ØMQ/2.2 - The Guide

By Pieter Hintjens

With thanks to the hundred or so people who contributed examples in two dozen programming languages, who helped with suggestions and fixes, and who kept pushing for more examples of how to connect your code.

Thanks to Bill Desmarais, Brian Dorsey, Daniel Lin, Eric Desgranges, Gonzalo Diethelm, Guido Goldstein, Hunter Ford, Kamil Shakirov, Martin Sustrik, Mike Castleman, Naveen Chawla, Nicola Peduzzi, Oliver Smith, Olivier Chamoux, Peter Alexander, Pierre Rouleau, Randy Dryburgh, John Unwin, Alex Thomas, Mihail Minkov, Jeremy Avnet, Michael Compton, Kamil Kisiel, Mark Kharitonov, Guillaume Aubert, Ian Barber, Mike Sheridan, Faruk Akgul, Oleg Sidorov, Lev Givon, Allister MacLeod, Alexander D'Archangel, Andreas Hoelzlwimmer, Han Holl, Robert G. Jakabosky, Felipe Cruz, Marcus McCurdy, Mikhail Kulemin, Dr. Gergő Érdi, Pavel Zhukov, Alexander Else, Giovanni Ruggiero, Rick "Technoweenie", Daniel Lundin, Dave Hoover, Simon Jefford, Benjamin Peterson, Justin Case, Devon Weller, Richard Smith, Alexander Morland, Wadim Grasza, Michael Jakl, Uwe Dauernheim, Sebastian Nowicki, Simone Deponti, Aaron Raddon, Dan Colish, Markus Schirp, Benoit Larroque, Jonathan Palardy, Isaiah Peng, Arkadiusz Orzechowski, Umut Aydin, Matthew Horsfall, Jeremy W. Sherman, Eric Pugh, Tyler Sellon, John E. Vincent, Pavel Mitin, Min RK, Igor Wiedler, Olof Åkesson, Patrick Lucas, Heow Goodman, Senthil Palanisami, John Gallagher, Tomas Roos, Stephen McQuay, Erik Allik, Arnaud Cogoluègnes, Rob Gagnon, Dan Williams, Edward Smith, James Tucker, Kristian Kristensen, Vadim Shalts, Martin Trojer, Tom van Leeuwen, Pandya Hiten, Harm Aarts, Marc Harter, Iskren Ivov Chernev, Jay Han, Sonia Hamilton, and Zed Shaw.

Thanks to Stathis Sideris for Ditaa, which I used for the diagrams.

Please use the issue tracker for all comments and errata. This version covers the latest stable release of ØMQ (3.2) and was published on Tue 23 October, 2012. If you are using older versions of ØMQ then some of the examples and explanations won't be accurate.

The Guide is originally in C, but also in PHP, Python, Lua, and Haxe. We've also translated most of the examples into C++, C#, CL, Erlang, F#, Felix, Haskell, Java, Objective-C, Ruby, Ada, Basic, Clojure, Go, Haxe, Node.js, ooc, Perl, and Scala.

How to explain ØMQ? Some of us start by saying all the wonderful things it does. It's sockets on steroids. It's like mailboxes with routing. It's fast! Others try to share their moment of enlightenment, that zap-pow-kaboom satori paradigm-shift moment when it all became obvious. Things just become simpler. Complexity goes away. It opens the mind. Others try to explain by comparison. It's smaller, simpler, but still looks familiar. Personally, I like to remember why we made ØMQ at all, because that's most likely where you, the reader, still are today.

Programming is a science dressed up as art, because most of us don't understand the physics of software, and it's rarely if ever taught. The physics of software is not algorithms, data structures, languages and abstractions. These are just tools we make, use, throw away. The real physics of software is the physics of people.

Specifically, our limitations when it comes to complexity, and our desire to work together to solve large problems in pieces. This is the science of programming: make building blocks that people can understand and use easily, and people will work together to solve the very largest problems.

We live in a connected world, and modern software has to navigate this world. So the building blocks for tomorrow's very largest solutions are connected and massively parallel. It's not enough for code to be "strong and silent" any more. Code has to talk to code. Code has to be chatty, sociable, well-connected. Code has to run like the human brain, trillions of individual neurons firing off messages to each other, a massively parallel network with no central control, no single point of failure, yet able to solve immensely difficult problems. And it's no accident that the future of code looks like the human brain, because the endpoints of every network are, at some level, human brains.

If you've done any work with threads, protocols, or networks, you'll realize this is pretty much impossible. It's a dream. Even connecting a few programs across a few sockets is plain nasty, when you start to handle real life situations. Trillions? The cost would be unimaginable. Connecting computers is so difficult that software and services to do this is a multi-billion dollar business.

So we live in a world where the wiring is years ahead of our ability to use it. We had a software crisis in the 1980s, when leading software engineers like Fred Brooks believed there was no "Silver Bullet" to "promise even one order of magnitude of improvement in productivity, reliability, or simplicity".

Brooks missed free and open source software, which solved that crisis, enabling us to share knowledge efficiently. Today we face another software crisis, but it's one we don't talk about much. Only the largest, richest firms can afford to create connected applications. There is a cloud, but it's proprietary. Our data, our knowledge is disappearing from our personal computers into clouds that we cannot access, cannot compete with. Who owns our social networks? It is like the mainframe-PC revolution in reverse.

We can leave the political philosophy for another book. The point is that while the Internet offers the potential of massively connected code, the reality is that this is out of reach for most of us, and so, large interesting problems (in health, education, economics, transport, and so on) remain unsolved because there is no way to connect the code, and thus no way to connect the brains that could work together to solve these problems.

There have been many attempts to solve the challenge of connected software. There are thousands of IETF specifications, each solving part of the puzzle. For application developers, HTTP is perhaps the one solution to have been simple enough to work, but it arguably makes the problem worse, by encouraging developers and architects to think in terms of big servers and thin, stupid clients.

So today people are still connecting applications using raw UDP and TCP, proprietary protocols, HTTP, Websockets. It remains painful, slow, hard to scale, and essentially centralized. Distributed P2P architectures are mostly for play, not work. How many applications use Skype or Bittorrent to exchange data?

Which brings us back to the science of programming. To fix the world, we needed to do two things. One, to solve the general problem of "how to connect any code to any code, anywhere". Two, to wrap that up in the simplest possible building blocks that people could understand and use easily.

It sounds ridiculously simple. And maybe it is. That's kind of the whole point.

ØMQ (ZeroMQ, 0MQ, zmq) looks like an embeddable networking library but acts like a concurrency framework. It gives you sockets that carry atomic messages across various transports like in-process, inter-process, TCP, and multicast. You can connect sockets N-to-N with patterns like fanout, pub-sub, task distribution, and request-reply. It's fast enough to be the fabric for clustered products. Its asynchronous I/O model gives you scalable multicore applications, built as asynchronous message-processing tasks. It has a score of language APIs and runs on most operating systems. ØMQ is from iMatix and is LGPLv3 open source.

We assume you are using the latest 3.2 release of ØMQ. We assume you are using a Linux box or something similar. We assume you can read C code, more or less, that's the default language for the examples. We assume that when we write constants like PUSH or SUBSCRIBE you can imagine they are really called ZMQ_PUSH or ZMQ_SUBSCRIBE if the programming language needs it.

The Guide examples live in the Guide's git repository. The simplest way to get all the examples is to clone this repository:

git clone --depth=1 git://github.com/imatix/zguide2.git

And then browse the examples subdirectory. You'll find examples by language. If there are examples missing in a language you use, you're encouraged to submit a translation. This is how the Guide became so useful, thanks to the work of many people. All examples are licensed under MIT/X11.

So let's start with some code. We start of course with a Hello World example. We'll make a client and a server. The client sends "Hello" to the server, which replies with "World". Here's the server in C, which opens a ØMQ socket on port 5555, reads requests on it, and replies with "World" to each request:

Figure 1 - Request-Reply

The REQ-REP socket pair is lockstep. The client does zmq_msg_send(3) and then zmq_msg_recv(3), in a loop (or once if that's all it needs). Doing any other sequence (e.g. sending two messages in a row) will result in a return code of -1 from the send or recv call. Similarly the service does zmq_msg_recv(3) and then zmq_msg_send(3) in that order, and as often as it needs to.

ØMQ uses C as its reference language and this is the main language we'll use for examples. If you're reading this on-line, the link below the example takes you to translations into other programming languages. Let's compare the same server in C++:

You can see that the ØMQ API is similar in C and C++. In a language like PHP, we can hide even more and the code becomes even easier to read:

Here's the client code (click the link below the source to look at, or contribute a translation in your favorite programming language):

Now this looks too simple to be realistic, but a ØMQ socket is what you get when you take a normal TCP socket, inject it with a mix of radioactive isotopes stolen from a secret Soviet atomic research project, bombard it with 1950-era cosmic rays, and put it into the hands of a drug-addled comic book author with a badly-disguised fetish for bulging muscles clad in spandex. Yes, ØMQ sockets are the world-saving superheroes of the networking world.

Figure 2 - A terrible accident…

You could literally throw thousands of clients at this server, all at once, and it would continue to work happily and quickly. For fun, try starting the client and then starting the server, see how it all still works, then think for a second what this means.

Let me explain briefly what these two programs are actually doing. They create a ØMQ context to work with, and a socket. Don't worry what the words mean. You'll pick it up. The server binds its REP (reply) socket to port 5555. The server waits for a request, in a loop, and responds each time with a reply. The client sends a request and reads the reply back from the server.

If you kill the server (Ctrl-C) and restart it, the client won't recover properly. Recovering from crashing processes isn't quite that easy. Making a reliable request-reply flow is complex enough that I won't cover it until Chapter Four.

There is a lot happening behind the scenes but what matters to us programmers is how short and sweet the code is, and how often it doesn't crash, even under heavy load. This is the request-reply pattern, probably the simplest way to use ØMQ. It maps to RPC and the classic client-server model.

ØMQ doesn't know anything about the data you send except its size in bytes. That means you are responsible for formatting it safely so that applications can read it back. Doing this for objects and complex data types is a job for specialized libraries like Protocol Buffers. But even for strings you need to take care.

In C and some other languages, strings are terminated with a null byte. We could send a string like "HELLO" with that extra null byte:

However if you send a string from another language it probably will not include that null byte. For example, when we send that same string in Python, we do this:

Then what goes onto the wire is a length (one byte for shorter strings) and the string contents, as individual characters.

Figure 3 - A ØMQ string

And if you read this from a C program, you will get something that looks like a string, and might by accident act like a string (if by luck the five bytes find themselves followed by an innocently lurking null), but isn't a proper string. Which means that your client and server don't agree on the string format, you will get weird results.

When you receive string data from ØMQ, in C, you simply cannot trust that it's safely terminated. Every single time you read a string you should allocate a new buffer with space for an extra byte, copy the string, and terminate it properly with a null.

So let's establish the rule that ØMQ strings are length-specified, and are sent on the wire without a trailing null. In the simplest case (and we'll do this in our examples) a ØMQ string maps neatly to a ØMQ message frame, which looks like the above figure, a length and some bytes.

Here is what we need to do, in C, to receive a ØMQ string and deliver it to the application as a valid C string:

This makes a very handy helper function and in the spirit of making things we can reuse profitably, let's write a similar 's_send' function that sends strings in the correct ØMQ format, and package this into a header file we can reuse.

The result is zhelpers.h, which lets us write sweeter and shorter ØMQ applications in C. It is a fairly long source, and only fun for C developers, so read it at leisure.

ØMQ does come in several versions and quite often, if you hit a problem, it'll be something that's been fixed in a later version. So it's a useful trick to know exactly what version of ØMQ you're actually linking with. Here is a tiny program that does that:

C | C++ | C# | CL | Erlang | F# | Felix | Go | Haskell | Java | Lua | Node.js | Objective-C | Perl | PHP | Python | Q | Ruby | Scala | Tcl | Haxe | Ada | Basic | Clojure | ooc | Racket

The second classic pattern is one-way data distribution, in which a server pushes updates to a set of clients. Let's see an example that pushes out weather updates consisting of a zip code, temperature, and relative humidity. We'll generate random values, just like the real weather stations do.

Here's the server. We'll use port 5556 for this application:

There's no start, and no end to this stream of updates, it's like a never ending broadcast.

Figure 4 - Publish-Subscribe

Here is client application, which listens to the stream of updates and grabs anything to do with a specified zip code, by default New York City because that's a great place to start any adventure:

Note that when you use a SUB socket you must set a subscription using zmq_setsockopt(3) and SUBSCRIBE, as in this code. If you don't set any subscription, you won't get any messages. It's a common mistake for beginners. The subscriber can set many subscriptions, which are added together. That is, if a update matches ANY subscription, the subscriber receives it. The subscriber can also unsubscribe specific subscriptions. Subscriptions are length-specified blobs. See zmq_setsockopt(3) for how this works.

The PUB-SUB socket pair is asynchronous. The client does zmq_msg_recv(3), in a loop (or once if that's all it needs). Trying to send a message to a SUB socket will cause an error. Similarly the service does zmq_msg_send(3) as often as it needs to, but must not do zmq_msg_recv(3) on a PUB socket.

In theory with ØMQ sockets, it does not matter which end connects, and which end binds. However in practice there are undocumented differences that I'll come to later. For now, bind the PUB and connect the SUB, unless your network design makes that impossible.

There is one more important thing to know about PUB-SUB sockets: you do not know precisely when a subscriber starts to get messages. Even if you start a subscriber, wait a while, and then start the publisher, the subscriber will always miss the first messages that the publisher sends. This is because as the subscriber connects to the publisher (something that takes a small but non-zero time), the publisher may already be sending messages out.

This "slow joiner" symptom hits enough people, often enough, that I'm going to explain it in detail. Remember that ØMQ does asynchronous I/O, i.e. in the background. Say you have two nodes doing this, in this order:

• Subscriber connects to an endpoint and receives and counts messages.
• Publisher binds to an endpoint and immediately sends 1,000 messages.

Then the subscriber will most likely not receive anything. You'll blink, check that you set a correct filter, and try again, and the subscriber will still not receive anything.

Making a TCP connection involves to and fro handshaking that takes several milliseconds depending on your network and the number of hops between peers. In that time, ØMQ can send very many messages. For sake of argument assume it takes 5 msecs to establish a connection, and that same link can handle 1M messages per second. During the 5 msecs that the subscriber is connecting to the publisher, it takes the publisher only 1 msec to send out those 1K messages.

In Chapter Two I'll explain how to synchronize a publisher and subscribers so that you don't start to publish data until the subscriber(s) really are connected and ready. There is a simple and stupid way to delay the publisher, which is to sleep. I'd never do this in a real application though, it is extremely fragile as well as inelegant and slow. Use sleeps to prove to yourself what's happening, and then wait for Chapter 2 to see how to do this right.

The alternative to synchronization is to simply assume that the published data stream is infinite and has no start, and no end. This is how we built our weather client example.

So the client subscribes to its chosen zip code and collects a thousand updates for that zip code. That means about ten million updates from the server, if zip codes are randomly distributed. You can start the client, and then the server, and the client will keep working. You can stop and restart the server as often as you like, and the client will keep working. When the client has collected its thousand updates, it calculates the average, prints it, and exits.

Some points about the publish-subscribe pattern:

• A subscriber can connect to more than one publisher, using one 'connect' call each time. Data will then arrive and be interleaved ("fair-queued") so that no single publisher drowns out the others.

• If a publisher has no connected subscribers, then it will simply drop all messages.

• If you're using TCP, and a subscriber is slow, messages will queue up on the publisher. We'll look at how to protect publishers against this, using the "high-water mark" later.

• In the current versions of ØMQ, filtering happens at the subscriber side, not the publisher side. This means, over TCP, that a publisher will send all messages to all subscribers, which will then drop messages they don't want.

This is how long it takes to receive and filter 10M messages on my laptop, which is an 2011-era Intel I7, fast but nothing special:

ph@nb201103:~/work/git/zguide/examples/c$ time wuclient
Collecting updates from weather server...
Average temperature for zipcode '10001 ' was 28F

real    0m4.470s
user    0m0.000s
sys     0m0.008s

As a final example (you are surely getting tired of juicy code and want to delve back into philological discussions about comparative abstractive norms), let's do a little supercomputing. Then coffee. Our supercomputing application is a fairly typical parallel processing model:

• We have a ventilator that produces tasks that can be done in parallel.
• We have a set of workers that process tasks.
• We have a sink that collects results back from the worker processes.

In reality, workers run on superfast boxes, perhaps using GPUs (graphic processing units) to do the hard maths. Here is the ventilator. It generates 100 tasks, each is a message telling the worker to sleep for some number of milliseconds:

Figure 5 - Parallel Pipeline

Here is the worker application. It receives a message, sleeps for that number of seconds, then signals that it's finished:

Here is the sink application. It collects the 100 tasks, then calculates how long the overall processing took, so we can confirm that the workers really were running in parallel, if there are more than one of them:

The average cost of a batch is 5 seconds. When we start 1, 2, 4 workers we get results like this from the sink:

#   1 worker
Total elapsed time: 5034 msec
#   2 workers
Total elapsed time: 2421 msec
#   4 workers
Total elapsed time: 1018 msec

Let's look at some aspects of this code in more detail:

• The workers connect upstream to the ventilator, and downstream to the sink. This means you can add workers arbitrarily. If the workers bound to their endpoints, you would need (a) more endpoints and (b) to modify the ventilator and/or the sink each time you added a worker. We say that the ventilator and sink are 'stable' parts of our architecture and the workers are 'dynamic' parts of it.

• We have to synchronize the start of the batch with all workers being up and running. This is a fairly common gotcha in ØMQ and there is no easy solution. The 'connect' method takes a certain time. So when a set of workers connect to the ventilator, the first one to successfully connect will get a whole load of messages in that short time while the others are also connecting. If you don't synchronize the start of the batch somehow, the system won't run in parallel at all. Try removing the wait, and see.

• The ventilator's PUSH socket distributes tasks to workers (assuming they are all connected before the batch starts going out) evenly. This is called load-balancing and it's something we'll look at again in more detail.

• The sink's PULL socket collects results from workers evenly. This is called fair-queuing.

Figure 6 - Fair Queuing

The pipeline pattern also exhibits the "slow joiner" syndrome, leading to accusations that PUSH sockets don't load balance properly. If you are using PUSH and PULL, and one of your workers gets way more messages than the others, it's because that PULL socket has joined faster than the others, and grabs a lot of messages before the others manage to connect.

Having seen some examples, you're eager to start using ØMQ in some apps. Before you start that, take a deep breath, chillax, and reflect on some basic advice that will save you stress and confusion.

• Learn ØMQ step by step. It's just one simple API but it hides a world of possibilities. Take the possibilities slowly, master each one.

• Write nice code. Ugly code hides problems and makes it hard for others to help you. You might get used to meaningless variable names, but people reading your code won't. Use names that are real words, that say something other than "I'm too careless to tell you what this variable is really for". Use consistent indentation, clean layout. Write nice code and your world will be more comfortable.

• Test what you make as you make it. When your program doesn't work, you should know what five lines are to blame. This is especially true when you do ØMQ magic, which just won't work the first few times you try it.

• When you find that things don't work as expected, break your code into pieces, test each one, see which one is not working. ØMQ lets you make essentially modular code, use that to your advantage.

• Make abstractions (classes, methods, whatever) as you need them. If you copy/paste a lot of code you're going to copy/paste errors too.

To illustrate, here is a fragment of code someone asked me to help fix:

//  NOTE: do NOT reuse this example code!
static char *topic_str = "msg.x|";

void* pub_worker(void* arg){
    void *ctx = arg;
    assert(ctx);

    void *qskt = zmq_socket(ctx, ZMQ_REP);
    assert(qskt);

    int rc = zmq_connect(qskt, "inproc://querys");
    assert(rc == 0);

    void *pubskt = zmq_socket(ctx, ZMQ_PUB);
    assert(pubskt);

    rc = zmq_bind(pubskt, "inproc://publish");
    assert(rc == 0);

    uint8_t cmd;
    uint32_t nb;
    zmq_msg_t topic_msg, cmd_msg, nb_msg, resp_msg;

    zmq_msg_init_data(&topic_msg, topic_str, strlen(topic_str) , NULL, NULL);

    fprintf(stdout,"WORKER: ready to receive messages\n");
    //  NOTE: do NOT reuse this example code, It's broken.
    //  e.g. topic_msg will be invalid the second time through
    while (1){
    zmq_msg_send(pubskt, &topic_msg, ZMQ_SNDMORE);

    zmq_msg_init(&cmd_msg);
    zmq_msg_recv(qskt, &cmd_msg, 0);
    memcpy(&cmd, zmq_msg_data(&cmd_msg), sizeof(uint8_t));
    zmq_msg_send(pubskt, &cmd_msg, ZMQ_SNDMORE);
    zmq_msg_close(&cmd_msg);

    fprintf(stdout, "received cmd %u\n", cmd);

    zmq_msg_init(&nb_msg);
    zmq_msg_recv(qskt, &nb_msg, 0);
    memcpy(&nb, zmq_msg_data(&nb_msg), sizeof(uint32_t));
    zmq_msg_send(pubskt, &nb_msg, 0);
    zmq_msg_close(&nb_msg);

    fprintf(stdout, "received nb %u\n", nb);

    zmq_msg_init_size(&resp_msg, sizeof(uint8_t));
    memset(zmq_msg_data(&resp_msg), 0, sizeof(uint8_t));
    zmq_msg_send(qskt, &resp_msg, 0);
    zmq_msg_close(&resp_msg);

    }
    return NULL;
}

This is what I rewrote it to, as part of finding the bug:

In the end, the problem was that the application was passing sockets between threads, which crashes weirdly. Sockets are not threadsafe. It became legal behavior to migrate sockets from one thread to another in ØMQ/2.1, but this remains dangerous unless you use a "full memory barrier". If you don't know what that means, don't attempt socket migration.

ØMQ applications always start by creating a context, and then using that for creating sockets. In C, it's the zmq_ctx_new(3) call. You should create and use exactly one context in your process. Technically, the context is the container for all sockets in a single process, and acts as the transport for inproc sockets, which are the fastest way to connect threads in one process. If at runtime a process has two contexts, these are like separate ØMQ instances. If that's explicitly what you want, OK, but otherwise remember:

Do one zmq_ctx_new(3) at the start of your main line code, and one zmq_ctx_destroy(3) at the end.

If you're using the fork() system call, each process needs its own context. If you do zmq_ctx_new(3) in the main process before calling fork(), the child processes get their own contexts. In general you want to do the interesting stuff in the child processes, and just manage these from the parent process.

Classy programmers share the same motto as classy hit men: always clean-up when you finish the job. When you use ØMQ in a language like Python, stuff gets automatically freed for you. But when using C you have to carefully free objects when you're finished with them, or you get memory leaks, unstable applications, and generally bad karma.

Memory leaks are one thing, but ØMQ is quite finicky about how you exit an application. The reasons are technical and painful but the upshot is that if you leave any sockets open, the zmq_ctx_destroy(3) function will hang forever. And even if you close all sockets, zmq_ctx_destroy(3) will by default wait forever if there are pending connects or sends. Unless you set the LINGER to zero on those sockets before closing them.

The ØMQ objects we need to worry about are messages, sockets, and contexts. Luckily it's quite simple, at least in simple programs:

• Always close a message the moment you are done with it, using zmq_msg_close(3).

• If you are opening and closing a lot of sockets, that's probably a sign you need to redesign your application.

• When you exit the program, close your sockets and then call zmq_ctx_destroy(3). This destroys the context.

If you're doing multithreaded work, it gets rather more complex than this. We'll get to multithreading in the next chapter, but because some of you will, despite warnings, will try to run before you can safely walk, below is the quick and dirty guide to making a clean exit in a multithreaded ØMQ application.

First, do not try to use the same socket from multiple threads. No, don't explain why you think this would be excellent fun, just please don't do it. Next, you need to shut down each socket that has ongoing requests. The proper way is to set a low LINGER value (1 second), then close the socket. If your language binding doesn't do this for you automatically when you destroy a context, I'd suggest sending a patch.

Finally, destroy the context. This will cause any blocking receives or polls or sends in attached threads (i.e. which share the same context) to return with an error. Catch that error, and then set linger on, and close sockets in that thread, and exit. Do not destroy the same context twice. The zmq_ctx_destroy in the main thread will block until all sockets it knows about are safely closed.

Voila! It's complex and painful enough that any language binding author worth his or her salt will do this automatically and make the socket closing dance unnecessary.

Now that you've seen ØMQ in action, let's go back to the "why".

Many applications these days consist of components that stretch across some kind of network, either a LAN or the Internet. So many application developers end up doing some kind of messaging. Some developers use message queuing products, but most of the time they do it themselves, using TCP or UDP. These protocols are not hard to use, but there is a great difference between sending a few bytes from A to B, and doing messaging in any kind of reliable way.

Let's look at the typical problems we face when we start to connect pieces using raw TCP. Any reusable messaging layer would need to solve all or most these:

• How do we handle I/O? Does our application block, or do we handle I/O in the background? This is a key design decision. Blocking I/O creates architectures that do not scale well. But background I/O can be very hard to do right.

• How do we handle dynamic components, i.e. pieces that go away temporarily? Do we formally split components into "clients" and "servers" and mandate that servers cannot disappear? What then if we want to connect servers to servers? Do we try to reconnect every few seconds?

• How do we represent a message on the wire? How do we frame data so it's easy to write and read, safe from buffer overflows, efficient for small messages, yet adequate for the very largest videos of dancing cats wearing party hats?

• How do we handle messages that we can't deliver immediately? Particularly, if we're waiting for a component to come back on-line? Do we discard messages, put them into a database, or into a memory queue?

• Where do we store message queues? What happens if the component reading from a queue is very slow, and causes our queues to build up? What's our strategy then?

• How do we handle lost messages? Do we wait for fresh data, request a resend, or do we build some kind of reliability layer that ensures messages cannot be lost? What if that layer itself crashes?

• What if we need to use a different network transport. Say, multicast instead of TCP unicast? Or IPv6? Do we need to rewrite the applications, or is the transport abstracted in some layer?

• How do we route messages? Can we send the same message to multiple peers? Can we send replies back to an original requester?

• How do we write an API for another language? Do we re-implement a wire-level protocol or do we repackage a library? If the former, how can we guarantee efficient and stable stacks? If the latter, how can we guarantee interoperability?

• How do we represent data so that it can be read between different architectures? Do we enforce a particular encoding for data types? How far is this the job of the messaging system rather than a higher layer?

• How do we handle network errors? Do we wait and retry, ignore them silently, or abort?

Take a typical open source project like Hadoop Zookeeper and read the C API code in src/c/src/zookeeper.c. As I write this, in 2010, the code is 3,200 lines of mystery and in there is an undocumented, client-server network communication protocol. I see it's efficient because it uses poll() instead of select(). But really, Zookeeper should be using a generic messaging layer and an explicitly documented wire level protocol. It is incredibly wasteful for teams to be building this particular wheel over and over.

Figure 7 - Messaging as it Starts

But how to make a reusable messaging layer? Why, when so many projects need this technology, are people still doing it the hard way, by driving TCP sockets in their code, and solving the problems in that long list, over and over?

It turns out that building reusable messaging systems is really difficult, which is why few FOSS projects ever tried, and why commercial messaging products are complex, expensive, inflexible, and brittle. In 2006 iMatix designed AMQP which started to give FOSS developers perhaps the first reusable recipe for a messaging system. AMQP works better than many other designs but remains relatively complex, expensive, and brittle. It takes weeks to learn to use, and months to create stable architectures that don't crash when things get hairy.

Most messaging projects, like AMQP, that try to solve this long list of problems in a reusable way do so by inventing a new concept, the "broker", that does addressing, routing, and queuing. This results in a client-server protocol or a set of APIs on top of some undocumented protocol, that let applications speak to this broker. Brokers are an excellent thing in reducing the complexity of large networks. But adding broker-based messaging to a product like Zookeeper would make it worse, not better. It would mean adding an additional big box, and a new single point of failure. A broker rapidly becomes a bottleneck and a new risk to manage. If the software supports it, we can add a second, third, fourth broker and make some fail-over scheme. People do this. It creates more moving pieces, more complexity, more things to break.

And a broker-centric set-up needs its own operations team. You literally need to watch the brokers day and night, and beat them with a stick when they start misbehaving. You need boxes, and you need backup boxes, and you need people to manage those boxes. It is only worth doing for large applications with many moving pieces, built by several teams of people, over several years.

So small to medium application developers are trapped. Either they avoid network programming, and make monolithic applications that do not scale. Or they jump into network programming and make brittle, complex applications that are hard to maintain. Or they bet on a messaging product, and end up with scalable applications that depend on expensive, easily broken technology. There has been no really good choice, which is maybe why messaging is largely stuck in the last century and stirs strong emotions. Negative ones for users, gleeful joy for those selling support and licenses.

Figure 8 - Messaging as it Becomes

What we need is something that does the job of messaging but does it in such a simple and cheap way that it can work in any application, with close to zero cost. It should be a library that you just link with, without any other dependencies. No additional moving pieces, so no additional risk. It should run on any OS and work with any programming language.

And this is ØMQ: an efficient, embeddable library that solves most of the problems an application needs to become nicely elastic across a network, without much cost.

Specifically:

• It handles I/O asynchronously, in background threads. These communicate with application threads using lock-free data structures, so concurrent ØMQ applications need no locks, semaphores, or other wait states.

• Components can come and go dynamically and ØMQ will automatically reconnect. This means you can start components in any order. You can create "service-oriented architectures" (SOAs) where services can join and leave the network at any time.

• It queues messages automatically when needed. It does this intelligently, pushing messages as close as possible to the receiver before queuing them.

• It has ways of dealing with over-full queues (called "high water mark"). When a queue is full, ØMQ automatically blocks senders, or throws away messages, depending on the kind of messaging you are doing (the so-called "pattern").

• It lets your applications talk to each other over arbitrary transports: TCP, multicast, in-process, inter-process. You don't need to change your code to use a different transport.

• It handles slow/blocked readers safely, using different strategies that depend on the messaging pattern.

• It lets you route messages using a variety of patterns such as request-reply and publish-subscribe. These patterns are how you create the topology, the structure of your network.

• It lets you create proxies to queue, forward, or capture messages with a single call. Proxies can reduce the interconnection complexity of a network.

• It delivers whole messages exactly as they were sent, using a simple framing on the wire. If you write a 10k message, you will receive a 10k message.

• It does not impose any format on messages. They are blobs of zero to gigabytes large. When you want to represent data you choose some other product on top, such as Google's protocol buffers, XDR, and others.

• It handles network errors intelligently. Sometimes it retries, sometimes it tells you an operation failed.

• It reduces your carbon footprint. Doing more with less CPU means your boxes use less power, and you can keep your old boxes in use for longer. Al Gore would love ØMQ.

Actually ØMQ does rather more than this. It has a subversive effect on how you develop network-capable applications. Superficially it's a socket-inspired API on which you do zmq_msg_recv(3) and zmq_msg_send(3). But message processing rapidly becomes the central loop, and your application soon breaks down into a set of message processing tasks. It is elegant and natural. And it scales: each of these tasks maps to a node, and the nodes talk to each other across arbitrary transports. Two nodes in one process (node is a thread), two nodes on one box (node is a process), or two boxes on one network (node is a box) - it's all the same, with no application code changes.

Let's see ØMQ's scalability in action. Here is a shell script that starts the weather server and then a bunch of clients in parallel:

wuserver &
wuclient 12345 &
wuclient 23456 &
wuclient 34567 &
wuclient 45678 &
wuclient 56789 &

As the clients run, we take a look at the active processes using 'top', and we see something like (on a 4-core box):

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
 7136 ph        20   0 1040m 959m 1156 R  157 12.0  16:25.47 wuserver
 7966 ph        20   0 98608 1804 1372 S   33  0.0   0:03.94 wuclient
 7963 ph        20   0 33116 1748 1372 S   14  0.0   0:00.76 wuclient
 7965 ph        20   0 33116 1784 1372 S    6  0.0   0:00.47 wuclient
 7964 ph        20   0 33116 1788 1372 S    5  0.0   0:00.25 wuclient
 7967 ph        20   0 33072 1740 1372 S    5  0.0   0:00.35 wuclient

Let's think for a second about what is happening here. The weather server has a single socket, and yet here we have it sending data to five clients in parallel. We could have thousands of concurrent clients. The server application doesn't see them, doesn't talk to them directly. So the ØMQ socket is acting like a little server, silently accepting client requests and shoving data out to them as fast as the network can handle it. And it's a multithreaded server, squeezing more juice out of your CPU.

As you start to program with ØMQ you will come across one problem more than once: you lose messages that you expect to receive. Here is a basic problem solver that walks through the most common causes for this. Don't worry if some of the terminology is unfamiliar still, it'll become clearer in the next chapters.

Figure 9 - Missing Message Problem Solver

If you're using ØMQ in a context where failures are expensive, then you want to plan properly. First, build prototypes that let you learn and test the different aspects of your design. Stress them until they break, so that you know exactly how strong your designs are. Second, invest in testing. This means building test frameworks, ensuring you have access to realistic setups with sufficient computer power, and getting time or help to actually test seriously. Ideally, one team writes the code, a second team tries to break it. Lastly, do get your organization to contact iMatix to discuss how we can help to make sure things work properly, and can be fixed rapidly if they break.

In short: if you have not proven an architecture works in realistic conditions, it will most likely break at the worst possible moment.

In early 2012, ØMQ/3.2 became stable enough for live use and by the time you're reading this, it's what you really should be using. If you are still using 2.2, here's a quick summary of the changes, and how to migrate your code.

The main change in 3.x is that PUB-SUB works properly, as in, the publisher only sends subscribers stuff they actually want. In 2.x, publishers send everything and the subscribers filter. Simple, but not ideal for performance on a TCP network.

Most of the API is backwards compatible, except a few blockheaded changes that went into 3.0 with no real regard to the cost of breaking existing code. The syntax of zmq_send(3) and zmq_recv(3) changed, and ZMQ_NOBLOCK got rebaptised to ZMQ_DONTWAIT. So although I'd love to say, "you just recompile your code with the latest libzmq and everything will work", that's not how it is. For what it's worth, we banned such API breakage afterwards.

So the minimal change for C/C++ apps that use the low-level libzmq API is to replace all calls to zmq_send with zmq_msg_send, and zmq_recv with zmq_msg_recv. In other languages, your binding author may have done the work already. Note that these two functions now return -1 in case of error, and zero or more according to how many bytes were sent or received.

Other parts of the libzmq API became more consistent. We deprecated zmq_init(3) and zmq_term(3), replacing them with zmq_ctx_new(3) and zmq_ctx_destroy(3). We added zmq_ctx_set(3) to let you configure a context before starting to work with it.

Finally, we added context monitoring via the zmq_ctx_set_monitor(3) call, which lets you track connections and disconnections, and other events on sockets.

Traditional network programming is built on the general assumption that one socket talks to one connection, one peer. There are multicast protocols but these are exotic. When we assume "one socket = one connection", we scale our architectures in certain ways. We create threads of logic where each thread work with one socket, one peer. We place intelligence and state in these threads.

In the ØMQ universe, sockets are doorways to fast little background communications engines that manage a whole set of connections automagically for you. You can't see, work with, open, close, or attach state to these connections. Whether you use blocking send or receive, or poll, all you can talk to is the socket, not the connections it manages for you. The connections are private and invisible, and this is the key to ØMQ's scalability.

Because your code, talking to a socket, can then handle any number of connections across whatever network protocols are around, without change. A messaging pattern sitting in ØMQ can scale more cheaply than a messaging pattern sitting in your application code.

So the general assumption no longer applies. As you read the code examples, your brain will try to map them to what you know. You will read "socket" and think "ah, that represents a connection to another node". That is wrong. You will read "thread" and your brain will again think, "ah, a thread represents a connection to another node", and again your brain will be wrong.

If you're reading this Guide for the first time, realize that until you actually write ØMQ code for a day or two (and maybe three or four days), you may feel confused, especially by how simple ØMQ makes things for you, and you may try to impose that general assumption on ØMQ, and it won't work. And then you will experience your moment of enlightenment and trust, that zap-pow-kaboom satori paradigm-shift moment when it all becomes clear.

In Chapter One we took ØMQ for a drive, with some basic examples of the main ØMQ patterns: request-reply, publish-subscribe, and pipeline. In this chapter we're going to get our hands dirty and start to learn how to use these tools in real programs.

We'll cover:

• How to create and work with ØMQ sockets.
• How to send and receive messages on sockets.
• How to build your apps around ØMQ's asynchronous I/O model.
• How to handle multiple sockets in one thread.
• How to handle fatal and non-fatal errors properly.
• How to handle interrupt signals like Ctrl-C.
• How to shutdown a ØMQ application cleanly.
• How to check a ØMQ application for memory leaks.
• How to send and receive multi-part messages.
• How to forward messages across networks.
• How to build a simple message queuing broker.
• How to write multithreaded applications with ØMQ.
• How to use ØMQ to signal between threads.
• How to use ØMQ to coordinate a network of nodes.
• How to create and use message envelopes for publish-subscribe.
• Using the high-water mark (HWM) to protect against memory overflows.

The Ø in ØMQ is all about tradeoffs. On the one hand this strange name lowers ØMQ's visibility on Google and Twitter. On the other hand it annoys the heck out of some Danish folk who write us things like "ØMG røtfl", and "Ø is not a funny looking zero!" and "Rødgrød med Fløde!", which is apparently an insult that means "may your neighbours be the direct descendants of Grendel!" Seems like a fair trade.

Originally the zero in ØMQ was meant as "zero broker" and (as close to) "zero latency" (as possible). In the meantime it has come to cover different goals: zero administration, zero cost, zero waste. More generally, "zero" refers to the culture of minimalism that permeates the project. We add power by removing complexity rather than exposing new functionality.

To be perfectly honest, ØMQ does a kind of switch-and-bait on you. Which we don't apologize for, it's for your own good and hurts us more than it hurts you. It presents a familiar socket-based API but that hides a bunch of message-processing engines that will slowly fix your world-view about how to design and write distributed software.

Sockets are the de-facto standard API for network programming, as well as being useful for stopping your eyes from falling onto your cheeks. One thing that makes ØMQ especially tasty to developers is that it uses sockets and messages instead of some other arbitrary set of concepts. Kudos to Martin Sustrik for pulling this off. It turns "Message Oriented Middleware", a phrase guaranteed to send the whole room off to Catatonia, into "Extra Spicy Sockets!" which leaves us with a strange craving for pizza, and a desire to know more.

Like a nice pepperoni pizza, ØMQ sockets are easy to digest. Sockets have a life in four parts, just like BSD sockets:

• Creating and destroying sockets, which go together to form a karmic circle of socket life (see zmq_socket(3), zmq_close(3)).

• Configuring sockets by setting options on them and checking them if necessary (see zmq_setsockopt(3), zmq_getsockopt(3)).

• Plugging sockets onto the network topology by creating ØMQ connections to and from them (see zmq_bind(3), zmq_connect(3)).

• Using the sockets to carry data by writing and receiving messages on them (see zmq_msg_send(3), zmq_msg_recv(3)).

Which looks like this, in C:

Note that sockets are always void pointers, and messages (which we'll come to very soon) are structures. So in C you pass sockets as-such, but you pass addresses of messages in all functions that work with messages, like zmq_msg_send(3) and zmq_msg_recv(3). As a mnemonic, realize that "in ØMQ all your sockets are belong to us", but messages are things you actually own in your code.

Creating, destroying, and configuring sockets works as you'd expect for any object. But remember that ØMQ is an asynchronous, elastic fabric. This has some impact on how we plug sockets into the network topology, and how we use the sockets after that.

To create a connection between two nodes you use zmq_bind(3) in one node, and zmq_connect(3) in the other. As a general rule of thumb, the node which does zmq_bind(3) is a "server", sitting on a well-known network address, and the node which does zmq_connect(3) is a "client", with unknown or arbitrary network addresses. Thus we say that we "bind a socket to an endpoint" and "connect a socket to an endpoint", the endpoint being that well-known network address.

ØMQ connections are somewhat different from old-fashioned TCP connections. The main notable differences are:

• They go across an arbitrary transport (inproc, ipc, tcp, pgm or epgm). See zmq_inproc(7), zmq_ipc(7), zmq_tcp(7), zmq_pgm(7), and zmq_epgm(7).

• They exist when a client does zmq_connect(3) to an endpoint, whether or not a server has already done zmq_bind(3) to that endpoint.

• They are asynchronous, and have queues that magically exist where and when needed.

• They may express a certain "messaging pattern", according to the type of socket used at each end.

• One socket may have many outgoing and many incoming connections.

• There is no zmq_accept() method. When a socket is bound to an endpoint it automatically starts accepting connections.

• Your application code cannot work with these connections directly; they are encapsulated under the socket.

Many architectures follow some kind of client-server model, where the server is the component that is most static, and the clients are the components that are most dynamic, i.e. they come and go the most. There are sometimes issues of addressing: servers will be visible to clients, but not necessarily vice-versa. So mostly it's obvious which node should be doing zmq_bind(3) (the server) and which should be doing zmq_connect(3) (the client). It also depends on the kind of sockets you're using, with some exceptions for unusual network architectures. We'll look at socket types later.

Now, imagine we start the client before we start the server. In traditional networking we get a big red Fail flag. But ØMQ lets us start and stop pieces arbitrarily. As soon as the client node does zmq_connect(3) the connection exists and that node can start to write messages to the socket. At some stage (hopefully before messages queue up so much that they start to get discarded, or the client blocks), the server comes alive, does a zmq_bind(3) and ØMQ starts to deliver messages.

A server node can bind to many endpoints and it can do this using a single socket. This means it will accept connections across different transports:

You cannot bind to the same endpoint twice, that will cause an exception.

Each time a client node does a zmq_connect(3) to any of these endpoints, the server node's socket gets another connection. There is no inherent limit to how many connections a socket can have. A client node can also connect to many endpoints using a single socket.

In most cases, which node acts as client, and which as server, is about network topology rather than message flow. However, there are cases (resending when connections are broken) where the same socket type will behave differently if it's a server or if it's a client.

What this means is that you should always think in terms of "servers" as static parts of your topology, with more-or-less fixed endpoint addresses, and "clients" as dynamic parts that come and go. Then, design your application around this model. The chances that it will "just work" are much better like that.

Sockets have types. The socket type defines the semantics of the socket, its policies for routing messages inwards and outwards, queuing, etc. You can connect certain types of socket together, e.g. a publisher socket and a subscriber socket. Sockets work together in "messaging patterns". We'll look at this in more detail later.

It's the ability to connect sockets in these different ways that gives ØMQ its basic power as a message queuing system. There are layers on top of this, such as proxies, which we'll get to later. But essentially, with ØMQ you define your network architecture by plugging pieces together like a child's construction toy.

To send and receive messages you use the zmq_msg_send(3) and zmq_msg_recv(3) methods. The names are conventional but ØMQ's I/O model is different enough from the TCP model that you will need time to get your head around it.

Figure 10 - TCP sockets are 1 to 1

Let's look at the main differences between TCP sockets and ØMQ sockets when it comes to carrying data:

• ØMQ sockets carry messages, rather than bytes (as in TCP) or frames (as in UDP). A message is a length-specified blob of binary data. We'll come to messages shortly, their design is optimized for performance and thus somewhat tricky to understand.

• ØMQ sockets do their I/O in a background thread. This means that messages arrive in a local input queue, and are sent from a local output queue, no matter what your application is busy doing. These are configurable memory queues, by the way.

• ØMQ sockets can, depending on the socket type, be connected to (or from, it's the same) many other sockets. Where TCP emulates a one-to-one phone call, ØMQ implements one-to-many (like a radio broadcast), many-to-many (like a post office), many-to-one (like a mail box), and even one-to-one.

• ØMQ sockets can send to many endpoints (creating a fan-out model), or receive from many endpoints (creating a fan-in model).

Figure 11 - ØMQ Sockets are N to N

So writing a message to a socket may send the message to one or many other places at once, and conversely, one socket will collect messages from all connections sending messages to it. The zmq_msg_recv(3) method uses a fair-queuing algorithm so each sender gets an even chance.

The zmq_msg_send(3) method does not actually send the message to the socket connection(s). It queues the message so that the I/O thread can send it asynchronously. It does not block except in some exception cases. So the message is not necessarily sent when zmq_msg_send(3) returns to your application. If you created a message using zmq_msg_init_data(3) you cannot reuse the data or free it, otherwise the I/O thread will rapidly find itself writing overwritten or unallocated garbage. This is a common mistake for beginners. We'll see a little later how to properly work with messages.

ØMQ provides a set of unicast transports (inproc, ipc, and tcp) and multicast transports (epgm, pgm). Multicast is an advanced technique that we'll come to later. Don't even start using it unless you know that your fanout ratios will make 1-to-N unicast impossible.

For most common cases, use tcp, which is a disconnected TCP transport. It is elastic, portable, and fast enough for most cases. We call this 'disconnected' because ØMQ's tcp transport doesn't require that the endpoint exists before you connect to it. Clients and servers can connect and bind at any time, can go and come back, and it remains transparent to applications.

The inter-process transport, ipc, is like tcp except that it is abstracted from the LAN, so you don't need to specify IP addresses or domain names. This makes it better for some purposes, and we use it quite often in the examples in this book. ØMQ's ipc transport is disconnected, like tcp. It has one limitation: it does not work on Windows. This may be fixed in future versions of ØMQ. By convention we use endpoint names with an ".ipc" extension to avoid potential conflict with other file names. On UNIX systems, if you use ipc endpoints you need to create these with appropriate permissions otherwise they may not be shareable between processes running under different user ids. You must also make sure all processes can access the files, e.g. by running in the same working directory.

The inter-thread transport, inproc, is a connected signaling transport. It is much faster than tcp or ipc. This transport has a specific limitation compared to ipc and tcp: you must do bind before connect. This is something future versions of ØMQ may fix, but at present this defines you use inproc sockets. We create and bind one socket, start the child threads, which create and connect the other sockets.

A common question that newcomers to ØMQ ask (it's one I asked myself) is something like, "how do I write a XYZ server in ØMQ?" For example, "how do I write an HTTP server in ØMQ?"

The implication is that if we use normal sockets to carry HTTP requests and responses, we should be able to use ØMQ sockets to do the same, only much faster and better.

Sadly the answer is "this is not how it works". ØMQ is not a neutral carrier, it imposes a framing on the transport protocols it uses. This framing is not compatible with existing protocols, which tend to use their own framing. For example, compare an HTTP request, and a ØMQ request, both over TCP/IP.

Figure 12 - HTTP On the Wire

Where the HTTP request uses CR-LF as its simplest framing delimiter, and ØMQ uses a length-specified frame.

Figure 13 - ØMQ On the Wire

So you could write a HTTP-like protocol using ØMQ, using for example the request-reply socket pattern. But it would not be HTTP.

There is however a good answer to the question, "How can I make profitable use of ØMQ when making my new XYZ server?" You need to implement whatever protocol you want to speak in any case, but you can connect that protocol server (which can be extremely thin) to a ØMQ backend that does the real work. The beautiful part here is that you can then extend your backend with code in any language, running locally or remotely, as you wish. Zed Shaw's Mongrel2 web server is a great example of such an architecture.

We said that ØMQ does I/O in a background thread. One I/O thread (for all sockets) is sufficient for all but the most extreme applications. When you create a new context it starts with one I/O thread. The general rule of thumb is to allow one I/O thread per gigabyte of data in or out per second. To raise the number of I/O threads, use the zmq_ctx_set(3) call before creating any sockets:

There is a major difference between a ØMQ application and a conventional networked application, which is that you don't create one socket per connection. One socket handles all incoming and outgoing connections for a particular point of work. E.g. when you publish to a thousand subscribers, it's via one socket. When you distribute work among twenty services, it's via one socket. When you collect data from a thousand web applications, it's via one socket.

This has a fundamental impact on how you write applications. A traditional networked application has one process or one thread per remote connection, and that process or thread handles one socket. ØMQ lets you collapse this entire structure into a single thread, and then break it up as necessary for scaling.

By default, a ØMQ socket will continue to accept connections until your operating system runs out of file handles. This isn't always the best policy for public-facing services as it leaves you open to a simple denial-of-service attack. You can set a limit using another zmq_ctx_set(3) call:

Underneath the brown paper wrapping of ØMQ's socket API lies the world of messaging patterns. If you have a background in enterprise messaging, or know UDP well, these will be vaguely familiar. But to most ØMQ newcomers they are a surprise, we're so used to the TCP paradigm where a socket maps one-to-one to another node.

Let's recap briefly what ØMQ does for you. It delivers blobs of data (messages) to nodes, quickly and efficiently. You can map nodes to threads, processes, or boxes. It gives your applications a single socket API to work with, no matter what the actual transport (like in-process, inter-process, TCP, or multicast). It automatically reconnects to peers as they come and go. It queues messages at both sender and receiver, as needed. It manages these queues carefully to ensure processes don't run out of memory, overflowing to disk when appropriate. It handles socket errors. It does all I/O in background threads. It uses lock-free techniques for talking between nodes, so there are never locks, waits, semaphores, or deadlocks.

But cutting through that, it routes and queues messages according to precise recipes called patterns. It is these patterns that provide ØMQ's intelligence. They encapsulate our hard-earned experience of the best ways to distribute data and work. ØMQ's patterns are hard-coded but future versions may allow user-definable patterns.

ØMQ patterns are implemented by pairs of sockets with matching types. In other words, to understand ØMQ patterns you need to understand socket types and how they work together. Mostly this just takes learning, there is little that is obvious at this level.

The built-in core ØMQ patterns are:

• Request-reply, which connects a set of clients to a set of services. This is a remote procedure call and task distribution pattern.

• Publish-subscribe, which connects a set of publishers to a set of subscribers. This is a data distribution pattern.

• Pipeline, connects nodes in a fan-out / fan-in pattern that can have multiple steps, and loops. This is a parallel task distribution and collection pattern.

We looked at each of these in the first chapter. There's one more pattern that people tend to try to use when they still think of ØMQ in terms of traditional TCP sockets:

• Exclusive pair, which connects two sockets in an exclusive pair. This is a pattern you should use only to connect two threads in a process. We'll see an example at the end of this chapter.

The zmq_socket(3) man page is fairly clear about the patterns, it's worth reading several times until it starts to make sense. These are the socket combinations that are valid for a connect-bind pair (either side can bind):

• PUB and SUB
• REQ and REP
• REQ and ROUTER
• DEALER and REP
• DEALER and ROUTER
• DEALER and DEALER
• ROUTER and ROUTER
• PUSH and PULL
• PAIR and PAIR

You'll also see references to XPUB and XSUB sockets, which we'll come to later (they're like raw versions of PUB and SUB). Any other combination will produce undocumented and unreliable results and future versions of ØMQ will probably return errors if you try them. You can and will of course bridge other socket types via code, i.e. read from one socket type and write to another.

These four core patterns are cooked-in to ØMQ. They are part of the ØMQ API, implemented in the core C++ library, and guaranteed to be available in all fine retail stores.

On top, we add high-level patterns. We build these high-level patterns on top of ØMQ and implement them in whatever language we're using for our application. They are not part of the core library, do not come with the ØMQ package, and exist in their own space, as part of the ØMQ community. For example the Majordomo pattern, which we explore in Chapter Four, sits in the github Majordomo project in the ZeroMQ organization.

One of the things we aim to provide you with this guide are a set of such high-level patterns, both small (how to handle messages sanely) to large (how to make a reliable publish-subscribe architecture).

On the wire, ØMQ messages are blobs of any size from zero upwards, fitting in memory. You do your own serialization using protobufs, msgpack, JSON, or whatever else your applications need to speak. It's wise to choose a data representation that is portable and fast, but you can make your own decisions about trade-offs.

In memory, ØMQ messages are zmq_msg_t structures (or classes depending on your language). Here are the basic ground rules for using ØMQ messages in C:

• You create and pass around zmq_msg_t objects, not blocks of data.

• To read a message you use zmq_msg_init(3) to create an empty message, and then you pass that to zmq_msg_recv(3).

• To write a message from new data, you use zmq_msg_init_size(3) to create a message and at the same time allocate a block of data of some size. You then fill that data using memcpy[3], and pass the message to zmq_msg_send(3).

• To release (not destroy) a message you call zmq_msg_close(3). This drops a reference, and eventually ØMQ will destroy the message.

• To access the message content you use zmq_msg_data(3). To know how much data the message contains, use zmq_msg_size(3).

• Do not use zmq_msg_move(3), zmq_msg_copy(3), or zmq_msg_init_data(3) unless you read the man pages and know precisely why you need these.

Here is a typical chunk of code working with messages, which should be familiar if you have been paying attention. This is from the zhelpers.h file we use in all the examples:

You can easily extend this code to send and receive blobs of arbitrary length.

Note than when you have passed a message to zmq_msg_send(3), ØMQ will clear the message, i.e. set the size to zero. You cannot send the same message twice, and you cannot access the message data after sending it.

If you want to send the same message more than once, create a second message, initialize it using zmq_msg_init(3) and then use zmq_msg_copy(3) to create a copy of the first message. This does not copy the data but the reference. You can then send the message twice (or more, if you create more copies) and the message will only be finally destroyed when the last copy is sent or closed.

ØMQ also supports multi-part messages, which let you send or receive a list of frames as a single on-the-wire message. This is widely used in real applications and we'll look at that later in this chapter and in Chapter Three.

Some other things that are worth knowing about messages:

• ØMQ sends and receives them atomically, i.e. you get a whole message, or you don't get it at all. This is also true for multi-part messages.

• ØMQ does not send a message right away but at some indeterminate later time.

• You may send zero-length messages, e.g. for sending a signal from one thread to another.

• A message must fit in memory. If you want to send files of arbitrary sizes, you should break them into pieces and send each piece as a separate message.

• You must call zmq_msg_close(3) when finished with a message, in languages that don't automatically destroy objects when a scope closes.

And to be necessarily repetitive, do not use zmq_msg_init_data(3), yet. This is a zero-copy method and guaranteed to create trouble for you. There are far more important things to learn about ØMQ before you start to worry about shaving off microseconds.

In all the examples so far, the main loop of most examples has been:

1. wait for message on socket
2. process message
3. repeat

What if we want to read from multiple sockets at the same time? The simplest way is to connect one socket to multiple endpoints and get ØMQ to do the fan-in for us. This is legal if the remote endpoints are in the same pattern but it would be wrong to e.g. connect a PULL socket to a PUB endpoint.

The right way is to use zmq_poll(3). An even better way might be to wrap zmq_poll(3) in a framework that turns it into a nice event-driven reactor, but it's significantly more work than we want to cover here.

Let's start with a dirty hack, partly for the fun of not doing it right, but mainly because it lets me show you how to do non-blocking socket reads. Here is a simple example of reading from two sockets using non-blocking reads. This rather confused program acts both as a subscriber to weather updates, and a worker for parallel tasks:

C | C++ | C# | Clojure | CL | Erlang | F# | Felix | Go | Java | Lua | Objective-C | Perl | PHP | Python | Ruby | Scala | Tcl | Haxe | Ada | Basic | Haskell | Node.js | ooc | Q | Racket

The cost of this approach is some additional latency on the first message (the sleep at the end of the loop, when there are no waiting messages to process). This would be a problem in applications where sub-millisecond latency was vital. Also, you need to check the documentation for nanosleep() or whatever function you use to make sure it does not busy-loop.

You can treat the sockets fairly by reading first from one, then the second rather than prioritizing them as we did in this example. This is called "fair-queuing", something that ØMQ does automatically when one socket receives messages from more than one source.

Now let's see the same little senseless application done right, using zmq_poll(3):

C | C++ | C# | Clojure | CL | Erlang | F# | Felix | Go | Haskell | Java | Lua | Node.js | Objective-C | Perl | PHP | Python | Ruby | Scala | Tcl | Haxe | Ada | Basic | ooc | Q | Racket

ØMQ's error handling philosophy is a mix of fail-fast and resilience. Processes, we believe, should be as vulnerable as possible to internal errors, and as robust as possible against external attacks and errors. To give an analogy, a living cell will self-destruct if it detects a single internal error, yet it will resist attack from the outside by all means possible.

Assertions, which pepper the ØMQ code, are absolutely vital to robust code, they just have to be on the right side of the cellular wall. And there should be such a wall. If it is unclear whether a fault is internal or external, that is a design flaw to be fixed. In C/C++, assertions stop the application immediately with an error. In other languages you may get exceptions or halts.

When ØMQ detects an external fault it returns an error to the calling code. In some rare cases it drops messages silently, if there is no obvious strategy for recovering from the error.

In most of the C examples we've seen so far there's been no error handling. Real code should do error handling on every single ØMQ call. If you're using a language binding other than C, the binding may handle errors for you. In C you do need to do this yourself. There are some simple rules, starting with POSIX conventions:

• Methods that create objects will return NULL if they fail.

• Methods that process data may return the number of bytes processed, or -1 on an error or failure.

• Other methods will return 0 on success and -1 on an error or failure.

• The error code is provided in errno or zmq_errno(3).

• A descriptive error text for logging is provided by zmq_strerror(3).

There are two main exceptional conditions that you may want to handle as non-fatal:

• When a thread calls zmq_msg_recv(3) with the ZMQ_DONTWAIT option and there is no waiting data. ØMQ will return -1 and set errno to EAGAIN.

• When a thread calls zmq_ctx_destroy(3) and other threads are doing blocking work. The zmq_ctx_destroy(3) call closes the context and all blocking calls exit with -1, and errno set to ETERM.

What this boils down to is that in most cases you can use assertions on ØMQ calls, like this, in C:

In C/C++, asserts can be removed entirely in optimized code, so don't make the mistake of wrapping the whole ØMQ call in an assert(). It looks neat, then the optimizer removes all the asserts and the calls you want to make, and your application breaks in impressive ways.

Let's see how to shut down a process cleanly. We'll take the parallel pipeline example from the previous section. If we've started a whole lot of workers in the background, we now want to kill them when the batch is finished. Let's do this by sending a kill message to the workers. The best place to do this is the sink, since it really knows when the batch is done.

How do we connect the sink to the workers? The PUSH/PULL sockets are one-way only. The standard ØMQ answer is: create a new socket flow for each type of problem you need to solve. We'll use a publish-subscribe model to send kill messages to the workers:

• The sink creates a PUB socket on a new endpoint.
• Workers bind their input socket to this endpoint.
• When the sink detects the end of the batch it sends a kill to its PUB socket.
• When a worker detects this kill message, it exits.

It doesn't take much new code in the sink:

Figure 14 - Parallel Pipeline with Kill Signaling

Here is the worker process, which manages two sockets (a PULL socket getting tasks, and a SUB socket getting control commands) using the zmq_poll(3) technique we saw earlier:

Here is the modified sink application. When it's finished collecting results it broadcasts a KILL message to all workers:

Realistic applications need to shutdown cleanly when interrupted with Ctrl-C or another signal such as SIGTERM. By default, these simply kill the process, meaning messages won't be flushed, files won't be closed cleanly, etc.

Here is how we handle a signal in various languages:

The program provides s_catch_signals(), which traps Ctrl-C (SIGINT) and SIGTERM. When either of these signals arrive, the s_catch_signals() handler sets the global variable s_interrupted. Your application will not die automatically, you have to now explicitly check for an interrupt, and handle it properly. Here's how:

• Call s_catch_signals() (copy this from interrupt.c) at the start of your main code. This sets-up the signal handling.
• If your code is blocking in zmq_msg_recv(3), zmq_poll(3), or zmq_msg_send(3), when a signal arrives, the call will return with EINTR.
• Wrappers like s_recv() return NULL if they are interrupted.
• So, your application checks for an EINTR return code, a NULL return, and/or s_interrupted.

Here is a typical code fragment:

s_catch_signals ();
client = zmq_socket (...);
while (!s_interrupted) {
    char *message = s_recv (client);
    if (!message)
        break;          //  Ctrl-C used
}
zmq_close (client);

If you call s_catch_signals() and don't test for interrupts, the your application will become immune to Ctrl-C and SIGTERM, which may be useful, but is usually not.

Any long-running application has to manage memory correctly, or eventually it'll use up all available memory and crash. If you use a language that handles this automatically for you, congratulations. If you program in C or C++ or any other language where you're responsible for memory management, here's a short tutorial on using valgrind, which among other things will report on any leaks your programs have.

• To install valgrind, e.g. on Ubuntu or Debian: sudo apt-get install valgrind.

• By default, ØMQ will cause valgrind to complain a lot. To remove these warnings, create a file valgrind.supp that contains this:

{
   <socketcall_sendto>
   Memcheck:Param
   socketcall.sendto(msg)
   fun:send
   ...
}
{
   <socketcall_sendto>
   Memcheck:Param
   socketcall.send(msg)
   fun:send
   ...
}

• Fix your applications to exit cleanly after Ctrl-C. For any application that exits by itself, that's not needed, but for long-running applications, this is essential, otherwise valgrind will complain about all currently allocated memory.

• Build your application with -DDEBUG, if it's not your default setting. That ensures valgrind can tell you exactly where memory is being leaked.

• Finally, run valgrind thus:

valgrind --tool=memcheck --leak-check=full --suppressions=valgrind.supp someprog

And after fixing any errors it reported, you should get the pleasant message:

==30536== ERROR SUMMARY: 0 errors from 0 contexts...

ØMQ lets us compose a message out of several frames, giving us a 'multi-part message'. Realistic applications use multi-part messages heavily, both for wrapping messages with address information, and for simple serialization. We'll look at address envelopes later.

What we'll learn now is simply how to safely (but blindly) read and write multi-part messages in any application (like a proxy) that needs to forward messages without inspecting them.

When you work with multi-part messages, each part is a zmq_msg item. E.g. if you are sending a message with five parts, you must construct, send, and destroy five zmq_msg items. You can do this in advance (and store the zmq_msg items in an array or structure), or as you send them, one by one.

Here is how we send the frames in a multi-part message (we receive each frame into a message object):

Here is how we receive and process all the parts in a message, be it single part or multi-part:

Some things to know about multi-part messages:

• When you send a multi-part message, the first part (and all following parts) are only actually sent on the wire when you send the final part.
• If you are using zmq_poll(3), when you receive the first part of a message, all the rest has also arrived.
• You will receive all parts of a message, or none at all.
• Each part of a message is a separate zmq_msg item.
• You will receive all parts of a message whether or not you check the RCVMORE option.
• On sending, ØMQ queues message frames in memory until the last is received, then sends them all.
• There is no way to cancel a partially sent message, except by closing the socket.

ØMQ aims for decentralized intelligence but that doesn't mean your network is empty space in the middle. It's filled with message-aware infrastructure and quite often, we build that infrastructure with ØMQ. The ØMQ plumbing can range from tiny pipes to full-blown service-oriented brokers. The messaging industry calls this "intermediation", meaning that the stuff in the middle deals with either side. In ØMQ we call these proxies, queues, forwarders, device, or brokers, depending on the context.

This pattern is extremely common in the real world and is why our societies and economies are filled with intermediaries who have no other real function than to reduce the complexity and scaling costs of larger networks. Real-world intermediaries are typically called wholesalers, distributors, managers, etc.

One of the problems you will hit as you design larger distributed architectures is discovery. That is, how do pieces know about each other? It's especially difficult if pieces come and go, thus we can call this the "dynamic discovery problem".

There are several solutions to dynamic discovery. The simplest is to entirely avoid it by hard-coding (or configuring) the network architecture so discovery is done by hand. That is, when you add a new piece, you reconfigure the network to know about it.

In practice this leads to increasingly fragile and hard-to-manage architectures. Let's say you have one publisher and a hundred subscribers. You connect each subscriber to the publisher by configuring a publisher endpoint in each subscriber. That's easy. Subscribers are dynamic, the publisher is static. Now say you add more publishers. Suddenly it's not so easy any more. If you continue to connect each subscriber to each publisher, the cost of avoiding dynamic discovery gets higher and higher.

Figure 15 - Small-scale Pub-Sub Network

There are quite a few answers to this but the very simplest answer is to add an intermediary, that is, a static point in the network to which all other nodes connect. In classic messaging, this is the job of the "message broker". ØMQ doesn't come with a message broker as such, but it lets us build intermediaries quite easily.

You might wonder, if all networks eventually get large enough to need intermediaries, why don't we simply always design around a message broker? For beginners, it's a fair compromise. Just always use a star topology, forget about performance, and things will usually work. However message brokers are greedy things; in their role as central intermediaries, they become too complex, too stateful, and eventually a problem.

It's better to think of intermediaries as simple stateless message switches. The best analogy is an HTTP proxy; it's there but doesn't have any special role. Adding a pub-sub proxy solves the dynamic discovery problem in our example. We set the proxy in the "middle" of the network. The proxy opens an XSUB socket, an XPUB socket, and binds each to well-known IP addresses and ports. Then all other processes connect to the proxy, instead of to each other. It becomes trivial to add more subscribers or publishers.

Figure 16 - Pub-Sub Network with a Proxy

We need XPUB and XSUB sockets because ØMQ does subscription forwarding: SUB sockets actually send subscriptions to PUB sockets as special messages. The proxy has to forward these as well, by reading them from the XPUB socket and writing them to the XSUB socket. This is the main use-case for XSUB and XPUB.

Figure 17 - Extended Publish-Subscribe

In the Hello World client-server application we have one client that talks to one service. However in real cases we usually need to allow multiple services as well as multiple clients. This lets us scale up the power of the service (many threads or processes or boxes rather than just one). The only constraint is that services must be stateless, all state being in the request or in some shared storage such as a database.

There are two ways to connect multiple clients to multiple servers. The brute-force way is to connect each client socket to multiple service endpoints. One client socket can connect to multiple service sockets, and the REQ socket will then load-balance requests among these services. Let's say you connect a client socket to three service endpoints, A, B, and C. The client makes requests R1, R2, R3, R4. R1 and R4 go to service A, R2 goes to B, and R3 goes to service C.

Figure 18 - Load-balancing of Requests

This design lets you add more clients cheaply. You can also add more services. Each client will load-balance its requests to the services. But each client has to know the service topology. If you have 100 clients and then you decide to add three more services, you need to reconfigure and restart 100 clients in order for the clients to know about the three new services.

That's clearly not the kind of thing we want to be doing at 3am when our supercomputing cluster has run out of resources and we desperately need to add a couple of hundred new service nodes. Too many static pieces are like liquid concrete: knowledge is distributed and the more static pieces you have, the more effort it is to change the topology. What we want is something sitting in between clients and services that centralizes all knowledge of the topology. Ideally, we should be able to add and remove services or clients at any time without touching any other part of the topology.

So we'll write a little message queuing broker that gives us this flexibility. The broker binds to two endpoints, a frontend for clients and a backend for services. It then uses zmq_poll(3) to monitor these two sockets for activity and when it has some, it shuttles messages between its two sockets. It doesn't actually manage any queues explicitly — ØMQ does that automatically on each socket.

When you use REQ to talk to REP you get a strictly synchronous request-reply dialog. The client sends a request, the service reads the request and sends a reply. The client then reads the reply. If either the client or the service try to do anything else (e.g. sending two requests in a row without waiting for a response) they will get an error.

But our broker has to be non-blocking. Obviously we can use zmq_poll(3) to wait for activity on either socket, but we can't use REP and REQ.

Luckily there are two sockets called DEALER and ROUTER that let you do non-blocking request-response. You'll see in Chapter Three how DEALER and ROUTER sockets let you build all kinds of asynchronous request-reply flows. For now, we're just going to see how DEALER and ROUTER let us extend REQ-REP across an intermediary, that is, our little broker.

In this simple stretched request-reply pattern, REQ talks to ROUTER and DEALER talks to REP. In between the DEALER and ROUTER we have to have code (like our broker) that pulls messages off the one socket and shoves them onto the other.

Figure 19 - Extended Request-reply

The request-reply broker binds to two endpoints, one for clients to connect to (the frontend socket) and one for workers to connect to (the backend). To test this broker, you will want to change your workers so they connect to the backend socket. Here are a client and worker that show what I mean:

Here is the worker:

And here is the broker, which properly handles multi-part messages:

Using a request-reply broker makes your client-server architectures easier to scale since clients don't see workers, and workers don't see clients. The only static node is the broker in the middle.

Figure 20 - Request-reply Broker

It turns out that that core loop in rrbroker is very useful, and reusable. It lets us build pub-sub forwarders and shared queues and other little intermediaries, with very little effort. ØMQ wraps this up in a single method, zmq_proxy(3):

The two (or three sockets, if we want to capture data) must be properly connected, bound, configured. When we call the zmq_proxy method it's exactly like starting the main loop of rrbroker. Let's rewrite the request-reply broker to call zmq_proxy, and re-badge this as an expensive-sounding "message queue" (people have charged houses for code that did less):

If you're like most ØMQ users, at this stage your mind is starting to think, "what kind of evil stuff can I do if I plug random socket types into the proxy?" The short answer is: try it and work out what is happening. In practice you would usually stick to ROUTER/DEALER, XSUB/XPUB, or PULL/PUSH.

A frequent request from ØMQ users is "how do I connect my ØMQ network with technology X?" where X is some other networking or messaging technology. The simple answer is to build a "bridge". A bridge is a small application that speaks one protocol at one socket, and converts to/from a second protocol at another socket. A protocol interpreter, if you like. A common bridging problem in ØMQ is to bridge two transports or networks.

As example, we're going to write a little proxy that sits in between a publisher and a set of subscribers, bridging two networks. The frontend socket (SUB) faces the internal network, where the weather server is sitting, and the backend (PUB) faces subscribers on the external network. It subscribes to the weather service on the frontend socket, and republishes its data on the backend socket.

Figure 21 - Pub-Sub Forwarder Proxy

ØMQ is perhaps the nicest way ever to write multithreaded (MT) applications. Whereas as ØMQ sockets require some readjustment if you are used to traditional sockets, ØMQ multithreading will take everything you know about writing MT applications, throw it into a heap in the garden, pour gasoline over it, and set it alight. It's a rare book that deserves burning, but most books on concurrent programming do.

To make utterly perfect MT programs (and I mean that literally) we don't need mutexes, locks, or any other form of inter-thread communication except messages sent across ØMQ sockets.

By "perfect" MT programs I mean code that's easy to write and understand, that works with one design language in any programming language, and on any operating system, and that scales across any number of CPUs with zero wait states and no point of diminishing returns.

If you've spent years learning tricks to make your MT code work at all, let alone rapidly, with locks and semaphores and critical sections, you will be disgusted when you realize it was all for nothing. If there's one lesson we've learned from 30+ years of concurrent programming it is: just don't share state. It's like two drunkards trying to share a beer. It doesn't matter if they're good buddies. Sooner or later they're going to get into a fight. And the more drunkards you add to the table, the more they fight each other over the beer. The tragic majority of MT applications look like drunken bar fights.

The list of weird problems that you need to fight as you write classic shared-state MT code would be hilarious if it didn't translate directly into stress and risk, as code that seems to work suddenly fails under pressure. Here is a list of "11 Likely Problems In Your Multithreaded Code" from a large firm with world-beating experience in buggy code: forgotten synchronization, incorrect granularity, read and write tearing, lock-free reordering, lock convoys, two-step dance, and priority inversion.

Yeah, we also counted seven, not eleven. That's not the point though. The point is, do you really want that code running the power grid or stock market to start getting two-step lock convoys at 3pm on a busy Thursday? Who cares what the terms actually mean. This is not what turned us on to programming, fighting ever more complex side-effects with ever more complex hacks.

Some widely used models, despite being the basis for entire industries, are fundamentally broken, and shared state concurrency is one of them. Code that wants to scale without limit does it like the Internet does, by sending messages and sharing nothing except a common contempt for broken programming models.

You should follow some rules to write happy multithreaded code with ØMQ:

• You MUST NOT access the same data from multiple threads. Using classic MT techniques like mutexes are an anti-pattern in ØMQ applications. The only exception to this is a ØMQ context object, which is threadsafe.

• You MUST create a ØMQ context for your process, and pass that to all threads that you want to connect via inproc sockets.

• You MAY treat threads as separate tasks, with their own context, but these threads cannot communicate over inproc. However they will be easier to break into standalone processes afterwards.

• You MUST NOT share ØMQ sockets between threads. ØMQ sockets are not threadsafe. Technically it's possible to do this, but it demands semaphores, locks, or mutexes. This will make your application slow and fragile. The only place where it's remotely sane to share sockets between threads are in language bindings that need to do magic like garbage collection on sockets.

If you need to start more than one proxy in an application, for example, you will want to run each in their own thread. It is easy to make the error of creating the proxy frontend and backend sockets in one thread, and then passing the sockets to the proxy in another thread. This may appear to work but will fail randomly. Remember: Do not use or close sockets except in the thread that created them.

If you follow these rules, you can quite easily split threads into separate processes, when you need to. Application logic can sit in threads, processes, boxes: whatever your scale needs.

ØMQ uses native OS threads rather than virtual "green" threads. The advantage is that you don't need to learn any new threading API, and that ØMQ threads map cleanly to your operating system. You can use standard tools like Intel's ThreadChecker to see what your application is doing. The disadvantages are that your code, when it for instance starts new threads, won't be portable, and that if you have a huge number of threads (thousands), some operating systems will get stressed.

Let's see how this works in practice. We'll turn our old Hello World server into something more capable. The original server was a single thread. If the work per request is low, that's fine: one ØMQ thread can run at full speed on a CPU core, with no waits, doing an awful lot of work. But realistic servers have to do non-trivial work per request. A single core may not be enough when 10,000 clients hit the server all at once. So a realistic server must start multiple worker threads. It then accepts requests as fast as it can, and distributes these to its worker threads. The worker threads grind through the work, and eventually send their replies back.

You can of course do all this using a proxy broker and external worker processes, but often it's easier to start one process that gobbles up sixteen cores, than sixteen processes, each gobbling up one core. Further, running workers as threads will cut out a network hop, latency, and network traffic.

The MT version of the Hello World service basically collapses the broker and workers into a single process:

All the code should be recognizable to you by now. How it works:

• The server starts a set of worker threads. Each worker thread creates a REP socket and then processes requests on this socket. Worker threads are just like single-threaded servers. The only differences are the transport (inproc instead of tcp), and the bind-connect direction.

• The server creates a ROUTER socket to talk to clients and binds this to its external interface (over tcp).

• The server creates a DEALER socket to talk to the workers and binds this to its internal interface (over inproc).

• The server starts a proxy that connects the two sockets. The proxy pulls incoming requests fairly from all clients, and distributes those out to workers. It also routes replies back to their origin.

Note that creating threads is not portable in most programming languages. The POSIX library is pthreads, but on Windows you have to use a different API. We'll see in Chapter Three how to wrap this in a portable API.

Here the 'work' is just a one-second pause. We could do anything in the workers, including talking to other nodes. This is what the MT server looks like in terms of ØMQ sockets and nodes. Note how the request-reply chain is REQ-ROUTER-queue-DEALER-REP.

Figure 22 - Multithreaded Server

When you start making multithreaded applications with ØMQ, you'll hit the question of how to coordinate your threads. Though you might be tempted to insert 'sleep' statements, or use multithreading techniques such as semaphores or mutexes, the only mechanism that you should use are ØMQ messages. Remember the story of The Drunkards and the Beer Bottle.

Let's make three threads that signal each other when they are ready. In this example we use PAIR sockets over the inproc transport: