This document is for Kombu's development version, which can be significantly different from previous releases. Get the stable docs here: 5.0.

Introduction

What is messaging?

In times long ago people didn’t have email. They had the postal service, which with great courage would deliver mail from hand to hand all over the globe. Soldiers deployed at wars far away could only communicate with their families through the postal service, and posting a letter would mean that the recipient wouldn’t actually receive the letter until weeks or months, sometimes years later.

It’s hard to imagine this today when people are expected to be available for phone calls every minute of the day.

So humans need to communicate with each other, this shouldn’t be news to anyone, but why would applications?

One example is banks. When you transfer money from one bank to another, your bank sends a message to a central clearinghouse. The clearinghouse then records and coordinates the transaction. Banks need to send and receive millions and millions of messages every day, and losing a single message would mean either losing your money (bad) or the banks money (very bad)

Another example is the stock exchanges, which also have a need for very high message throughputs and have strict reliability requirements.

Email is a great way for people to communicate. It is much faster than using the postal service, but still using email as a means for programs to communicate would be like the soldier above, waiting for signs of life from his girlfriend back home.

Messaging Scenarios

  • Request/Reply

    The request/reply pattern works like the postal service example. A message is addressed to a single recipient, with a return address printed on the back. The recipient may or may not reply to the message by sending it back to the original sender.

    Request-Reply is achieved using direct exchanges.

  • Broadcast

    In a broadcast scenario a message is sent to all parties. This could be none, one or many recipients.

    Broadcast is achieved using fanout exchanges.

  • Publish/Subscribe

    In a publish/subscribe scenario producers publish messages to topics, and consumers subscribe to the topics they are interested in.

    If no consumers subscribe to the topic, then the message will not be delivered to anyone. If several consumers subscribe to the topic, then the message will be delivered to all of them.

    Pub-sub is achieved using topic exchanges.

Reliability

For some applications reliability is very important. Losing a message is a critical situation that must never happen. For other applications losing a message is fine, it can maybe recover in other ways, or the message is resent anyway as periodic updates.

AMQP defines two built-in delivery modes:

  • persistent

    Messages are written to disk and survives a broker restart.

  • transient

    Messages may or may not be written to disk, as the broker sees fit to optimize memory contents. The messages won’t survive a broker restart.

Transient messaging is by far the fastest way to send and receive messages, so having persistent messages comes with a price, but for some applications this is a necessary cost.