phoenix-ops

Phoenix Operations and Deployment (Elixir/BEAM)

Production-ready Phoenix apps rely on releases, runtime configuration, telemetry, clustering, and secure endpoints. The BEAM enables rolling restarts and supervision resilience when configured correctly.

Releases and Runtime Config

MIX_ENV=prod PHX_SERVER=true mix assets.deploy
MIX_ENV=prod mix release
_build/prod/rel/my_app/bin/my_app eval "IO.puts(:os.type())"
_build/prod/rel/my_app/bin/my_app start

config/runtime.exs for env-driven settings:

config :my_app, MyApp.Repo,
  url: System.fetch_env!("DATABASE_URL"),
  pool_size: String.to_integer(System.get_env("POOL_SIZE", "10")),
  ssl: true

config :my_app, MyAppWeb.Endpoint,
  url: [host: System.fetch_env!("PHX_HOST"), port: 443, scheme: "https"],
  http: [ip: {0,0,0,0}, port: String.to_integer(System.get_env("PORT", "4000"))],
  secret_key_base: System.fetch_env!("SECRET_KEY_BASE"),
  server: true

Secrets

• Prefer env vars or secret stores (AWS/GCP KMS, Vault); avoid embedding in configs.
• Generate SECRET_KEY_BASE with mix phx.gen.secret.

Clustering and PubSub/Presence

Add libcluster for automatic node discovery:

# mix.exs deps
{:libcluster, "~> 3.3"},
{:phoenix_pubsub, "~> 2.1"},

# application.ex
topologies = [
  dns_poll: [
    strategy: Cluster.Strategy.DNSPoll,
    config: [poll_interval: 5_000, query: "my-app.internal"],
    connect: {:net_adm, :ping}
  ]
]

children = [
  {Cluster.Supervisor, [topologies, [name: MyApp.ClusterSupervisor]]},
  {Phoenix.PubSub, name: MyApp.PubSub},
  MyAppWeb.Endpoint
]

Guidelines

• Share secret_key_base across nodes for consistent session signing.
• Use distributed PubSub for Presence; ensure node connectivity before enabling Presence-heavy features.
• For blue/green, keep cookies compatible between versions.

Telemetry, Logging, and Metrics

• Install opentelemetry_phoenix and opentelemetry_ecto for traces/metrics.
• Add Plug.Telemetry and LoggerJSON or structured logging.
• Export metrics (Prometheus/OpenTelemetry) via :telemetry_poller for VM stats (reductions, memory, schedulers).
• Set LOGGER_LEVEL=info in prod; use :debug only for troubleshooting.

HTTP and Network Hardening

• Enforce HTTPS (force_ssl), HSTS, secure cookies (same_site, secure), and proper content_security_policy.
• CORS: configure cors_plug for API origins.
• Rate limiting: apply plugs (ETS/Cachex token bucket) or edge (NGINX/Cloudflare).
• Uploads: prefer presigned URLs; limit request body size (:max_request_line_length, :max_header_value_length).

Assets and Static Delivery

• mix assets.deploy runs npm/tailwind/esbuild and digests assets.
• Serve static files via CDN/reverse proxy; ensure cache-control headers set in Endpoint.
• Disable unused watchers in production to trim image size.

Background Jobs

• Oban recommended for retries/backoff, scheduled jobs, and isolation; supervise in application.ex.
• Configure queues via runtime env; monitor with Oban Web/Pro or telemetry.
• For CPU-heavy tasks, consider pooling or external workers to avoid blocking schedulers.

Deployment Patterns

• Containers: multi-stage builds; run mix deps.get --only prod, mix compile, mix assets.deploy, then mix release.
• Systemd: run release binary as service with Environment= secrets; add Restart=on-failure.
• Fly/Gigalixir/Render: supply env vars, attach Postgres/Redis, open long-lived WebSocket ports.
• Blue/green or canary: keep DB migrations compatible; deploy code first, then run migrations; keep feature flags for schema changes.

Observability and Health

• Add /health and /ready endpoints (Repo check + PubSub/Presence check).
• Export VM metrics: run :telemetry_poller for scheduler utilization and memory.
• Alert on error rates, DB timeouts, queue depths, and VM memory.

Common Pitfalls

• Building releases without PHX_SERVER=true (endpoint won’t start).
• Missing runtime config in config/runtime.exs; relying on compile-time config for secrets.
• No cluster discovery configured → Presence inconsistencies across nodes.
• Leaving default secret_key_base or per-node keys → invalid sessions after deploy.
• Large assets without digests/CDN → slow cold loads.