Internally, we have a program using librados C++ api to periodiclly check ceph's cluster status and send alerts to our internal system.
The little program is written using c++, and linked with librados and our base c++ library. The program works well with K version, but it core dumped with L version. Here is the minmum code to reproduce:
#include <iostream>
#include <rados/librados.hpp>
// Thread class from our base library
class Thread {
public:
Thread() {
}
};
int main() {
Thread t;
librados::Rados cluster;
int error_code = cluster.init2("client.admin", "ceph", /* flags = */0);
// ... code omitted
return 0;
}
# g++ -I/path/to/ceph/src/include ./test.cpp -L/path/to/ceph/build/lib -lrados
# export LD_LIBRARY_PATH=/path/to/ceph/build/lib
# ./a.out
Here is backtrace for the coredump:
ceph/src/log/Log.cc: In function 'void ceph::logging::Log::start()' thread 7f839ed72180 time 2018-07-30 14:55:58.340036
ceph/src/log/Log.cc: 509: FAILED assert(!is_started())
ceph version 14.0.0-1698-gcecbf3e (cecbf3e5dda2dc132367d6d5d74b1b384846c31a) nautilus (dev)
1: (ceph::__ceph_assert_fail(char const*, char const*, int, char const*)+0x14e) [0x7f83951305ce]
2: (()+0x2e8787) [0x7f8395130787]
3: (ceph::logging::Log::stop()+0) [0x7f839534b130]
4: (CephContext::CephContext(unsigned int, code_environment_t, int)+0x1bd7) [0x7f839513f677]
5: (common_preinit(CephInitParameters const&, code_environment_t, int)+0x4e) [0x7f83951677be]
6: (()+0x884aa) [0x7f839e9c54aa]
7: (rados_create2()+0xa8) [0x7f839e9c5888]
8: ./a.out() [0x400be4]
9: (__libc_start_main()+0xf5) [0x7f839dd02b35]
10: ./a.out() [0x400af9]
Aborted
The reason for the core dump is the Thread::Thread in our base c++ library conflicts with librados from ceph's common Thread https://github.com/ceph/ceph/blob/master/src/common/Thread.h#L41
Both of them are in global namespace.
Maybe we'd better put https://github.com/ceph/ceph/blob/master/src/common in ceph or rados namespace. Or build librados using -Wl,-Bsymbolic -Wl,-Bsymbolic-functions flag.
Best wishes,
Yao Zongyou
