The import-rados feature (#8276) uses librados so in my wip-8231 branch I now link with librados. It is hard to reproduce, but I’ll play with that commit and branch. David Zafman Senior Developer http://www.inktank.com http://www.redhat.com On Aug 21, 2014, at 4:56 PM, Sage Weil <sweil@xxxxxxxxxx> wrote: > On Thu, 21 Aug 2014, Gregory Farnum wrote: >> On Thu, Aug 21, 2014 at 4:37 PM, David Zafman <david.zafman@xxxxxxxxxxx> wrote: >>> >>> Has anyone seen anything like this from an application linked with librados using valgrind? Or a Segmentation fault on exit from such an application? >>> >>> Invalid free() / delete / delete[] / realloc() >>> at 0x4C2A4BC: operator delete(void*) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) >>> by 0x8195C12: std::basic_string<char, std::char_traits<char>, std::allocator<char> >::~basic_string() (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.16) >>> by 0x13890F3: coll_t::~coll_t() (osd_types.h:468) >>> by 0x8944DEC: __cxa_finalize (cxa_finalize.c:56) >>> by 0x6E1CEC5: ??? (in /src/ceph/src/.libs/librados.so.2.0.0) >>> by 0x725F400: ??? (in /src/ceph/src/.libs/librados.so.2.0.0) >>> by 0x89449D0: __run_exit_handlers (exit.c:78) >>> by 0x8944A54: exit (exit.c:100) >>> by 0x137FF37: usage(boost::program_options::options_description&) (ceph_objectstore_tool.cc:1794) >>> by 0x1380572: main (ceph_objectstore_tool.cc:1849) >> >> This looks fairly strange to me ? why does ceph_objectstore_tool do >> anything with librados? I thought it was just hitting the OSD >> filesystem structure directly. >> Also note that the crash appears to be underneath the coll_t >> destructor, probably in destroying its string. That combined with the >> weird librados presence makes me think memory corruption is running >> over the stack somewhere. > > Ah, this was fixed in 5d79605319fcde330bccce5e1b07276a98be02de in the > wip-libcommon branch. The problem is partly when we link libcommon > staticaly (ceph-objectstore-tool) and dynamically (librados) at teh same > time. The easy fix here is not linking librados at all. > > Not sure why we see this sometimes and not always.. maybe link order? In > any case, wip-libcommon moves libcommon.la into a .so shared between > librados and the binary using it to avoid the problem. Makes things > slightly more restrictive with mixed versions, but i suspect it is worth > avoiding this sort of pain. > > Can you cherry-pick that commit and see if it resolves this for you? > And/or merge in that entire branch? > > sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
