In APIs create_slave_connection() and create_master_connection(), they will return directly after establishing only one connection, so we have at most 2 connections available any time.

Why not make the connection to all Masters & Slaves, then redis3m users have more connections available, which might be better for load sharing.

We're using a redis installation that requires authentication but it seems there is no support for authentication in the current version. Is this planned?

Hi, I was trying to get cmake to build a .a instead of a .so but I could not seem to be able to do it. However, it seems if you change the add_library call in CMakeLists.txt to:

add_library(${PROJECT_NAME} ${SRC_FILES} ${CMAKE_CURRENT_BINARY_DIR}/datadir.cpp ${INCLUDE_FILES})

(ie, remove SHARED), then the default behavior will still be to generate a shared library. But if you want a static library, you can run cmake -DBUILD_SHARED_LIBS:BOOL=OFF and then it will build a static library when running "make." Do you think this change is acceptable?

Lets say my use case is, There are 2 processes trying to acquire lock on some key. One of them succeed, while other will just wait. Once first one is done updating the value it should release lock and then 2nd process can acquire lock if want or use the value updated by 1st one. I am talking about similar methods from StackExchange.Redis for c#

1. LockTake
2. LockRelease
3. LockExtend
4. LockQuery

hi, I added Visual studio 2013 solution and also updated the code to c++11 where I could.

I have defined a NO_BOOST preprocessor macro (most of the functionality except boost::format seems to be in c++11 and for various reasons I can't use boost in my code). This does litter the code a bit, so feel free to make suggestions on how to resolve those.

My preference would be to remove boost completely where it can be removed. The one big area I didn't touch is scheduler.cpp/.h where I believe std::chrono could be used instead of boost..

thanks.

I'm looking at the connection_pool::create() function that receives a list of sentinel hostnames separated by commas. But then it seems that the port number has to be the same for all sentinels. Is there a way to specify different port numbers for each sentinel?

Thank you!

Hi,

The subscriber.cpp example shows how to make an infinite loop that receives messages, but how do I break out of that loop?

I'm running the subcriber in a separate thread, with a connection that is not being used for anything else. Whenever the get_reply() function returns, it always returns 3 elements: {"message", channelname, content }. But how do I make it stop?

I tried the following in the main thread: conn->run(command("UNSUBSCRIBE") << channelname);

That is, using the same connection that is being used by the subscriber thread. After this line of code is executed, the get_reply() function in the subscriber thread returns this: { "unsubscribe", channelname, "" }.

So, I found a way for the subcriber thread to unblock and know when to break out of the loop and return: it just needs to check if the first element of the reply is "message" or "unsubscribe".

However, now the call to the unsubscribe command in the main thread blocks forever! How am I supposed to do this then?

Thank you very much!

I just wonder, why sometimes is explicit locking on a mutex object is used instead of a unique_lock with explicit unlock call.

In the file: redis3m/src/connection_pool.cpp or redis3m/src/simple_pool.cpp I code like that:

connection::ptr_t simple_pool::get()
{
    connection::ptr_t ret;

    access_mutex.lock();
    std::set<connection::ptr_t>::iterator it = connections.begin();
    if (it != connections.end())
    {
        ret = *it;
        connections.erase(it);
    }

    access_mutex.unlock();

    if (!ret)
    {
        ret = connection::create(_host, _port);
        // Setup connections selecting db
        if (_database != 0)
        {
            ret->run(command("SELECT") << _database);
        }
    }
    return ret;
}

Why not writing it like:

connection::ptr_t simple_pool::get()
{
    connection::ptr_t ret;

    std::unique_lock<std::mutex> lock(access_mutex);
    std::set<connection::ptr_t>::iterator it = connections.begin();
    if (it != connections.end())
    {
        ret = *it;
        connections.erase(it);
    }

    lock.unlock();

    if (!ret)
    {
        ret = connection::create(_host, _port);
        // Setup connections selecting db
        if (_database != 0)
        {
            ret->run(command("SELECT") << _database);
        }
    }
    return ret;
}

In fact unique_lock is used 30 lines below...

I understand that set::erase(iterator) version has the nothrow guarantee, but still I think consistency would be better. And there is no performance drawback in that case.

Hi,

When I store a binary string (in redis-cli this looks like "\r\x04\x00\x04\x05") it will not be properly retrieved. The reply will be cut off and look like this. {_type=STRING (1) _str="\r\x4" _integer=0 ...}

Is there any workaround or pending fix? Otherwise I have to use another library :(

Regards

Hey, I was trying to set a key value having "-" in it. But it doesn't work with reidis3m library.

Below command doesn't set a key-value pair in redis. However I am able to set this value using Redis-client. connection->run(command("SET")<<"mykey"<<"myvalue-1");

Hello,

When I'm attempting to build the package, the cmake works perfectly fine. When I run make however, it gets to about 17% then starts erroring hiredis/hiredis.h no such file or directory.

I've included all output from the git clone to the make in the following pastebin.

Thanks!! http://pastebin.com/XfxkdwW0

PS. I handbuilt this on a different system (worked fine) and it caused a DRASTIC reduction in wait times & greatly improved performance. So great job & thanks again.

This is a great project and very solid. Thank you. Also, I have a query...

I build the VS solution and have a problem with the makefile examples project. The lib builds fine.

Also when I try and build with CMAKE I am struggling to disable BOOST using a flag. Seems strange.

Best wishes

sudo apt-get install libmsgpack-dev libboost-thread-dev libboost-date-time-dev libboost-test-dev libboost-filesystem-dev libboost-system-dev libhiredis-dev cmake build-essential

Unable to install these dependencies on redis3m on Centos 7. Is there a way to get around with this?

Below code in cluster_pool.h is to define less operator for host_t. The class host_t is used as key in a map.
I think the implemetion has a error. Below two redis host are different address, they should be as different values and can be inserted into a map. But the current implemention will treat them as the same value ,since the port are the same, the less function will return false. Please have a look at it. redis1 : address: 192.168.10.10, port: 6379 redis2 : address: 192.168.10.11, prot: 6379

inline bool operator < (const host_t& rvalue) const
{
      return (address <= rvalue.address) && ( port < rvalue.port);
}

I use redis3m as git submodule in my project:

$ tree -L 1 ch1/
ch1/
├── build
├── CMakeCache.txt
├── CMakeFiles
├── cmake_install.cmake
├── CMakeLists.txt
├── gtest/
├── main.cc
├── Makefile
├── redis3m/
└── test

I add CMake add_subdirectory to compile redis3m:

add_subdirectory(redis3m)

Then I got the following error message:

[ 41%] Built target redis3m
make[2]: *** No rule to make target `Doxyfile', needed by `redis3m/docs'.  Stop.
make[1]: *** [redis3m/CMakeFiles/apidoc.dir/all] Error 2
make: *** [all] Error 2

I fix the error by changing CMAKE_SOURCE_DIR to CMAKE_CURRENT_SOURCE_DIR:

 FIND_PACKAGE(Doxygen QUIET)
 IF (DOXYGEN_FOUND)
-        SET(DOXYGEN_INPUT "${CMAKE_SOURCE_DIR}/Doxyfile")
+      SET(DOXYGEN_INPUT "${CMAKE_CURRENT_SOURCE_DIR}/Doxyfile")
   SET(DOXYGEN_OUTPUT "docs")

I think CMAKE_CURRENT_SOURCE_DIR better than CMAKE_SOURCE_DIR when compiling redis3m in a project with multiple dependencies (When compiling redis3m alone, the two variable is same).

I am running multiple commands and reusing reply object sometimes i find abnormal behavior e.g.

command t = command("SET") << "foor" << "bar" ; reply myreply = conn->run(t);

t = command("GET") << "foor"; myreply = conn->run(t);