, etc.

Authentication

Protect the metrics endpoint with HTTP Basic Auth:

ember --expose :9191 --daemon --metrics-auth admin:secret

Prefer the environment variable to avoid exposing credentials in ps output:

export EMBER_METRICS_AUTH=admin:secret
ember --expose :9191 --daemon

Prometheus scrape configuration

scrape_configs:
  - job_name: ember
    scrape_interval: 5s
    static_configs:
      - targets: ["localhost:9191"]

With authentication:

scrape_configs:
  - job_name: ember
    scrape_interval: 5s
    basic_auth:
      username: admin
      password: secret
    static_configs:
      - targets: ["localhost:9191"]

Health endpoint

GET /healthz returns JSON:

• 200 OK: {"status": "ok", "last_fetch": "...", "age_seconds": 1.2}, data is fresh
• 503: {"status": "stale", ...}, data older than 3x the polling interval (minimum 5s)
• 503: {"status": "no data yet"}, no data collected yet

Use /healthz as a Kubernetes liveness probe to detect when Ember loses contact with Caddy.

Docker Deployment

The Ember image is published to Docker Hub as alexandredaubois/ember and mirrored on GHCR as ghcr.io/alexandre-daubois/ember. It is built from scratch: no OS, no shell, just the static binary and CA certificates.

Default behavior: --daemon --expose :9191

Host network (simplest)

docker run --rm --network host alexandredaubois/ember

Custom address

docker run --rm alexandredaubois/ember \
  --daemon --expose :9191 --addr http://caddy:2019

Docker Compose (sidecar pattern)

services:
  caddy:
    image: caddy:latest
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile

  ember:
    image: alexandredaubois/ember
    network_mode: "service:caddy"
    depends_on:
      - caddy

With network_mode: "service:caddy", Ember shares Caddy's network namespace and can reach localhost:2019 directly.

Custom flags in Docker

Override the default CMD by appending flags:

docker run --rm --network host alexandredaubois/ember \
  --daemon --expose :9191 --interval 2s --metrics-prefix myapp

The image has no shell: use docker logs to read Ember's stderr output.

Environment Variables

All main flags have environment variable equivalents, convenient for containers where flags are less practical:

EMBER_METRICS_AUTH is recommended over the flag to avoid leaking credentials in ps output.