redox

Modern, asynchronous, and wicked fast C++11 client for Redis [] (https://travis-ci.org/hmartiro/redox)

Redox is a C++ interface to the Redis key-value store that makes it easy to write applications that are both elegant and high-performance. Communication should be a means to an end, not something we spend a lot of time worrying about. Redox takes care of the details so you can move on to the interesting part of your project.

Features:

• Expressive asynchronous and synchronous API, templated by return value
• Callbacks can be lambdas, class methods, bind expressions, or any std::function
• Thread-safe - use one client in multiple threads or multiple clients in one
• Automatic pipelining, even for synchronous calls from separate threads
• Low-level access when needed
• Accessible and robust error handling
• Configurable logging level and output to any ostream
• Full support for binary data (keys and values)
• Fast - developed for robotics applications
• 100% clean Valgrind reports

Redox is built on top of hiredis and libev. It uses only the asynchronous API of hiredis, even for synchronous commands. There is no dependency on Boost or any other libraries.

Benchmarks

Benchmarks are given by averaging the results of ten trials of the speed tests in examples/ on an AWS t2.medium instance running Ubuntu 14.04 (64-bit) and a local Redis server.

• speed_test_async_multi over TCP: 879,589 commands/s
• speed_test_async_multi over Unix socket: 901,683 commands/s
• speed_test_async over TCP: 203,285 commands/s
• speed_test_async over Unix socket: 301,823 commands/s
• speed_test_sync over TCP: 21,072 commands/s
• speed_test_sync over Unix socket: 24,911 commands/s

A mid-range laptop gives comparable results. Numbers can be much higher on a high-end machine.

Tutorial

This section introduces the main features of redox. Look in examples/ for more inspiration.

Hello world

Here is the simplest possible redox program:

#include <iostream>
#include <redox.hpp>

using namespace std;
using namespace redox;

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

  Redox rdx;
  if(!rdx.connect("localhost", 6379)) return 1;

  rdx.set("hello", "world!");
  cout << "Hello, " << rdx.get("hello") << endl;

  rdx.disconnect();
  return 0;
}

Compile and run:

$ g++ hello.cpp -o hello -std=c++11 -lredox -lev -lhiredis
$ ./hello
Hello, world!

This example is synchronous, in the sense that the commands don't return until a reply is received from the server.

Asynchronous commands

In a high-performance application, we don't want to wait for a reply, but instead do other work. At the core of Redox is a generic asynchronous API for executing any Redis command and providing a reply callback. The command method accepts a Redis command in the form of an STL vector of strings, and a callback to be invoked when a reply is received or if there is an error.

rdx.command<string>({"GET", "hello"}, [](Command<string>& c) {
  if(c.ok()) {
    cout << "Hello, async " << c.reply() << endl;
  } else {
    cerr << "Command has error code " << c.status() << endl;
  }
});

This statement tells redox to run the command GET hello. The <string> template parameter means that we want the reply to be put into a string and that we expect the server to respond with something that can be put into a string. The full list of reply types is listed in this document and covers convenient access to anything returned from the Redis protocol. The input vector can contain arbitrary binary data.

The second argument is a callback function that accepts a reference to a Command object of the requested reply type. The Command object contains the reply and any error information. If c.ok() is true, the expected reply is accessed from c.reply() (a string in this case). If c.ok() is false, then the error code is given by c.status(), which can report an error or nil reply, a reply of the wrong type, a send error, etc. The callback is guaranteed to be invoked exactly once, and the memory for the Command object is freed automatically once the callback returns.

Here is a simple example of running GET hello asynchronously ten times:

Redox rdx;

// Block until connected, localhost by default
if(!rdx.connect()) return 1;

auto got_reply = [](Command<string>& c) {
  if(!c.ok()) return;
  cout << c.cmd() << ": " << c.reply() << endl;
};

for(int i = 0; i < 10; i++) rdx.command<string>({"GET", "hello"}, got_reply);

// Do useful work
this_thread::sleep_for(chrono::milliseconds(10));

rdx.disconnect(); // Block until disconnected

The .command() method returns immediately, so this program doesn't wait for a reply from the server - it just pauses for ten milliseconds and then shuts down. If we want to shut down after we get all replies, we could do something like this:

Redox rdx;
if(!rdx.connect()) return 1;

int total = 10; // Number of commands to run
atomic_int count(0); // Number of replies expected
auto got_reply = [&](Command<string>& c) {
  count++;
  if(c.ok()) cout << c.cmd() << " #" << count << ": " << c.reply() << endl;
  if(count == total) rdx.stop(); // Signal to shut down
};

for(int i = 0; i < total; i++) rdx.command<string>({"GET", "hello"}, got_reply);

// Do useful work

rdx.wait(); // Block until shut down complete

This example tracks of how how many replies are received and signals the Redox instance to stop once they all process. We use an std::atomic_int to be safe because the callback is invoked from a separate thread. The stop() method signals Redox to shut down its event loop and disconnect from Redis. The wait() method blocks until stop() has been called and everything is brought down. The disconnect() method used earlier is just a call to stop() and then a call to wait().

Synchronous commands

Redox implements synchronous commands by running asynchronous commands and waiting on them with condition variables. That way, we can reap the benefits of pipelining between synchronous commands in different threads. The commandSync method provides a similar API to command, but instead of a callback returns a Command object when a reply is received.

Command<string>& c = rdx.commandSync<string>({"GET", "hello"});
if(c.ok()) cout << c.cmd() << ": " << c.reply() << endl;
c.free();

When using synchronous commands, the user is responsible for freeing the memory of the Command object by calling c.free(). The c.cmd() method just returns a string representation of the command (GET hello in this case).

Looping and delayed commands

We often want to run commands on regular invervals. Redox provides the commandLoop method to accomplish this. It is easier to use and more efficient than running individual commands in a loop, because it only creates a single Command object. commandLoop takes a command vector, a callback, and an interval (in seconds) to repeat the command. It then runs the command on the given interval until the user calls c.free().

Command<string>& cmd = rdx.commandLoop<string>({"GET", "hello"}, [](Command<string>& c) {
  if(c.ok()) cout << c.cmd() << ": " << c.reply() << endl;
}, 0.1);

this_thread::sleep_for(chrono::seconds(1));
cmd.free();
rdx.disconnect();

Finally, commandDelayed runs a command after a specified delay (in seconds). It does not return a command object, because the memory is automatically freed after the callback is invoked.

rdx.commandDelayed<string>({"GET", "hello"}, [](Command<string>& c) {
  if(c.ok()) cout << c.cmd() << ": " << c.reply() << endl;
}, 1);
this_thread::sleep_for(chrono::seconds(2));

Convenience methods

The four methods command, commandSync, commandLoop, and commandDelayed form the core of Redox's functionality. There are convenience methods provided that are simple wrappers over the core methods. Some examples of those are .get(), .set(), .del(), and .publish(). These methods are nice because they return simple values, and there are no Command objects or template parameters. However, they make strong assumptions about how to deal with errors (ignore or throw exceptions), and since their implementations are a few lines of code it is often easier to create custom convenience methods for your application.

Publisher / Subscriber

Redox provides an API for the pub/sub functionality of Redis. Publishing is done just like any other command using a Redox instance. There is a separate Subscriber class that receives messages and provides subscribe/unsubscribe and psubscribe/punsubscribe methods.

Redox rdx; Subscriber sub;
if(!rdx.connect() || !sub.connect()) return 1;

sub.subscribe("hello", [](const string& topic, const string& msg) {
  cout << topic << ": " << msg << endl;
});

for(int i = 0; i < 10; i++) {
  rdx.publish("hello", "this is a pubsub message");
  this_thread::sleep_for(chrono::milliseconds(500));
}

sub.disconnect(); rdx.disconnect();

strToVec and vecToStr

Redox provides helper methods to convert between a string command and a vector of strings as needed by its API. rdx.strToVec("GET foo") will return an std::vector<std::string> containing GET and foo as entries. rdx.vecToStr({"GET", "foo"}) will return the string GET foo.

No-Wait Mode

Redox provides a no-wait mode, which tells the event loop not to sleep in between processing events. It means that the event thread will run at 100% CPU, but it can greatly improve performance when critical. It is disabled by default and can be enabled with rdx.noWait(true);.

Reply types

These the available template parameters in redox and the Redis return types they can hold. If a given command returns an incompatible type you will get a WRONG_TYPE or NIL_REPLY status.

• <redisReply*>: All reply types, returns the hiredis struct directly
• <char*>: Simple Strings, Bulk Strings
• <std::string>: Simple Strings, Bulk Strings
• <long long int>: Integers
• <int>: Integers (careful about overflow, long long int recommended)
• <std::nullptr_t>: Null Bulk Strings, any other receiving a nil reply will get a NIL_REPLY status
• <std::vector<std::string>>: Arrays of Simple Strings or Bulk Strings (in received order)
• <std::set<std::string>>: Arrays of Simple Strings or Bulk Strings (in sorted order)
• <std::unordered_set<std::string>>: Arrays of Simple Strings or Bulk Strings (in no order)

Installation

Instructions provided are for Ubuntu, but all components are platform-independent.

Build from source

Get the build environment and dependencies:

sudo apt-get install git cmake build-essential
sudo apt-get install libhiredis-dev libev-dev

Build the library:

mkdir build && cd build
cmake ..
make

Install into system directories (optional):

sudo make install

Build examples and test suite

Enable examples using ccmake or the following:

cmake -Dexamples=ON ..
make examples

To run the test suite, first make sure you have gtest set up, then:

cmake -Dtests=ON ..
make test_redox
./test_redox

Build documentation

Redox documentation is generated using doxygen.

cd docs
doxygen

The documentation can then be viewed in a browser at docs/html/index.html.

Build RPM and DEB packages

Basic support to build RPMs and DEBs is in the build system. To build them, issue the following commands:

mkdir release && cd release
cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo ..
make package

NOTE: To build RPM packages, you will need rpmbuild.

Contributing

Redox is in its early stages and I am looking for feedback and contributors to make it easier, faster, and more robust. Open issues on GitHub or message me directly.

Redox is not currently recommended for production use. It has no support yet for sentinels or clusters. Feel free to provide them!