Case Study: Asynchronous File Processing with Manager-Worker Actors

Analyze the file_processor example to learn how QB actors can manage and distribute blocking file I/O tasks asynchronously.

Case Study: Asynchronous File Processing with Manager-Worker Actors

• Location: example/core_io/file_processor/
• Objective: This example demonstrates a robust Manager-Worker pattern for handling potentially blocking file I/O operations (reads and writes) in an asynchronous manner within the QB Actor Framework. It showcases how to prevent VirtualCores from stalling by offloading blocking tasks to dedicated worker actors.

By studying this example, you will understand:

• How to structure a Manager-Worker pattern using actors.
• The technique of wrapping blocking system calls (like file operations) within qb::io::async::callback to maintain system responsiveness.
• Managing a pool of worker actors and distributing tasks based on availability.
• A request/response flow where workers can reply directly to the original client actor.

Architectural Overview: Distributing File I/O

The file_processor example uses three main actor types to achieve its goal:

1. FileManager Actor (file_manager.h)

• Core Assignment (Typical): Core 0 (acts as a central dispatcher).
• Role: The central coordinator and dispatcher. It receives file operation requests from ClientActor(s), maintains a queue of pending tasks, and assigns these tasks to available FileWorker actors.
• QB Integration: Standard qb::Actor.
• State Management:
  • _available_workers: An std::unordered_set<qb::ActorId> storing the IDs of FileWorker actors that are currently idle and ready for tasks.
  • _read_requests, _write_requests: std::queues holding pending ReadFileRequest and WriteFileRequest events respectively, when no workers are immediately available.
• Key Event Handling & Logic:
  • on(ReadFileRequest&) and on(WriteFileRequest&):
    • Receives a request. Assigns a unique request_id to track it (though in this example, the original request event is forwarded, already containing necessary IDs).
    • Checks _available_workers. If a worker is free:
      • It removes the worker's ID from _available_workers.
      • It pushes the original request event (which includes the requestor ID) directly to the chosen FileWorker. cpp // Simplified from FileManager::on(ReadFileRequest& request) if (!_available_workers.empty()) { qb::ActorId worker_id = *_available_workers.begin(); _available_workers.erase(_available_workers.begin()); qb::io::cout() << "FileManager: Assigning Read Request for \"" << request.filepath.c_str() << "\" to Worker " << worker_id << ".\n"; push<ReadFileRequest>(worker_id, request.filepath.c_str(), request.requestor, request.request_id); } else { _read_requests.push(request); // Forwarding the whole event implies copying or careful event design qb::io::cout() << "FileManager: No worker available for Read Request. Queued.\n"; }
    • If no worker is available, the request is enqueued in _read_requests or _write_requests (with reads typically prioritized).
  • on(WorkerAvailable&):
    • Called by a FileWorker when it finishes a task.
    • Adds the worker_id from the event back to _available_workers.
    • Immediately checks the task queues (reads first, then writes). If a task is pending, it dequeues it and assigns it to the now-available worker (similar logic to handling a new request).
  • on(ReadFileResponse&) and on(WriteFileResponse&):
    • Receives the response event from a FileWorker.
    • Crucially, it forwards the entire response event (which contains the result, original request_id, and the original requestor ID) to the requestor actor that initiated the operation. cpp // Inside FileManager::on(ReadFileResponse& response) qb::io::cout() << "FileManager: Forwarding ReadFileResponse for \"" << response.filepath.c_str() << "\" to original requestor " << response.requestor << ".\n"; push<ReadFileResponse>(response.requestor, response.filepath.c_str(), response.data, response.success, response.error_msg.c_str(), response.request_id);

2. FileWorker Actor (file_worker.h)

• Core Assignment (Typical): Cores 1+ (can have multiple instances distributed across cores for parallelism).
• Role: Executes a single file read or write operation. Designed to handle potentially blocking I/O without stalling its VirtualCore.
• QB Integration: Standard qb::Actor.
• Initialization (onInit()):
  • Stores its manager's ID (_manager_id).
  • Sends an initial WorkerAvailable message to the FileManager to signal its readiness.

• **Key Event Handling & Logic (on(ReadFileRequest&) and on(WriteFileRequest&)):

  1. Receives a task request from FileManager.
  2. Sets an internal _is_busy = true; flag (though not explicitly used to prevent new tasks in this example, as FileManager manages availability).
  3. **Asynchronous I/O Offload:

  This is the core of its non-blocking nature. It captures the necessary request details (filepath, data for write, original requestor ID, request ID) into a lambda and schedules this lambda for execution using qb::io::async::callback([this, captured_request, ...]() { ... });. ```cpp // Simplified from FileWorker::on(ReadFileRequest& request) auto captured_request = request; // Capture needed data auto file_content = std::make_shared<std::vector<char>>(); // Buffer for read

  qb::io::async::callback([this, captured_request, file_content]() { if (!this->is_alive()) return; // Actor might have been killed

  qb::io::sys::file file_op; bool success = false; qb::string<256> error_message; // — PERFORM BLOCKING FILE I/O — if (file_op.open(captured_request.filepath.c_str(), O_RDONLY) >= 0) { // ... read file into file_content using file_op.read() ... // ... handle potential read errors ... success = true; file_op.close(); } else { error_message = "Failed to open file for reading"; } // — END OF BLOCKING I/O —

  // Send response DIRECTLY to the original client actor this->push<ReadFileResponse>(captured_request.requestor, captured_request.filepath.c_str(), file_content, success, error_message.c_str(), captured_request.request_id); notifyAvailable(); // Tell FileManager this worker is free }); ``

  1. **Inside the

  async::callback Lambda:**

  • Performs the actual, potentially blocking, file operation using

  qb::io::sys::file (open, read/write, close).

  • Constructs a response event (

  ReadFileResponse or WriteFileResponse) containing the result (data read, bytes written, success status, error message), the original request_id, and importantly, the requestor ID from the captured request.

  • **Direct Reply to Client:**

  pushes the response event *directly* to the original requestor (captured_request.requestor). This bypasses the FileManager for the response path, reducing load on the manager.

  • Calls

  notifyAvailable(), which pushes a WorkerAvailable event (with its own id()) back to the FileManager`.

3. ClientActor (main.cpp)

• Core Assignment (Typical): Core 0 (can co-exist with FileManager).
• Role: Simulates one or more clients that make requests for file operations.
• QB Integration: Standard qb::Actor.
• Interaction Flow:
  1. In onInit() or other handlers, sends ReadFileRequest or WriteFileRequest events to the FileManager. It includes its own id() as the requestor field in these requests.
  2. Receives ReadFileResponse or WriteFileResponse events directly from the FileWorker that processed its specific request.
  3. Processes the response (e.g., prints content to console, checks status).
  4. Tracks pending requests and can initiate a system shutdown when all its simulated operations are complete.

Key QB Concepts Illustrated

• Manager-Worker Pattern: A common and effective pattern for distributing tasks and managing a pool of worker components.
• Asynchronicity for Blocking Operations: Demonstrates the crucial technique of wrapping potentially blocking system calls (like synchronous file I/O) inside qb::io::async::callback. This prevents the FileWorker actor from stalling its VirtualCore, maintaining overall system responsiveness.
• Task Queuing & Basic Load Balancing: The FileManager acts as a simple load balancer by assigning tasks to the next free worker or by queueing requests if all workers are busy.
• Decoupled Request/Response Flow (Direct Reply): The pattern where a client sends a request to a manager, the manager delegates to a worker, and the worker sends the response directly back to the original client is a powerful way to reduce load on central manager actors and improve response latency. The original requestor ID is passed along through the events to enable this.
• State Management for Workers: FileWorkers manage their busy/available state implicitly by only notifying the FileManager when they complete a task.
• Multi-Core Parallelism: Multiple FileWorker actors can be instantiated on different VirtualCores, allowing for concurrent file I/O operations (though actual parallelism will be limited by the underlying disk subsystem's performance).
• Custom Event Design: Shows practical examples of events for requests (ReadFileRequest), responses (ReadFileResponse), and internal coordination (WorkerAvailable).

This example provides a clear blueprint for how to integrate blocking tasks into a QB actor system without sacrificing the benefits of asynchronous processing.

(Next Example Analysis: message_broker Example Analysis**)