RabbitMQ client that automatically reconnects when the connection fails, and has a confirmed push method (i.e. the server is guaranteed to recieve the message)

queue.go

package main

import (
	"errors"
	"github.com/streadway/amqp"
	"log"
	"os"
	"time"
)

// Queue represents a connection to a specific queue.
type Queue struct {
	name          string
	logger        *log.Logger
	connection    *amqp.Connection
	channel       *amqp.Channel
	done          chan bool
	notifyClose   chan *amqp.Error
	notifyConfirm chan amqp.Confirmation
	isConnected   bool
}

const (
	// When reconnecting to the server after connection failure
	reconnectDelay = 5 * time.Second

	// When resending messages the server didn't confirm
	resendDelay = 5 * time.Second
)

var (
	errNotConnected  = errors.New("not connected to the queue")
	errNotConfirmed  = errors.New("message not confirmed")
	errAlreadyClosed = errors.New("already closed: not connected to the queue")
)

// New creates a new queue instance, and automatically
// attempts to connect to the server.
func New(name string, addr string) *Queue {
	queue := Queue{
		logger: log.New(os.Stdout, "", log.LstdFlags),
		name:   name,
		done:   make(chan bool),
	}
	go queue.handleReconnect(addr)
	return &queue
}

// handleReconnect will wait for a connection error on
// notifyClose, and then continously attempt to reconnect.
func (queue *Queue) handleReconnect(addr string) {
	for {
		queue.isConnected = false
		log.Println("Attempting to connect")
		for !queue.connect(addr) {
			log.Println("Failed to connect. Retrying...")
			time.Sleep(reconnectDelay)
		}
		select {
		case <-queue.done:
			return
		case <-queue.notifyClose:
		}
	}
}

// connect will make a single attempt to connect to
// RabbitMQ. It returns the success of the attempt.
func (queue *Queue) connect(addr string) bool {
	conn, err := amqp.Dial(addr)
	if err != nil {
		return false
	}
	ch, err := conn.Channel()
	if err != nil {
		return false
	}
	ch.Confirm(false)
	_, err = ch.QueueDeclare(
		queue.name,
		false, // Durable
		false, // Delete when unused
		false, // Exclusive
		false, // No-wait
		nil,   // Arguments
	)
	if err != nil {
		return false
	}
	queue.changeConnection(conn, ch)
	queue.isConnected = true
	log.Println("Connected!")
	return true
}

// changeConnection takes a new connection to the queue,
// and updates the channel listeners to reflect this.
func (queue *Queue) changeConnection(connection *amqp.Connection, channel *amqp.Channel) {
	queue.connection = connection
	queue.channel = channel
	queue.notifyClose = make(chan *amqp.Error)
	queue.notifyConfirm = make(chan amqp.Confirmation)
	queue.channel.NotifyClose(queue.notifyClose)
	queue.channel.NotifyPublish(queue.notifyConfirm)
}

// Push will push data onto the queue, and wait for a confirm.
// If no confirms are recieved until within the resendTimeout,
// it continuously resends messages until a confirm is recieved.
// This will block until the server sends a confirm. Errors are
// only returned if the push action itself fails, see UnsafePush.
func (queue *Queue) Push(data []byte) error {
	if !queue.isConnected {
		return errors.New("failed to push push: not connected")
	}
	for {
		err := queue.UnsafePush(data)
		if err != nil {
			queue.logger.Println("Push failed. Retrying...")
			continue
		}
		select {
		case confirm := <-queue.notifyConfirm:
			if confirm.Ack {
				queue.logger.Println("Push confirmed!")
				return nil
			}
		case <-time.After(resendDelay):
		}
		queue.logger.Println("Push didn't confirm. Retrying...")
	}
}

// UnsafePush will push to the queue without checking for
// confirmation. It returns an error if it fails to connect.
// No guarantees are provided for whether the server will
// recieve the message.
func (queue *Queue) UnsafePush(data []byte) error {
	if !queue.isConnected {
		return errNotConnected
	}
	return queue.channel.Publish(
		"",         // Exchange
		queue.name, // Routing key
		false,      // Mandatory
		false,      // Immediate
		amqp.Publishing{
			ContentType: "text/plain",
			Body:        data,
		},
	)
}

// Stream will continuously put queue items on the channel.
// It is required to call delivery.Ack when it has been
// successfully processed, or delivery.Nack when it fails.
// Ignoring this will cause data to build up on the server.
func (queue *Queue) Stream() (<-chan amqp.Delivery, error) {
	if !queue.isConnected {
		return nil, errNotConnected
	}
	return queue.channel.Consume(
		queue.name,
		"",    // Consumer
		false, // Auto-Ack
		false, // Exclusive
		false, // No-local
		false, // No-Wait
		nil,   // Args
	)
}

// Close will cleanly shutdown the channel and connection.
func (queue *Queue) Close() error {
	if !queue.isConnected {
		return errAlreadyClosed
	}
	err := queue.channel.Close()
	if err != nil {
		return err
	}
	err = queue.connection.Close()
	if err != nil {
		return err
	}
	close(queue.done)
	queue.isConnected = false
	return nil
}

@mshddev Good catch! Though isConnected is always written safely (since only one goroutine writes to it), functions like UnsafePush and Stream and Close may read stale versions of isConnected and fail to return the error response, potentially leading to lost data when it's pushed at the wrong time.

I'll try to update the code. It seems that the same bug is present in the upstream streadway/amqp repo, so I'll create an issue and tag you there.

Thanks a lot for the example.

I'm implementing an abstraction for streadway/amqp and struggle with implementing support for the confirmations.

How is it ensured in your example that when Push() is called in parallel, the notifyConfirm message is associated with the correct publishing?

As a poor timing diagram, I hope it's understandable what I mean: :-)

goroutine1-> Push()... ... queue.channel.Publish() (message has deliveryTag 2)...confirm := <-queue.notifyConfirm() (received ack for deliveryTag 1, msg from goroutine2)
goroutine2-> Push()... queue.channel.Publish() (message has deliveryTag 1)... ... confirm := <-queue.notifyConfirm() (received NACK for deliveryTag 2, msg from goroutine1)

This implementation is extraordinarily buggy. Please don't use it! Leaving it up for posterity, because it's one of my first attempts at "real" multithreaded programming.