PhoenixGenApi

A framework for rapidly building backend APIs on top of Phoenix Channels and Erlang clustering. Define your business logic on any node — the framework handles routing, validation, permissions, retries, and observability. No HTTP endpoints to write, no routes to configure, no restarts to deploy.

Version: 2.18.0 | Elixir: ~> 1.18 | OTP: ~> 27 | License: MPL-2.0

Why PhoenixGenApi?

Traditional Phoenix backends require you to: define routes, write controllers, serialize responses, handle errors, and restart to deploy changes. PhoenixGenApi eliminates all of that:

• No routes, no controllers — Clients call functions by name over WebSocket. The framework routes requests to the right node automatically.
• No restarts to deploy — Register new APIs at runtime from any cluster node. The gateway picks them up instantly (push) or on the next pull interval.
• Built-in cross-cutting concerns — Permissions, rate limiting, retries, circuit breakers, hooks, and telemetry come standard. No plugs, no middleware.
• Multi-node by default — Erlang distribution means your APIs run across the cluster. The framework handles node selection, RPC, and failover.
• Real-time ready — Async responses, streaming, and relay messages (group chat) are first-class features, not afterthoughts.

How It Works

┌──────────┐     WebSocket      ┌──────────────────┐     RPC      ┌──────────────┐
│  Client  │ ◄──────────────► │  Phoenix Gateway  │ ◄──────────► │ Service Node │
└──────────┘   (Phoenix Ch.)   │  (uses this lib)  │   (Erlang)   │  (your app)  │
                               └──────────────────┘              └──────────────┘

1. Clients send requests through a Phoenix Channel.
2. The Gateway looks up the matching FunConfig, selects a node, validates arguments & permissions, then executes the remote function.
3. The Service Node runs the function and returns the result.

Service nodes can register new APIs at any time — the gateway picks them up automatically (pull) or receives them immediately (push).

Features

• Dynamic Configuration — Add, update, or remove APIs at runtime from any cluster node
• Multiple Response Modes — Sync, async, streaming, and fire-and-forget
• Node Selection — Random, hash-based, round-robin, or custom strategies
• Function Versioning — Run multiple API versions side-by-side; enable/disable per version
• Rate Limiting — Sliding-window rate limiter with global and per-API limits
• Permission System — Authenticated-only, argument-based, or role-based access control
• Retry — Configurable retry on the same node or across all nodes, with exhaustion telemetry
• Hooks — before_execute / after_execute callbacks with per-hook timeout protection (default 5s)
• Telemetry — 31 events across 5 categories for observability
• Relay Messages — Group-based message relaying with automatic cleanup on channel disconnect
• Circuit Breaker — Configurable pool-level and worker-level circuit breakers
• Structured Errors — Request.decode!/1 returns structured error codes (:invalid_payload, :missing_field)

Installation

Requires Elixir ~> 1.18 and OTP ~> 27.

def deps do
  [
    {:phoenix_gen_api, "~> 2.16"}
  ]
end

Use :libcluster to form the Erlang cluster.

Quick Start

1. Define your API on the service node

defmodule MyApp.Api do
  def get_user(user_id) do
    %{id: user_id, name: "Alice"}
  end
end

2. Create a FunConfig

A FunConfig tells the gateway how to call your function:

alias PhoenixGenApi.Structs.FunConfig

%FunConfig{
  request_type: "get_user",
  service: "user_service",
  nodes: [:"app@host"],
  choose_node_mode: :random,
  timeout: 5_000,
  mfa: {MyApp.Api, :get_user, []},
  arg_types: %{"user_id" => :string},
  response_type: :sync,
  version: "1.0.0"
}

3. Register the config on the gateway

Option A — Pull mode (gateway fetches from service node on startup):

On the service node, define a supporter module:

defmodule MyApp.GenApi.Supporter do
  alias PhoenixGenApi.Structs.FunConfig

  def get_config(_arg) do
    {:ok, [
      %FunConfig{
        request_type: "get_user",
        service: "user_service",
        nodes: [Node.self()],
        choose_node_mode: :random,
        timeout: 5_000,
        mfa: {MyApp.Api, :get_user, []},
        arg_types: %{"user_id" => :string},
        response_type: :sync,
        version: "1.0.0"
      }
    ]}
  end
end

Then configure the gateway to pull from it (see Gateway Configuration below).

Option B — Push mode (service node pushes to gateway on startup):

alias PhoenixGenApi.ConfigPusher

fun_configs = [%FunConfig{ ... }]

push_config = ConfigPusher.from_service_config(
  :user_service,
  [Node.self()],
  fun_configs,
  config_version: "1.0.0"
)

ConfigPusher.push_on_startup(:"gateway@host", push_config)

Option C — Add directly on the gateway node at runtime:

PhoenixGenApi.ConfigDb.add(%FunConfig{ ... })

4. Add PhoenixGenApi to your Channel

defmodule MyAppWeb.ApiChannel do
  use Phoenix.Channel
  use PhoenixGenApi, event: "api"

  # handle_in, handle_info for :push, :async_call, :stream_response
  # are automatically injected by `use PhoenixGenApi`
end

5. Call from the client

Send a JSON message over the channel:

{
  "service": "user_service",
  "request_type": "get_user",
  "request_id": "req_1",
  "args": { "user_id": "123" }
}

Receive the response:

{
  "request_id": "req_1",
  "result": { "id": "123", "name": "Alice" },
  "success": true,
  "async": false,
  "has_more": false
}

That's it — you have a working API gateway.

Core Concepts

FunConfig

FunConfig is the central data structure. It maps a {service, request_type} pair to a function call.

Request

The client payload is decoded into a %Request{}:

Request.decode!/1 validates the payload before deserialization:

• Payload size is checked before deserialization (default limit: 1MB, configurable via config :phoenix_gen_api, :request, max_payload_bytes: N)
• Required fields (request_type, request_id, service) must be present and non-empty
• On failure, raises PhoenixGenApi.Errors.DecodeError with a structured :code:
  • :invalid_payload — malformed data or oversized payload
  • :missing_field — one or more required fields are missing

The use PhoenixGenApi channel macro distinguishes these client errors from internal errors in its rescue block, returning "Invalid request: ..." for decode errors and "Request processing failed" for unexpected exceptions.

Response

The gateway replies with a %Response{}:

Configuration

Gateway Configuration

In the gateway node's config.exs:

config :phoenix_gen_api, :gen_api,
  pull_timeout: 5_000,
  pull_interval: 30_000,
  detail_error: false,
  service_configs: [
    %{
      service: "user_service",
      nodes: [:"app@host"],
      module: MyApp.GenApi.Supporter,
      function: :get_config,
      args: []
    }
  ]

Each service_config map supports:

Remote Node (Pull Mode)

Mark the node as a client so it doesn't start gateway-only services (RateLimiter, WorkerPool, ConfigDb, etc.):

config :phoenix_gen_api, :client_mode, true

When client_mode: true, the application starts with an empty supervision tree and no ETS tables. This is the standard setting for service/worker nodes that push configs to the gateway.

Pull startup behavior: On application start, the ConfigPuller schedules an initial pull after 1 second. If the initial pull fails (service nodes unreachable), it logs a warning and retries with exponential backoff (up to 300s). The gateway starts even if service nodes are down — APIs register on first successful pull.

Define a supporter module that returns FunConfig lists:

defmodule MyApp.GenApi.Supporter do
  alias PhoenixGenApi.Structs.FunConfig

  def get_config(_arg) do
    {:ok, my_fun_configs()}
  end

  def my_fun_configs do
    [
      %FunConfig{
        request_type: "get_user",
        service: "user_service",
        nodes: [Node.self()],
        choose_node_mode: :random,
        timeout: 5_000,
        mfa: {MyApp.Api, :get_user, []},
        arg_types: %{"user_id" => :string},
        response_type: :sync,
        version: "1.0.0"
      }
    ]
  end
end

Active Push (Remote → Gateway)

Remote nodes can push configs immediately on startup instead of waiting for a pull:

alias PhoenixGenApi.ConfigPusher
alias PhoenixGenApi.Structs.{FunConfig, PushConfig}

fun_configs = [%FunConfig{ ... }]

push_config = %PushConfig{
  service: :user_service,
  nodes: [Node.self()],
  config_version: "1.0.0",
  fun_configs: fun_configs,
  # Optional: enable periodic pull after initial push
  module: MyApp.GenApi.Supporter,
  function: :get_config,
  args: []
}

ConfigPusher.push_on_startup(:"gateway@host", push_config)

Push is idempotent — if config_version matches, the push is skipped. Use force: true to override:

ConfigPusher.push(:"gateway@host", push_config, force: true)

Verify a service's config version before pushing:

case ConfigPusher.verify(:"gateway@host", :user_service, "1.0.0") do
  {:ok, :matched} -> :already_registered
  {:ok, :mismatch, _} -> ConfigPusher.push(:"gateway@host", push_config)
  {:error, :not_found} -> ConfigPusher.push(:"gateway@host", push_config)
end

Gateway-side API:

# Receive a push (called via RPC from remote node)
{:ok, :accepted} = PhoenixGenApi.push_config(push_config)
{:ok, :skipped, :version_matches} = PhoenixGenApi.push_config(push_config)
{:ok, :accepted} = PhoenixGenApi.push_config(push_config, force: true)

# Verify a service's config version
{:ok, :matched} = PhoenixGenApi.verify_config("user_service", "1.0.0")
{:ok, :mismatch, "0.9.0"} = PhoenixGenApi.verify_config("user_service", "1.0.0")
{:error, :not_found} = PhoenixGenApi.verify_config("unknown", "1.0.0")

# Check pushed services (IEx helper)
PhoenixGenApi.pushed_services_status()

Channel Integration

defmodule MyAppWeb.ApiChannel do
  use Phoenix.Channel
  use PhoenixGenApi, event: "api"
end

Options:

The use macro injects these handlers:

• handle_in(event, payload, socket) — decodes the request via Request.decode!/1, executes it via Executor.execute!, replies to the client. Wraps errors in try/rescue to prevent channel crashes.
• handle_info({:push, result}, socket) — pushes sync results to the client
• handle_info({:stream_response, result}, socket) — pushes stream chunks
• handle_info({:async_call, result}, socket) — pushes async call results
• handle_info({:relay_message, result}, socket) — pushes relay messages to the client

It also derives JSON.Encoder for Response using the configured JSON library.

Function Versioning

Run multiple versions of the same API side-by-side:

# Version 1.0.0
%FunConfig{
  request_type: "get_user",
  service: "user_service",
  version: "1.0.0",
  mfa: {MyApp.Users, :get_user_v1, []},
  arg_types: %{"id" => :string},
  response_type: :sync
}

# Version 2.0.0 — adds a "fields" argument
%FunConfig{
  request_type: "get_user",
  service: "user_service",
  version: "2.0.0",
  mfa: {MyApp.Users, :get_user_v2, []},
  arg_types: %{"id" => :string, "fields" => :list_string},
  response_type: :sync
}

Clients request a specific version:

{
  "service": "user_service",
  "request_type": "get_user",
  "request_id": "req_1",
  "version": "2.0.0",
  "args": { "id": "123", "fields": ["name", "email"] }
}

If no version is sent, nil is used. The value "0.0.0" is reserved as a sentinel meaning "no version specified" and cannot be explicitly registered as a version. Use ConfigDb.get_latest/2 to resolve the highest enabled version.

Managing Versions

alias PhoenixGenApi.ConfigDb

# Get a specific version
{:ok, config} = ConfigDb.get("user_service", "get_user", "1.0.0")

# Get the latest enabled version
{:ok, latest} = ConfigDb.get_latest("user_service", "get_user")

# Disable a version (returns {:error, :disabled} when called)
:ok = ConfigDb.disable("user_service", "get_user", "1.0.0")

# Re-enable
:ok = ConfigDb.enable("user_service", "get_user", "1.0.0")

# Delete
:ok = ConfigDb.delete("user_service", "get_user", "1.0.0")

# List all functions with their versions
ConfigDb.get_all_functions()

Behavior:

• Disabled versions return {:error, :disabled}
• Non-existent versions return {:error, :not_found}
• get_latest/2 returns the highest enabled version number (sorted semantically via Version.compare/2). Nil-version configs are always skipped.
• get_fast/2 is an optimized lookup: returns the sole matching config directly, or the latest versioned config when multiple exist. Nil-version configs are returned when they are the sole match but excluded from "latest" resolution when versioned configs exist.
• Disabling one version does not affect others
• The value "0.0.0" is reserved — configs with this version are stored with a nil version key

Rate Limiter

Sliding-window rate limiter backed by ETS. Uses a multi-instance architecture for high concurrency — instances are supervised under a :one_for_one supervisor and requests are routed via consistent hashing (:erlang.phash2) by default.

Configuration

config :phoenix_gen_api, :rate_limiter,
  enabled: true,
  fail_open: true,
  # Multi-instance sharding (default: number of online schedulers)
  instance_count: :auto,  # :auto or positive integer
  routing_strategy: :hash,  # :hash (consistent) or :random
  global_limits: [
    # 2 000 requests per minute per user
    %{key: :user_id, max_requests: 2000, window_ms: 60_000},
    # 10 000 requests per minute per device
    %{key: :device_id, max_requests: 10000, window_ms: 60_000}
  ],
  api_limits: [
    # Expensive operation: 10 requests per minute per user
    %{
      service: "data_service",
      request_type: "export_data",
      key: :user_id,
      max_requests: 10,
      window_ms: 60_000
    }
  ]

Runtime API

alias PhoenixGenApi.RateLimiter

# Check rate limit for a request
case RateLimiter.check_rate_limit(request) do
  :ok ->
    # Proceed
    Executor.execute!(request)

  {:error, :rate_limited, details} ->
    # Reject
    %{error: "Rate limit exceeded", retry_after: details.retry_after_ms}
end

# Check global/API-specific limits directly
RateLimiter.check_rate_limit("user_123", :global, :user_id)
RateLimiter.check_rate_limit("user_123", {"my_service", "my_api"}, :user_id)

# Dynamic configuration
RateLimiter.add_global_limit(%{key: :ip_address, max_requests: 5000, window_ms: 60_000})
RateLimiter.update_config(%{enabled: true, global_limits: [...], api_limits: [...]})

Supported Keys

:user_id, :device_id, :ip_address, or any custom string key.

The rate limiter uses a sharded ETS architecture — multiple GenServer instances distribute the load. The routing strategy is :erlang.phash2/1 by default (configurable via routing_strategy). ETS tables (:rate_limiter_global and :rate_limiter_api) are shared across all instances with read_concurrency and write_concurrency enabled.

IEx Helpers

PhoenixGenApi.rl_status("user_123")   # Rate limit status for a user
PhoenixGenApi.rl_global()             # Show global limits
PhoenixGenApi.rl_config()             # Show full rate limiter config

Permission System

Four permission modes, configured per FunConfig via check_permission:

1. Disabled (false)

No check. Use for public endpoints.

%FunConfig{request_type: "search", check_permission: false}

2. Any Authenticated (:any_authenticated)

Requires a non-nil user_id.

%FunConfig{request_type: "get_profile", check_permission: :any_authenticated}

3. Argument-Based ({:arg, arg_name})

The specified argument must match user_id — users can only access their own data.

%FunConfig{request_type: "get_user_profile", check_permission: {:arg, "user_id"}}
# ✅ user_id: "user_123", args: %{"user_id" => "user_123"}
# ❌ user_id: "user_123", args: %{"user_id" => "user_999"}

4. Role-Based ({:role, allowed_roles})

User must have at least one of the specified roles.

%FunConfig{request_type: "delete_user", check_permission: {:role, ["admin", "moderator"]}}
# ✅ user_roles: ["admin"]
# ❌ user_roles: ["user"]

Notes:

• Permission checks run before argument validation and function execution
• All permission failures are logged for auditing
• Missing arguments result in denial

Relay Messages

Group-based message relaying: a user sends a message to a group, and all members (including the sender) receive it through their Phoenix Channel.

Group Types

Muted members in :strict_private groups can receive messages but cannot send them.

Architecture

• ETS table (:phoenix_gen_api_relay_groups) stores group metadata: type, members, roles, status.
• Registry (PhoenixGenApi.RelayRegistry) with :duplicate keys maps group_id to {user_id, channel_pid} for dispatching.
• Process monitoring — RelayServer monitors channel processes on join. When a channel dies, membership is auto-cleaned.

Example FunConfig Setup

# Create a group (one-time setup, e.g. in an admin API)
PhoenixGenApi.Relay.create_group("room_1", :public, "admin_user", channel_pid)

# Join a group
PhoenixGenApi.Relay.join_group("room_1", "user_2", user2_channel_pid)

# Accept a pending member (private/strict_private)
PhoenixGenApi.Relay.accept_member("room_1", "admin_user", "user_2")

# Mute a member (strict_private only)
PhoenixGenApi.Relay.mute_member("room_1", "admin_user", "user_2")

# Send a relay message (configured as FunConfig, called via WebSocket)
%FunConfig{
  request_type: "relay_msg",
  service: "chat_service",
  nodes: :local,
  mfa: {PhoenixGenApi.Relay, :handle_relay, []},
  arg_types: %{"group_id" => :string, "message" => :string},
  arg_orders: ["group_id", "message"],
  check_permission: :any_authenticated,
  response_type: :sync
}

Client → Server → All Members Flow

Client A                    Gateway (Phoenix Channel)           Client B
  │  WebSocket: relay_msg                                      │
  │  {group_id, message}                                       │
  │──────────────────────────►│                                 │
  │                            │ handle_in()                     │
  │                            │  └─ Executor.execute!(request)  │
  │                            │     └─ Relay.handle_relay()     │
  │                            │        ├─ ETS: validate member  │
  │                            │        ├─ Registry: dispatch     │
  │                            │        └─ send(pid, {:relay_message, response})
  │                            │                                 │
  │                            │ handle_info({:relay_message})  │
  │                            │  └─ push(socket, result)───────►│
  │◄───────────────────────────│                                 │
  │  {status: "relayed",       │                                 │
  │   recipients_count: 2}     │                                 │

Client WebSocket Payload

{
  "service": "chat_service",
  "request_type": "relay_msg",
  "request_id": "req_123",
  "args": {
    "group_id": "room_1",
    "message": "Hello everyone!"
  }
}

Client Receives (all members)

{
  "request_id": "req_123",
  "success": true,
  "result": {
    "group_id": "room_1",
    "from_user_id": "user_1",
    "message": "Hello everyone!",
    "timestamp": "2025-01-15T10:30:00Z"
  }
}

Direct API (outside WebSocket)

You can also manage groups programmatically:

# Create
:ok = PhoenixGenApi.Relay.create_group("room_1", :public, "admin", channel_pid)

# Join
{:ok, :active} = PhoenixGenApi.Relay.join_group("room_1", "user_2", pid)

# Leave
:ok = PhoenixGenApi.Relay.leave_group("room_1", "user_2")

# Accept (private/strict_private)
:ok = PhoenixGenApi.Relay.accept_member("room_1", "admin", "user_2")

# Mute / Unmute (strict_private only)
:ok = PhoenixGenApi.Relay.mute_member("room_1", "admin", "user_2")
:ok = PhoenixGenApi.Relay.unmute_member("room_1", "admin", "user_2")

# Inspect
{:ok, info} = PhoenixGenApi.Relay.get_group_info("room_1")

Retry

Configure retry behavior when execution fails:

%FunConfig{
  request_type: "get_data",
  service: "my_service",
  nodes: [:"node1@host", :"node2@host", :"node3@host"],
  mfa: {MyApp.Api, :get_data, []},
  retry: {:all_nodes, 3}  # Retry up to 3 times across all nodes
}

For nodes: :local, both :same_node and :all_nodes retry locally.

Retry attempts emit telemetry at [:phoenix_gen_api, :executor, :retry] with metadata including mode (:same_node or :all_nodes), type (:local or :remote), and the current attempt number in measurements.

When all retry attempts are exhausted, a [:phoenix_gen_api, :executor, :retry, :exhausted] event is emitted with the original retry configuration and the final error. This event is emitted for both local and remote retries.

Telemetry

31 events across 5 categories:

Note: The [:phoenix_gen_api, :worker_pool, :task, :rejected] event is emitted when a task is rejected due to the circuit breaker being open.

Quick Start

# Attach to all events
PhoenixGenApi.Telemetry.attach_all("my-app", fn event, measurements, metadata, _config ->
  Logger.info("[Telemetry] #{inspect(event)} #{inspect(measurements)}")
end)

# Attach to executor events only
PhoenixGenApi.Telemetry.attach_executor("my-app-exec", fn event, measurements, metadata, _config ->
  case event do
    [:phoenix_gen_api, :executor, :request, :stop] ->
      Logger.info("Request #{metadata.request_id} completed in #{measurements.duration_us}µs")
    _ ->
      :ok
  end
end)

# Built-in debug logger
PhoenixGenApi.Telemetry.attach_default_logger()

# List all available events
PhoenixGenApi.Telemetry.list_events()

# Detach when done
PhoenixGenApi.Telemetry.detach_all("my-app")

You can also use the convenience functions on the main module:

PhoenixGenApi.attach_telemetry("my-app", &MyHandler.handle_event/4)
PhoenixGenApi.detach_telemetry("my-app")

These attach to both executor and rate limiter events simultaneously.

Integration with Telemetry.Metrics

defmodule MyApp.Metrics do
  def metrics do
    [
      Telemetry.Metrics.distribution(
        "phoenix_gen_api.executor.request.duration_us",
        event_name: [:phoenix_gen_api, :executor, :request, :stop],
        measurement: :duration_us,
        tags: [:service, :request_type, :success]
      ),
      Telemetry.Metrics.counter(
        "phoenix_gen_api.executor.exceptions.count",
        event_name: [:phoenix_gen_api, :executor, :request, :exception],
        tags: [:service, :request_type]
      ),
      Telemetry.Metrics.counter(
        "phoenix_gen_api.rate_limiter.exceeded.count",
        event_name: [:phoenix_gen_api, :rate_limiter, :exceeded],
        tags: [:key, :scope]
      )
    ]
  end
end

📖 For the complete event reference, integration patterns, and best practices, see the Telemetry Guide.

IEx Helpers

PhoenixGenApi.rl_status("user_123")     # Rate limit status
PhoenixGenApi.rl_global()               # Global rate limits
PhoenixGenApi.rl_global(limits)         # Set global rate limits
PhoenixGenApi.rl_config()               # Rate limiter config
PhoenixGenApi.cache_status()            # Config cache status
PhoenixGenApi.pool_status()             # Worker pool status
PhoenixGenApi.pushed_services_status()  # Pushed services status

Related Packages

All packages in the PhoenixGenApi ecosystem:

AI Agent Support

Update usage rules from dependencies:

mix usage_rules.sync AGENTS.md --all --link-to-folder deps --inline usage_rules:all

Start the Tidewave MCP server:

mix tidewave

Connect at http://localhost:4114/tidewave/mcp. See Tidewave for details.