Struggling to get boost shared memory segments to construct

Question

I have some code with which I'm trying to make a shared memory segment. This segment is managed, on this end, from within a class. The shared segment will be used in a "bulletin board" fashion. That is, this one process will write to it, and many others will read from it. Without further ado:

#include <string>
#include <sys/types.h>
#include <boost/interprocess/managed_shared_memory.hpp>

static const std::string SHM_NAME("SharedMemory");
static const std::string SHM_STATUS("StatusArray");
static const std::string SHM_INFO1("MfgData");

class Manager {
    u_int8_t* _status;
    char* _info;

    boost::interprocess::managed_shared_memory _shm;

    Manager()
        : _shm(boost::interprocess::create_only, SHM_NAME.c_str(), 1024)
    {
        // the process goes out to lunch on this first call, it's like a deadlock
        status = _shm.construct<u_int8_t>(SHM_STATUS.c_str()) [128] (0); // array 128 bytes, init to 0
        info = _shm.construct<char>(SHM_INFO1.c_str()) [256] (0);
    }

public:
    ~Manager() {
        _shm.destroy<u_int8_t>(SHM_STATUS.c_str());
        _shm.destroy<char>(SHM_INFO1.c_str());
        boost::interprocess::managed_shared_memory_object::remove(SHM_NAME.c_str());
    }

    Manager* Builder() {
        // just in case a previous instance was abnormally terminated
        boost::interprocess::managed_shared_memory_object::remove(SHM_NAME.c_str());
        return new Manager(); // sort of a factory pattern
    }
};

The reading that I have done from the boost website on how to do this suggests that there might be a deadlock. In fact, that's why I put the ctor into the private section and made the builder function: so that, during construction, previous instances could be removed. However, this didn't alleviate the problem.

I tried changing the name of the shared memory segment so that it wouldn't be in use, but still, the process hangs when it gets to those lines of code.

I'm using this link (as well as others from the same documentation site) for my model. At this point, I need a second set of eyes and preferably those experienced eyes of boost interprocess and shared memory.

By the way, the program model that I'm using from that link I provided is the "named shared memory" program. What's quite irritating is that I've copied that program onto my Linux system, built and run it with no trouble. What am I missing?

Thanks for the help, Andy

A stack back trace from a debugger when it is hung would help. — brian beuning, Apr 18 '13 at 15:48
@brianbeuning Your suggestion helped me to find the real culprit. Thanks. It's been a while since I've used gdb. It's been wonderful to set aside the squeegee for a few days. — Andrew Falanga, Apr 18 '13 at 20:50
I ran into a similar symptom while during development I had created a managed_shared_memory object without calling destroy and remove and I later added it but the shared_memory_object I had created still had some sort of mutex or semaphore in place, so I rm -rf /dev/shm and started it over and that helped me. — Michael, Feb 27 '19 at 22:44

score 1 · Accepted Answer · answered Apr 18 '13 at 20:54

The problem appears to have been related to something not related to boost at this point. I asked the question, and tagged it thusly, because when I first encountered the problem, this is precisely where the hangup was happening. I believe that hangup was due a blocking request of a mutex buried deep within the boost libraries for interprocess communication. I believe the code construct I have, putting the ctor private and calling a factory function, has alleviated that issue.

Thanks to Brian's great suggestion, I was able to see in moments where the problem was. The issue in this case was that when opening the fifo I'd made earlier, the option for non-blocking wasn't or'ed into the flags and the process was blocking waiting for another party to open the same pipe for writing.

Struggling to get boost shared memory segments to construct

1 Answers1