Bug #10468
closedosd crashed at OpRequest::_dump_op_descriptor_unlocked
0%
Description
when test cephfs,osd crush.
ceph version 0.90 (08bd1e1eee37c80ee033c067f678236da38b1694)
coredump information:
(gdb) where
#0 0x00000000006d9c2a in OpRequest::_dump_op_descriptor_unlocked(std::basic_ostream<char, std::char_traits<char> >&) const ()
#1 0x00000000006dc67f in OpTracker::_mark_event(TrackedOp*, std::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, utime_t) ()
#2 0x00000000006dc861 in TrackedOp::mark_event(std::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) ()
#3 0x00000000006de528 in OpTracker::RemoveOnDelete::operator()(TrackedOp*) ()
#4 0x0000000000693719 in std::tr1::__shared_count<(_gnu_cxx::_Lock_policy)2>::~_shared_count() ()
#5 0x000000000067ab76 in OSD::ms_fast_dispatch(Message*) ()
#6 0x0000000000bb9959 in DispatchQueue::fast_dispatch(Message*) ()
#7 0x0000000000b220bc in Pipe::reader() ()
#8 0x0000000000b23cbd in Pipe::Reader::entry() ()
#9 0x0000003808007851 in start_thread () from /lib64/libpthread.so.0
#10 0x00000038078e767d in clone () from /lib64/libc.so.6
(gdb)
Updated by qf zhao over 9 years ago
We tested a long period of time about cephs,metadata poll and data poll are separated,we write 64KB file about 48 hours,then one osd is down, maybe the 'Message' parameter in OSD::ms_fast_dispatch(Message*) function is free before _dump_op_descriptor_unlocked function use it.