NAME

Beekeeper - Framework for building applications with a microservices architecture

VERSION

Version 0.01

SYNOPSIS

Create a service:

package My::Service::Worker;

use base 'Beekeeper::Worker';

sub on_startup {
    my $self = shift;
    
    $self->accept_jobs( 'my.service.echo' => 'echo' );
    
    $self->accept_notifications( 'my.service.msg' => 'msg' );
}

sub echo {
    my ($self, $params) = @_;
    return $params;
}

sub msg {
    my ($self, $params) = @_;
    warn $params->{msg};
}

Create an API for the service:

package My::Service;

use Beekeeper::Client;

sub msg {
    my ($class, $message) = @_;
    my $cli = Beekeeper::Client->instance;
    
    $cli->send_notification(
        method => "my.service.msg",
        params => { msg => $message },
    );
}

sub echo {
    my ($class, %args) = @_;
    my $cli = Beekeeper::Client->instance;
    
    my $result = $cli->do_job(
        method => "my.service.echo",
        params => { %args },
    );

    return $result;
}

Use the service from a client:

package main;
use My::Service;

My::Service->msg( "foo!" );

My::Service->echo( foo => "bar" );

DESCRIPTION

Beekeeper is a framework for building applications with a microservices architecture.

A pool of worker processes handle requests and communicate with each other through a common message bus.

Clients send requests through a different set of message buses, which are isolated for security reasons.

Requests and responses are shoveled between buses by a few router processes.

Benefits of this architecture:

- Scales horizontally very well. It is easy to add or remove workers or brokers.

- High availability. The system remains responsive even when several components fail.

- Easy integration of browsers via WebSockets or clients written in other languages.

Key characteristics:

- Broker is a messaging server like Apache ActiveMQ or RabbitMQ.

- Broker protocol is STOMP (see the specification at https://stomp.github.io/stomp-specification-1.2.html).

- RPC protocol is JSON-RPC 2.0 (see the specification at https://www.jsonrpc.org/specification).

- Message marshalling is JSON.

- No message persistence in the broker, it just passes on messages.

- No routing logic is defined in the broker.

- Efficient multicast and unicast notifications.

- Inherent load balancing.

What does this framework provides:

- Beekeeper::Worker, a base class for writing service workers with almost no additional code.

- Beekeeper::Client, a class for writing service clients.

- bkpr command which spawns and controls worker processes.

- Command line tools for monitoring and controlling remotely worker pools.

- A simple internal broker handy for development or running tests.

- Automatic message routing between frontend and backend buses.

- Centralized logging, which can be shoveled to an external monitoring application.

- Performance metrics gathering, which can be shoveled to an external monitoring application.

Getting Started

Writing workers

Workers provide a service accepting certain RPC calls from clients. The base class Beekeeper::Worker provides all the glue needed to accept requests and communicate trough the message bus with clients or another workers.

A worker class just declares on startup which methods it will accept, then implements them:

package MyApp::Worker;

use base 'Beekeeper::Worker';

sub on_startup {
    my $self = shift;

    $self->accept_jobs(
        'myapp.str.uc' => 'uppercase',
    );
}

sub uppercase {
    my ($self, $params) = @_;

    return uc $params->{'string'};
}

Writing clients

Clients of the service need an interface to use it without knowledge of the underlying RPC mechanisms. The class Beekeeper::Client provides simple methods to connect to the broker and make RPC calls.

This is the interface of the above service:

package MyApp::Client;

use Beekeeper::Client;

sub uppercase {
    my ($class, $str) = @_;

    my $client = Beekeeper::Client->instance;

    my $resp = $client->do_job(
        method => 'myapp.str.uc',
        params => { string => $str },
    );

    return $resp->result;
}

Then other workers or clients can just:

use MyApp::Client;

print MyApp::Client->uppercase("hello!");

Configuring

Beekeeper applications use two config files to define how clients, workers and brokers connect to each other. These files are searched for in ENV BEEKEEPER_CONFIG_DIR, ~/.config/beekeeper and then /etc/beekeeper. File format is relaxed JSON, which allows comments and trailings commas.

The file pool.config.json defines all worker pools running on a host, specifying which logical bus should be used and which services it will run. For example:

[{
    "pool-id" : "myapp",
    "bus-id"  : "backend",
    "workers" : {
        "MyApp::Worker" : { "workers_count" : 4 },
    },
}]

The file bus.config.json defines all logical buses used by the application, specifying the connection parameters to the brokers that will service them. For example:

{
   "bus-id"  : "backend",
   "host"    : "localhost",
   "user"    : "backend",
   "pass"    : "def456",
   "vhost"   : "/back",
]

Neither the worker code nor the client code have hardcoded references to the logical message bus or the broker connection parameters, they communicate to each other using the definitions in these two files.

Running

To start or stop a pool of workers you use the bkpr command. Given the above example config, this will start 4 processes running MyApp::Worker code:

bkpr --pool-id "myapp" start

When started it daemonizes itself and forks all worker processes, then continues monitoring those forked processes and immediately respawns defunct ones.

The framework includes these command line tools to manage worker pools:

- bkpr-top allows to monitor in real time the performance of all workers.

- bkpr-log allows to monitor in real time the log output of all workers.

- bkpr-restart gracefully restarts local or remote worker pools.

Examples

This distribution includes some examples that can be run out of the box using an internal `ToyBroker` (so no install of a proper broker is needed):

examples/basic is a barebones example of the usage of Beekeper.

examples/flood allows to estimate the performance of a Beekeper setup.

examples/webstomp use a service from a browser using WebSockets.

examples/chat implements a real world setup with isolated buses and redundancy.

TODO

Since this project was started (and even then) STOMP has been completely surpassed as a fast and simple messaging protocol by MQTT. And since 2019, when MQTT version 5.0 was released, many brokers started to implement the routing features needed by Beekeeper to run.

So the underlying broker protocol should be changed to MQTT, in order to take advantage of better supported modern brokers.

SEE ALSO

Beekeeper::WorkerPool, Beekeeper::Client, Beekeeper::Worker.

SOURCE REPOSITORY

The source code repository for Beekeeper can be found at https://github.com/jmico/beekeeper

AUTHOR

José Micó, jose.mico@gmail.com

COPYRIGHT AND LICENSE

Copyright 2015 José Micó.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language itself.

This software is distributed in the hope that it will be useful, but it is provided “as is” and without any express or implied warranties. For details, see the full text of the license in the file LICENSE.