Failed opening pool: flock, Resource temporarily unavailable

[xujiaqi01]

I run the following code with pybind

     int my_id = 0;
     const int POOLSIZE = 1024 * 1024 * 64;
     const std::string LAYOUT = "concurrent_hash_map";
     try {
       std::string file = "./my_test/test_pmemobj_" + std::to_string(getpid()) + "_" + std::to_string(my_id);
       if (file_exists(file) != 0) {
         pop_ = pmem::obj::pool<root>::create(file, LAYOUT, POOLSIZE);
       } else {
         pop_ = pmem::obj::pool<root>::open(file, LAYOUT);
       }
     } catch (pmem::pool_error &e) {
         std::cerr << e.what() << std::endl;
         assert(false);
     }

I run two process with root user, and see files are created

but there are errors:

Failed opening pool: flock "./test_pmemobj_1076_0": Resource temporarily unavailable

Failed opening pool: flock "./test_pmemobj_1075_0": Resource temporarily unavailable

I am also wondering why the files are "----------" instead of "-rw-------"

Thanks for your help!

[Łukasz Stolarczuk]

Hi,

As for the file mode - it's weird, by default, if mode is not given to create function, it should equal to DEFAULT_MODE = S_IWUSR | S_IRUSR. Please verify if these constants are available in sys/stat.h in your system. Your code creates file (on my Ubuntu 20.10) with rw access.
What OS are you running?

As for errors - when do you get these, exactly?

[Igor Chorążewicz]

Also, it would be good to know on what filesystem you're running this code.

[xujiaqi01]

OK, thanks, I will check the file open mode. The filesystem is disk, I first compile and test my code in a docker which does not have aep, and after the test is ok, I will try running my code on aep machine.

[xujiaqi01]

I find that when I compile my c++ code to executable, and run it, the mode of file created is ok, and my code runs well. But when I use SWIG, my code is a .so file , and run my code through python, then the same error occurs as before.

this is my python env:

which python
/usr/bin/python
-----------------------------------------
/usr/bin/python --version
Python 3.6.8
-----------------------------------------
ll /usr/bin/python
lrwxrwxrwx 1 root root 16 Sep 27 12:30 /usr/bin/python -> /usr/bin/python3

this is my .so file:

-rwxr-xr-x  1 root root 391193384 Mar 17 11:44 my_test.so

the OS is centos, and the error occurs at pmem::obj::pool<root>::create

[Igor Chorążewicz]

If the code works when compiled to executable I think this is some environmental issue. Could you show us the entire code (both c++ and python)?

Also, please clarify the filesystem which you are using. Is it ext4, tmpfs or something else?

[xujiaqi01]

Thanks! I will first check the filesystem, and then check my code again.

[xujiaqi01]

Thanks for your help ! I finally find the problem, it is because there are multi threads run the same code, and they create the same file name at same time, so I change my_id to static, and use mutex for my_id ++ for each thread to get its unique file name, and then the file mode is ok, as following:

[xujiaqi01]

👍 👍 Pmem is very useful, next I will try using hash map on it, to save memory, and observe the performance.

[Łukasz Stolarczuk]

It's great to hear that! If you have any more comments, questions or observations please share with us.

[xujiaqi01]

OK, thanks！

[Łukasz Stolarczuk]

I'll allow myself to close this issue, since the merit is solved.
Please contact us again in case you have more questions :)