Developing Custom Workers

Extend the iii engine by building custom workers that register functions, triggers, and integrate with external systems.

Custom workers in the iii Engine allow developers to extend the core functionality of the system. A worker acts as a container for logic that can register functions, triggers, and integrate with external systems.

Introduction

Workers are dynamically loaded and configured, often utilizing an Adapter pattern to allow for swappable backend implementations (e.g., swapping an in-memory event bus for a Redis-backed one).

The engine provides a trait-based system where workers implement the CoreWorker trait for lifecycle management and the ConfigurableWorker trait for handling configuration and adapter injection.

Worker Architecture

The worker system is built around two primary traits: CoreWorker and ConfigurableWorker.

Core Traits

Trait	Description	Key Methods
CoreWorker	The base trait for all workers. Handles lifecycle, identification, and function registration.	name(), create(), initialize(), register_functions(), start_background_tasks(), destroy()
ConfigurableWorker	Extends CoreWorker to support typed configuration and pluggable adapters.	build(), registry(), adapter_name_from_config()

Lifecycle Flow

The following diagram illustrates the lifecycle of a worker from creation to initialization.

Implementing a Configurable Worker

Developing a custom worker typically involves defining an adapter interface, implementing specific adapters, and then wrapping them in a worker structure.

Step 1: Define the Adapter Trait

Define an async_trait that specifies the behavior your worker’s backend must implement. This allows users to switch implementations via configuration.

use async_trait::async_trait;
use serde_json::Value;

#[async_trait]
pub trait CustomEventAdapter: Send + Sync + 'static {
    async fn emit(&self, topic: &str, event_data: Value);
    async fn subscribe(&self, topic: &str, id: &str, function_id: &str);
    async fn unsubscribe(&self, topic: &str, id: &str);
}

Why async_trait? Rust’s async traits require this macro to handle the complexity of async function pointers.

Step 2: Implement Adapter Registration

To make adapters discoverable by the configuration system, you must define a registration struct and use the inventory crate.

use std::sync::Arc;
use std::future::Future;
use std::pin::Pin;
use iii::Engine;

pub type CustomEventAdapterFuture = Pin<
    Box<dyn Future<Output = anyhow::Result<Arc<dyn CustomEventAdapter>>> + Send>
>;

pub struct CustomEventAdapterRegistration {
    pub name: &'static str,
    pub factory: fn(Arc<Engine>, Option<Value>) -> CustomEventAdapterFuture,
}

// Implement AdapterRegistrationEntry trait
impl AdapterRegistrationEntry<dyn CustomEventAdapter> for CustomEventAdapterRegistration {
    fn name(&self) -> &'static str {
        self.name
    }

    fn factory(&self) -> fn(Arc<Engine>, Option<Value>) -> CustomEventAdapterFuture {
        self.factory
    }
}

// Register the type with inventory
inventory::collect!(CustomEventAdapterRegistration);

Purpose: This registration system allows the engine to discover and instantiate adapters dynamically based on configuration.

Step 3: Create Adapter Factories

Define factory functions that instantiate your specific adapter implementations (e.g., InMemory or Logging).

use iii::register_adapter;

fn make_inmemory_adapter(
    engine: Arc<Engine>,
    config: Option<Value>
) -> CustomEventAdapterFuture {
    Box::pin(async move {
        Ok(Arc::new(InMemoryEventAdapter::new(config, engine).await?)
            as Arc<dyn CustomEventAdapter>)
    })
}

// Register the specific adapter implementation
register_adapter!(
    <CustomEventAdapterRegistration>
    "my::InMemoryEventAdapter",
    make_inmemory_adapter
);

Step 4: Implement Adapter Logic

Create the actual adapter implementations.

In-Memory Adapter

Simple in-memory implementation for development and testing.

use std::collections::HashMap;
use tokio::sync::RwLock;

pub struct InMemoryEventAdapter {
    subscriptions: Arc<RwLock<HashMap<String, HashMap<String, String>>>>,
    engine: Arc<Engine>,
}

impl InMemoryEventAdapter {
    pub async fn new(
        _config: Option<Value>,
        engine: Arc<Engine>
    ) -> anyhow::Result<Self> {
        Ok(Self {
            subscriptions: Arc::new(RwLock::new(HashMap::new())),
            engine,
        })
    }
}

#[async_trait]
impl CustomEventAdapter for InMemoryEventAdapter {
    async fn emit(&self, topic: &str, event_data: Value) {
        let subs = self.subscriptions.read().await;
        if let Some(by_id) = subs.get(topic) {
            for function_id in by_id.values() {
                let _ = self.engine.call(function_id, event_data.clone()).await;
            }
        }
    }

    async fn subscribe(&self, topic: &str, id: &str, function_id: &str) {
        let mut subs = self.subscriptions.write().await;
        subs.entry(topic.to_string())
            .or_insert_with(HashMap::new)
            .insert(id.to_string(), function_id.to_string());
    }

    async fn unsubscribe(&self, topic: &str, id: &str) {
        let mut subs = self.subscriptions.write().await;
        if let Some(by_id) = subs.get_mut(topic) {
            by_id.remove(id);
        }
    }
}

Logging Wrapper Adapter

Wrapper adapter that logs all events while delegating to another adapter.

pub struct LoggingEventAdapter {
    inner: Arc<dyn CustomEventAdapter>,
}

#[async_trait]
impl CustomEventAdapter for LoggingEventAdapter {
    async fn emit(&self, topic: &str, event_data: Value) {
        tracing::info!(
            topic = %topic,
            event_data = %event_data,
            "Emitting event"
        );
        self.inner.emit(topic, event_data).await;
    }

    async fn subscribe(&self, topic: &str, id: &str, function_id: &str) {
        tracing::info!(topic = %topic, "Subscribing to topic");
        self.inner.subscribe(topic, id, function_id).await;
    }

    async fn unsubscribe(&self, topic: &str, id: &str) {
        tracing::info!(topic = %topic, "Unsubscribing from topic");
        self.inner.unsubscribe(topic, id).await;
    }
}

Step 5: Implement the Worker Logic

The worker struct holds the Engine reference and the injected Adapter.

use serde::Deserialize;
use once_cell::sync::Lazy;
use tokio::sync::RwLock;

#[derive(Debug, Clone, Deserialize, Default)]
#[serde(deny_unknown_fields)]
pub struct CustomEventModuleConfig {
    #[serde(default)]
    pub adapter: Option<AdapterEntry>,
}

#[derive(Clone)]
pub struct CustomEventModule {
    adapter: Arc<dyn CustomEventAdapter>,
    engine: Arc<Engine>,
    _config: CustomEventModuleConfig,
}

#[async_trait]
impl ConfigurableWorker for CustomEventModule {
    type Config = CustomEventModuleConfig;
    type Adapter = dyn CustomEventAdapter;
    type AdapterRegistration = CustomEventAdapterRegistration;
    const DEFAULT_ADAPTER_NAME: &'static str = "my::InMemoryEventAdapter";

    // Define how to access the registry
    async fn registry() -> &'static RwLock<HashMap<String, AdapterFactory<Self::Adapter>>> {
        static REGISTRY: Lazy<RwLock<HashMap<String, AdapterFactory<dyn CustomEventAdapter>>>> =
            Lazy::new(|| RwLock::new(CustomEventModule::build_registry()));

        &REGISTRY
    }

    // Builder method
    fn build(
        engine: Arc<Engine>,
        config: Self::Config,
        adapter: Arc<Self::Adapter>
    ) -> Self {
        Self {
            engine,
            _config: config,
            adapter
        }
    }
}

Registering Functions

Workers expose functionality to the engine (and thus to SDK workers) by registering functions. This is typically done in the initialize method or register_functions.

Registration Request Structure

When registering a function, you must provide a RegisterFunctionRequest.

Field	Type	Description
function_id	String	Unique function ID (e.g., “custom::emit”)
description	Option<String>	Human-readable description of the function
request_format	Option<Value>	JSON Schema defining the expected input
response_format	Option<Value>	JSON Schema defining the expected output

When using the #[service] macro with #[function] attributes, request_format and response_format are auto-generated as standard JSON Schema from your Rust types (via schemars). Your input/output types must derive JsonSchema. Manual schema specification is only needed for custom worker registration.

Example Registration

use iii::RegisterFunctionRequest;

#[async_trait]
impl CoreWorker for CustomEventModule {
    fn name(&self) -> &str {
        "custom_event"
    }

    async fn initialize(&self) -> anyhow::Result<()> {
        self.engine.register_function(
            RegisterFunctionRequest {
                function_id: "custom::emit".to_string(),
                description: Some("Emit a custom event".to_string()),
                request_format: Some(serde_json::json!({
                    "type": "object",
                    "properties": {
                        "topic": { "type": "string" },
                        "data": { "type": "object" }
                    },
                    "required": ["topic", "data"]
                })),
                response_format: None,
                metadata: None,
            },
            Box::new(self.clone()), // The handler
        );
        Ok(())
    }

    async fn register_functions(&self, _engine: Arc<Engine>) -> anyhow::Result<()> {
        // Additional function registrations can go here
        Ok(())
    }
}

Handling Function Invocations

To handle invocations, the worker (or a specific handler struct) must implement the FunctionHandler trait.

use iii::{FunctionHandler, FunctionResult};

#[async_trait]
impl FunctionHandler for CustomEventModule {
    async fn handle(&self, input: Value) -> FunctionResult {
        // 1. Parse Input
        let topic = input.get("topic")
            .and_then(|v| v.as_str())
            .ok_or_else(|| anyhow::anyhow!("Missing 'topic' field"))?;

        let data = input.get("data")
            .cloned()
            .unwrap_or(Value::Null);

        // 2. Execute Logic (using the adapter)
        self.adapter.emit(topic, data).await;

        // 3. Return Result
        FunctionResult::Success(None)
    }
}

Registering Triggers

Workers can also act as sources of events by registering TriggerTypes. This allows the engine to route external events (like Cron ticks or HTTP requests) to specific functions.

Trigger Architecture

Implementation

To support triggers, a worker implements TriggerRegistrator.

use iii::{TriggerRegistrator, Trigger, TriggerType};
use std::future::Future;
use std::pin::Pin;

impl TriggerRegistrator for CustomEventModule {
    fn register_trigger(
        &self,
        trigger: Trigger,
    ) -> Pin<Box<dyn Future<Output = Result<(), anyhow::Error>> + Send + '_>> {
        Box::pin(async move {
            // Extract configuration
            let config = trigger.config;
            let subscribes = config.get("subscribes")
                .and_then(|v| v.as_array())
                .ok_or_else(|| anyhow::anyhow!("Missing 'subscribes' array"))?;

            // Subscribe to each topic
            for topic in subscribes {
                let topic_str = topic.as_str()
                    .ok_or_else(|| anyhow::anyhow!("Invalid topic"))?;

                self.adapter.subscribe(
                    topic_str,
                    &trigger.id,
                    &trigger.function_id
                ).await;
            }

            Ok(())
        })
    }
}

Then, register the trigger type during initialization:

async fn initialize(&self) -> anyhow::Result<()> {
    // Register functions
    // ... (function registration code)

    // Register trigger type
    let trigger_type = TriggerType {
        id: "event".to_string(),
        registrator: Box::new(self.clone()),
        description: Some("Event-based trigger".to_string()),
    };

    self.engine.register_trigger_type(trigger_type).await?;

    Ok(())
}

Configuration

Workers are configured via iii-config.yaml or JSON passed during initialization. The ConfigurableWorker trait maps this configuration to a Rust struct.

Configuration Struct

#[derive(Debug, Clone, Deserialize, Default)]
#[serde(deny_unknown_fields)]
pub struct CustomEventModuleConfig {
    #[serde(default)]
    pub adapter: Option<AdapterEntry>,
}

Usage in Config File

workers:
  - name: my::CustomEventModule
    config:
      adapter:
        name: my::LoggingEventAdapter
        config:
          inner_adapter: my::InMemoryEventAdapter

Nested Adapters: The logging adapter wraps the in-memory adapter, creating a decorator pattern for cross-cutting concerns.

Complete Example

Here’s a complete custom worker implementation:

examples/custom_event_module.rs

use async_trait::async_trait;
use serde::Deserialize;
use serde_json::Value;
use std::collections::HashMap;
use std::sync::Arc;
use tokio::sync::RwLock;
use iii::{
    Engine, CoreWorker, ConfigurableWorker, FunctionHandler,
    FunctionResult, RegisterFunctionRequest, TriggerRegistrator,
    Trigger, TriggerType, AdapterEntry
};

// 1. Define Adapter Trait
#[async_trait]
pub trait CustomEventAdapter: Send + Sync + 'static {
    async fn emit(&self, topic: &str, event_data: Value);
    async fn subscribe(&self, topic: &str, id: &str, function_id: &str);
    async fn unsubscribe(&self, topic: &str, id: &str);
}

// 2. Implement In-Memory Adapter
pub struct InMemoryEventAdapter {
    subscriptions: Arc<RwLock<HashMap<String, HashMap<String, String>>>>,
    engine: Arc<Engine>,
}

impl InMemoryEventAdapter {
    pub async fn new(_config: Option<Value>, engine: Arc<Engine>) -> anyhow::Result<Self> {
        Ok(Self {
            subscriptions: Arc::new(RwLock::new(HashMap::new())),
            engine,
        })
    }
}

#[async_trait]
impl CustomEventAdapter for InMemoryEventAdapter {
    async fn emit(&self, topic: &str, event_data: Value) {
        let subs = self.subscriptions.read().await;
        if let Some(by_id) = subs.get(topic) {
            for function_id in by_id.values() {
                let _ = self.engine.call(function_id, event_data.clone()).await;
            }
        }
    }

    async fn subscribe(&self, topic: &str, id: &str, function_id: &str) {
        let mut subs = self.subscriptions.write().await;
        subs.entry(topic.to_string())
            .or_insert_with(HashMap::new)
            .insert(id.to_string(), function_id.to_string());
    }

    async fn unsubscribe(&self, topic: &str, id: &str) {
        let mut subs = self.subscriptions.write().await;
        if let Some(by_id) = subs.get_mut(topic) {
            by_id.remove(id);
        }
    }
}

// 3. Define Worker Configuration
#[derive(Debug, Clone, Deserialize, Default)]
#[serde(deny_unknown_fields)]
pub struct CustomEventModuleConfig {
    #[serde(default)]
    pub adapter: Option<AdapterEntry>,
}

// 4. Implement Worker
#[derive(Clone)]
pub struct CustomEventModule {
    adapter: Arc<dyn CustomEventAdapter>,
    engine: Arc<Engine>,
    _config: CustomEventModuleConfig,
}

#[async_trait]
impl CoreWorker for CustomEventModule {
    fn name(&self) -> &str {
        "custom_event"
    }

    async fn initialize(&self) -> anyhow::Result<()> {
        // Register emit function
        self.engine.register_function(
            RegisterFunctionRequest {
                function_id: "custom::emit".to_string(),
                description: Some("Emit a custom event".to_string()),
                request_format: Some(serde_json::json!({
                    "type": "object",
                    "properties": {
                        "topic": { "type": "string" },
                        "data": { "type": "object" }
                    },
                    "required": ["topic", "data"]
                })),
                response_format: None,
                metadata: None,
            },
            Box::new(self.clone()),
        );

        // Register trigger type
        let trigger_type = TriggerType {
            id: "event".to_string(),
            registrator: Box::new(self.clone()),
            description: Some("Event-based trigger".to_string()),
        };
        self.engine.register_trigger_type(trigger_type).await?;

        Ok(())
    }
}

#[async_trait]
impl FunctionHandler for CustomEventModule {
    async fn handle(&self, input: Value) -> FunctionResult {
        let topic = input.get("topic").and_then(|v| v.as_str()).unwrap_or("");
        let data = input.get("data").cloned().unwrap_or(Value::Null);

        self.adapter.emit(topic, data).await;
        FunctionResult::Success(None)
    }
}

impl TriggerRegistrator for CustomEventModule {
    fn register_trigger(
        &self,
        trigger: Trigger,
    ) -> std::pin::Pin<Box<dyn std::future::Future<Output = Result<(), anyhow::Error>> + Send + '_>> {
        Box::pin(async move {
            let config = trigger.config;
            let subscribes = config.get("subscribes")
                .and_then(|v| v.as_array())
                .ok_or_else(|| anyhow::anyhow!("Missing 'subscribes' array"))?;

            for topic in subscribes {
                let topic_str = topic.as_str()
                    .ok_or_else(|| anyhow::anyhow!("Invalid topic"))?;
                self.adapter.subscribe(topic_str, &trigger.id, &trigger.function_id).await;
            }

            Ok(())
        })
    }
}

iii-config.yaml

workers:
  - name: my::CustomEventModule
    config:
      adapter:
        name: my::InMemoryEventAdapter

Best Practices

Use Adapter Pattern

Always use adapters for external integrations to allow swapping implementations.

// Good: Adapter-based design
pub trait StorageAdapter {
    async fn save(&self, key: &str, value: Value);
}

// Avoid: Hard-coded implementation
pub struct Worker {
    redis: RedisClient,  // Tightly coupled
}

Implement Graceful Shutdown

Handle cleanup in the worker’s drop implementation or provide shutdown hooks.

impl Drop for CustomEventModule {
    fn drop(&mut self) {
        // Clean up resources
        tracing::info!("Shutting down CustomEventModule");
    }
}

Provide JSON Schemas

Always define request and response formats for functions to enable validation and documentation.

request_format: Some(serde_json::json!({
    "type": "object",
    "properties": {
        "email": { "type": "string", "format": "email" },
        "age": { "type": "number", "minimum": 0 }
    },
    "required": ["email"]
}))

Use Structured Logging

Use the tracing crate for structured logging with context.

tracing::info!(
    topic = %topic,
    subscriber_count = by_id.len(),
    "Emitting event to subscribers"
);

Next Steps

Core Workers

Explore built-in core workers

Adapters

Learn more about the adapter pattern