[go: up one dir, main page]

Skip to content
/ iris Public

Lightweight Component Model and Messaging Framework based on ØMQ

License

Notifications You must be signed in to change notification settings

p-ranav/iris

Repository files navigation

iris is a C++17 header-only library that provides a component model and messaging framework based on ZeroMQ.

Prerequisites

  • Requires a compiler with C++17 standard compliance
  • Requires an installation of libzmq

Table of Contents

Component Model

An iris::Component is a building block - A reusable piece of software that can be instantiated and connected with other components. Think LEGO. Large and complex software systems can be assembled by composing small, tested component building blocks.

iris components support:

  • A variety of communication ports and patterns: Publisher, Subscriber, Client, Server, AsyncServer, and Brokers
  • Periodic and oneshot timers that can trigger the component into action
  • A speedy multi-threaded task system with task stealing
  • Cereal-based serialization and deserialization of complex structures
  • ZeroMQ-based messaging

Getting Started

Simply include #include <iris/iris.hpp> and you're good to go. Start by creating an iris::Component:

iris::Component my_component;

You can optionally specify the number of threads the component can use in its task system, e.g., this component will spawn 2 executor threads that process records in its respective message queues.

iris::Component my_component(iris::threads = 2);

NOTE: Here iris::threads is a NamedType parameter. It is not necessary to use named parameters but in certain cases they improve code readability.

Time-Triggered Operations

iris components can be triggered periodically by timers. To create a timer, call component.set_interval. The following component is triggered every 500ms. Timers are an excellent way to kickstart a communication pattern, e.g., publish messages periodically to multiple sinks.

Call Component.start() to start the component - This starts the component executor threads, listener threads, timers etc.

#include <iostream>
#include <iris/iris.hpp>
using namespace iris;

int main() {
  Component my_component;
  my_component.set_interval(period = 500,
                            on_triggered = [] { std::cout << "Timer fired!\n"; });
  my_component.start();
}

One-shot Timers

Use component.set_timeout to create a one-shot timer that triggers the component after a set delay.

// oneshot_timers.cpp
#include <iris/iris.hpp>
using namespace iris;
#include <iostream>

int main() {
  Component c;
  c.set_timeout(delay = 1000,
                on_triggered = [] { std::cout << "1.0 second Timeout!" << std::endl; 
                });
  c.set_timeout(delay = 2500,
                on_triggered = [] { std::cout << "2.5 second Timeout!" << std::endl; 
                });
  c.set_timeout(delay = 5000, 
                on_triggered = [&] {
                    std::cout << "Stopping component" << std::endl;
                    c.stop();
                });
  c.start();
}

Noice that the component is stopped after 5 seconds - component.stop() stops the task scheduler from further processing of tasks.

Publish-Subscribe Interactions

Publish/Subscribe is classic pattern where senders of messages, called publishers, do not program the messages to be sent directly to specific receivers, called subscribers. Messages are published without the knowledge of what or if any subscriber of that knowledge exists.

In this example (samples/nginx_log_publisher), we will be parsing an Nginx log file and publishing each log entry. Here's the log file format:

[{"time": "17/May/2015:08:05:32 +0000", "remote_ip": "93.180.71.3", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.21)"},
{"time": "17/May/2015:08:05:23 +0000", "remote_ip": "93.180.71.3", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.21)"},
{"time": "17/May/2015:08:05:24 +0000", "remote_ip": "80.91.33.133", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.17)"},
{"time": "17/May/2015:08:05:34 +0000", "remote_ip": "217.168.17.5", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 200, "bytes": 490, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.10.3)"},
{"time": "17/May/2015:08:05:09 +0000", "remote_ip": "217.168.17.5", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 200, "bytes": 490, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.10.3)"},
{"time": "17/May/2015:08:05:57 +0000", "remote_ip": "93.180.71.3", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.21)"},
{"time": "17/May/2015:08:05:02 +0000", "remote_ip": "217.168.17.5", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 404, "bytes": 337, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.10.3)"},
{"time": "17/May/2015:08:05:42 +0000", "remote_ip": "217.168.17.5", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 404, "bytes": 332, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.10.3)"},
{"time": "17/May/2015:08:05:01 +0000", "remote_ip": "80.91.33.133", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.17)"},
{"time": "17/May/2015:08:05:27 +0000", "remote_ip": "93.180.71.3", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.21)"},
{"time": "17/May/2015:08:05:12 +0000", "remote_ip": "217.168.17.5", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 200, "bytes": 3316, "referrer": "-", "agent": "-"},
{"time": "17/May/2015:08:05:49 +0000", "remote_ip": "188.138.60.101", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.9.7.9)"},
{"time": "17/May/2015:08:05:14 +0000", "remote_ip": "80.91.33.133", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.16)"},
{"time": "17/May/2015:08:05:45 +0000", "remote_ip": "46.4.66.76", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 404, "bytes": 318, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (1.0.1ubuntu2)"},
{"time": "17/May/2015:08:05:26 +0000", "remote_ip": "93.180.71.3", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 404, "bytes": 324, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.21)"},
{"time": "17/May/2015:08:05:22 +0000", "remote_ip": "91.234.194.89", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.9.7.9)"},
{"time": "17/May/2015:08:05:07 +0000", "remote_ip": "80.91.33.133", "remote_user": "-", "request": "GET /downloads/product_1 HTTP/1.1", "response": 304, "bytes": 0, "referrer": "-", "agent": "Debian APT-HTTP/1.3 (0.8.16~exp12ubuntu10.17)"},
{"time": "17/May/2015:08:05:38 +0000", "remote_ip": "37.26.93.214", "remote_user": "-", "request": "GET /downloads/product_2 HTTP/1.1", "response": 404, "bytes": 319, "referrer": "-", "agent": "Go 1.1 package http"},
..
...
....

First we can write a message struct NginxLogEntry. iris uses Cereal for serialization and deserialization of messages for transport. Our NginxLogEntry struct has a serialize method for this purpose.

// nginx_log_entry.hpp
#pragma once
#include <string>

struct NginxLogEntry {
  std::string time;
  std::string remote_ip;
  std::string remote_user;
  std::string request;
  unsigned response;
  unsigned bytes;
  std::string agent;

  template <class Archive> void serialize(Archive &ar) {
    ar(time, remote_ip, remote_user, request, response, bytes, agent);
  }
};

We can start by writing our subscriber.

  • Create a subscriber using component.create_subscriber
  • The subscriber port timeout is how long the subscriber's recv() call will wait before timing out and checking again. Timeouts are essential to keeping the component reactive to commands like component.stop(). See ZMQ_RCVTIMEO for more details.
  • The signature of a subscriber callback is std::function<void(Message)>
  • You can deserialize the received message using Message.get<T>()
  • Here, we are receiving log entries and printing select fields
// subscriber.cpp
#include <iostream>
#include <iris/iris.hpp>
using namespace iris;
#include "nginx_log_entry.hpp"

int main() {
  Component receiver(threads = 2);
  receiver.create_subscriber(
      endpoints = {"tcp://localhost:5555"}, 
      timeout = 5000,
      on_receive = [](Message msg) {
          auto entry = msg.get<NginxLogEntry>();
          std::cout << "[" << entry.time << "] "
                    << "{" << entry.remote_ip << "} "
                    << "-> " << entry.request
                    << "-> " << entry.response << std::endl;
      });
  receiver.start();
}

Now, for the publisher. When managing state, it is cleaner to inherit from iris::Component and write a class.

  • Create a class named NginxLogPublisher that inherits from iris::Component
  • Create a publisher by calling create_publisher - We have inherited this method
  • Parse the JSON log file
  • Create a periodic timer using set_interval and publish log messages
  • Call join() on the class destructor to join on the task system executor threads
// publisher.cpp
#include <iostream>
#include <iris/iris.hpp>
using namespace iris;
#include "nginx_log_entry.hpp"
#include "json.hpp"
#include <fstream>

class NginxLogPublisher : public Component {
  Publisher pub;
  nlohmann::json j;
  nlohmann::json::iterator it;

public:
  NginxLogPublisher(const std::string &filename) {
    // read a JSON file
    std::ifstream stream("nginx_logs.json");
    stream >> j;
    it = j.begin();

    // Craete publisher
    pub = create_publisher(endpoints = {"tcp://*:5555"});

    // Publish periodically
    set_interval(period = 200, 
                 on_triggered = [this] {
                     auto element = *it;
                     std::cout << "Published: " << element << std::endl;
                     pub.send(NginxLogEntry{
                         .time = element["time"].get<std::string>(),
                         .remote_ip = element["remote_ip"].get<std::string>(),
                         .remote_user = element["remote_user"].get<std::string>(),
                         .request = element["request"].get<std::string>(),
                         .response = element["response"].get<unsigned>(),
                         .bytes = element["bytes"].get<unsigned>(),
                         .agent = element["agent"].get<std::string>()
                     });
                     ++it;
                 });
  }

  ~NginxLogPublisher() {
    join();
  }

};

int main() {
  NginxLogPublisher publisher("nginx_logs.json");
  publisher.start();
}

Synchronous Request-Reply Interactions

The client-server model is another basic interaction pattern. Client sends a request to a remote server and waits for a reply. The server receives the request and calls a server-side callback to respond to the client.

In this example (samples/music_tag_server), we will create a music database server that can be queried for album metadata. Clients can request for album metadata using a key-value pair, e.g., {"name", "Dark Side of the Moon"}. The server will respond with an std::optional<Album> - If the album is found in its database, it will return it, else it will return std::nullopt.

Our JSON database looks like this:

[
    {
        "catalog": "R2 552927",
        "name": "Paranoid", 
        "artist": "Black Sabbath", 
        "year": 1970, 
        "genre": "Heavy Metal", 
        "tracks": [
            "War Pigs", 
            "Paranoid", 
            "Planet Caravan", 
            "Iron Man", 
            "Electric Funeral", 
            "Hand of Doom", 
            "Jack the Stripper / Fairies Wear Boots"]
    },
    {
        "catalog": "7243 8 35870 2 5",
        "name": "The Number of the Beast",
...
...
...

Let's start with the expected server response - the Album struct.

// album.hpp
#pragma once
#include <iris/cereal/types/vector.hpp>
#include <string>
#include <vector>

struct Album {
  std::string name;
  std::string artist;
  int year;
  std::string genre;
  std::vector<std::string> tracks;

  template <class Archive> void serialize(Archive &ar) {
    ar(name, artist, year, genre, tracks);
  }
};

To create a server port, call component.create_server.

  • Server callbacks have the signature std::function<void(Request, Response&)>
  • The server port timeout is how long the server's recv() call will wait before timing out and checking again. Timeouts are essential to keeping the component reactive to commands like component.stop(). See ZMQ_RCVTIMEO for more details.
  • OK so what's happening here?
    • The server receives a request that is actually an std::tuple<std::string, std::string>
    • We deserialize this request using request.get<T>
    • Then we search in our JSON database if this key-value pair exists.
    • If yes, we construct an Album object and set the server callback response
    • If not, we set the server callback response to std::nullopt
// server.cpp
#include "album.hpp"
#include <iris/iris.hpp>
using namespace iris;
#include <iostream>
#include <map>
#include "json.hpp"
#include <fstream>
#include <iris/cereal/types/optional.hpp>
#include <iris/cereal/types/tuple.hpp>

int main() {

  // Load JSON database
  nlohmann::json j;
  std::ifstream stream("database.json");
  stream >> j;

  Component server;
  server.create_server(
      endpoints = {"tcp://*:5510"}, 
      timeout = 500,
      on_request = [&](Request request, Response &response) {
          // Request from client
          auto kvpair = request.get<std::tuple<std::string, std::string>>();
          auto key = std::get<0>(kvpair);
          auto value = std::get<1>(kvpair);
          std::cout << "Received request {key: " 
                    << key << ", value: " << value << "}\n";

          // Response to be filled and sent back 
          // Either a valid album struct or empty
          std::optional<Album> album{};

          // Find the album in the JSON database
          auto it = std::find_if(j.begin(), j.end(), 
            [&key, &value](const auto& element) {
              if (key == "year")
                return element[key] == std::stoi(value);
              else
                return element[key] == value;
          });

          // Populate the response fields
          if (it != j.end()) {
            album = Album {
              .name = (*it)["name"].get<std::string>(),
              .artist = (*it)["artist"].get<std::string>(),
              .year = (*it)["year"].get<unsigned>(),
              .genre = (*it)["genre"].get<std::string>(),
              .tracks = (*it)["tracks"].get<std::vector<std::string>>()
            };
          }

          // Set response
          response.set(album);
      });
  server.start();
}

Now, we can write a client that calls this server. Create a client port using component.create_client.

NOTE iris clients implement the lazy pirate pattern - Rather than doing a blocking receive, iris clients:

  • Send a request to the server
  • Poll the REQ socket and receive from it only when it's sure a reply has arrived.
  • Resend a request, if no reply has arrived within a timeout period.
  • Abandon the transaction if there is still no reply after several requests.

So, iris::Clients require a timeout (waiting on server response) and a total number of retries when this timeout occurs.

// client.cpp
#include "album.hpp"
#include <iostream>
#include <optional>
#include <iris/iris.hpp>
using namespace iris;
#include <iris/cereal/types/optional.hpp>
#include <iris/cereal/types/tuple.hpp>

int main(int argc, char *argv[]) {

  std::tuple<std::string, std::string> request;

  if (argc != 3) {
    std::cout << "Usage: ./<executable> key value\n";
    return 0;
  } else {
    request = {argv[1], argv[2]};
  }

  Component c(threads = 1);
  c.start();
  auto client = c.create_client(endpoints = {"tcp://127.0.0.1:5510"},
                                timeout = 2500, 
                                retries = 3);

  // Send request to server
  auto response = client.send(request);

  // Check that the server has called `Response.set()`
  if (response.has_value()) {

    // Parse response and print result if available
    auto album = response.get<std::optional<Album>>();
    if (album.has_value()) {
      auto metadata = album.value();
      std::cout << "- Received album:\n";
      std::cout << "    Name: " << metadata.name << "\n";
      std::cout << "    Artist: " << metadata.artist << "\n";
      std::cout << "    Year: " << metadata.year << "\n";
      std::cout << "    Genre: " << metadata.genre << "\n";
      std::cout << "    Tracks:\n";
      for (size_t i = 0; i < metadata.tracks.size(); ++i) {
        std::cout << "      " << i << ". " << metadata.tracks[i] << "\n";
      }
    } else {
      std::cout << "Album not found!\n";
    }
    
  } else {
    std::cout << "Response not set by server\n";
  }

  c.stop();
}

Asynchronous Request-Reply Interactions

Rather than having one client request work from one worker can we get any number of clients to request work from any number of workers? We could pre load each client with a list of workers and have each client talk directly to a worker. This works, but what if we add or remove workers, we then need to update every client. A better solution would be to have a broker which both clients and workers connect to and is responsible for passing messages back and forth. Brokers can deal with many simultaneous requests and responses using routers and dealers. Routers are like asynchronous response sockets and dealers like asynchronous request sockets.

For this example (samples/number_stats_server), let's assume we have some servers (workers) that can calculate the mean and standard deviation of an array of numbers. The client will send an array of numbers to the broker. The broker will forward this request to one of many async servers. These async servers/workers are connected to the broker and waiting for work.

Here is the struct we will be passing to the server(s). On the client-side, we will prepare a random array of 3 doubles and pass to the server.

// numbers.hpp
#pragma once
#include <iris/cereal/types/vector.hpp>
#include <vector>

struct Numbers {
  std::vector<double> values;

  auto size() const { return values.size(); }
  auto begin() const { return values.begin(); }
  auto end() const { return values.end(); }

  template <class Archive> void serialize(Archive &ar) {
    ar(values);
  }
};

and here is the response we expect from the server. A Statistics object with mean and standard deviation:

// statistics.hpp
#pragma once

struct Statistics {
  double mean;
  double stdev;

  template <class Archive> void serialize(Archive &ar) {
    ar(mean, stdev);
  }
};

Our broker component is very simple. It forwards request on port 5510 to port 5515 where one of our workers will be waiting to receive work.

NOTE Below, we are creating a component with 0 threads - The task system is not needed for pure broker components as there are no tasks to execute. Brokers operate at the ZeroMQ level, simply forwarding requests and responses.

NOTE Broker components, like any other component, can have other communication ports and timers. If you have these, then you need executor threads.

#include <iostream>
#include <iris/iris.hpp>
using namespace iris;

int main() {
  Component b(threads = 0);
  b.create_broker(
    router_endpoints = {"tcp://*:5510"},
    dealer_endpoints = {"tcp://*:5515"}
  );
  b.start();
}

Here is our async server. Create workers like this using Component.create_async_server.

This server:

  • receives a Request object that is deserialized into a Numbers structures.
  • calculates the mean and standard deviation of the array.
  • sets the response using response.set

NOTE iris::AsyncServer is not very different from iris::Server. Instead of binding to a ZeroMQ socket and waiting to receive requests, an AsyncServer connects with a broker and waits for requests.

#include <iostream>
#include "numbers.hpp"
#include "statistics.hpp"
#include <iris/iris.hpp>
using namespace iris;
#include <algorithm>
#include <numeric>
#include <cmath>

int main() {
  Component worker(threads = 3);

  worker.create_async_server(
      endpoints = {"tcp://localhost:5515"}, 
      timeout = 500,
      on_request = [&](Request request, Response &res) {
          auto numbers = request.get<Numbers>();

          std::cout << "Received numbers: {" << numbers.values[0] 
                    << ", " << numbers.values[1] 
                    << ", " << numbers.values[2] << "}\n";

          // Calculate mean
          double sum = std::accumulate(numbers.begin(), numbers.end(), 0.0);
          double mean = sum / numbers.size();

          // Calculate standard deviation
          double accum = 0.0;
          std::for_each(numbers.begin(), numbers.end(), [&](const double d) {
            accum += (d - mean) * (d -mean);
          });
          double stdev = std::sqrt(accum / numbers.size());

          // Set the response
          res.set(Statistics{.mean = mean, .stdev = stdev});

          std::cout << "Calculated stats successfully\n";
      });
  worker.start();
}

Finally, here's our client. This client:

  • Creates an array of numbers and sends it to the broker
  • The broker will forward to one of the async servers at its disposal.
  • One of the servers will respond and the response is forwarded back to this client
#include <iostream>
#include "numbers.hpp"
#include "statistics.hpp"
#include <iris/iris.hpp>
using namespace iris;

int main() {
  Component c(threads = 2);
  auto client = c.create_client(endpoints = {"tcp://localhost:5510"},
                                timeout = 2500, 
                                retries = 3);

  double i = 0.0, j = 1.0, k = 2.0;

  c.set_interval(
      period = 2000, 
      on_triggered = [&] {
          std::cout << "[Sent] numbers = {" << i << ", " << j << ", " << k << "}\n";
          auto response = client.send(Numbers{.values = {i, j, k}});
          auto stats = response.get<Statistics>();
          std::cout << "[Received] mean = " << stats.mean 
                    << "; stdev = " << stats.stdev
                    << std::endl;
          i += 0.3;
          j += 0.5;
          k += 0.9;
      });
  c.start();
}

Building Samples

There are a number of samples in the samples/ directory. You can build these samples by running the following commands.

mkdir build && cd build
cmake -DIRIS_SAMPLES=ON ..
make

Contributing

Contributions are welcome, have a look at the CONTRIBUTING.md document for more information.

License

The project is available under the MIT license.