Ceph » CephFS

2013-09-18 01:44:39.259611 7f17b8379700 -1 mds/MDLog.h: In function 'LogSegment* MDLog::get_current_segment()' thread 7f17b8379700 time 2013-09-18 01:44:39.255228
mds/MDLog.h: 172: FAILED assert(!segments.empty())

 ceph version 0.68-231-g4216eac (4216eac0f59af60f60d4ce909b9ace87a7b64ccc)
 1: ceph-mds() [0x564128]
 2: (Locker::check_inode_max_size(CInode*, bool, bool, unsigned long, bool, unsigned long, utime_t)+0x136f) [0x6d3ecf]
 3: (MDCache::start_files_to_recover(std::vector<CInode*, std::allocator<CInode*> >&, std::vector<CInode*, std::allocator<CInode*> >&)+0x52) [0x622562]
 4: (MDCache::open_snap_parents()+0x9c8) [0x673288]
 5: (MDCache::rejoin_gather_finish()+0x115) [0x676055]
 6: (MDCache::rejoin_send_rejoins()+0x1319) [0x67f179]
 7: (MDCache::process_imported_caps()+0x635) [0x6816f5]
 8: (MDCache::rejoin_open_ino_finish(inodeno_t, int)+0x5d0) [0x6822b0]
 9: (Context::complete(int)+0x9) [0x56ce49]
 10: (finish_contexts(CephContext*, std::list<Context*, std::allocator<Context*> >&, int)+0x95) [0x5906c5]
 11: (MDCache::open_ino_finish(inodeno_t, MDCache::open_ino_info_t&, int)+0xf5) [0x656465]
 12: (MDCache::_open_ino_traverse_dir(inodeno_t, MDCache::open_ino_info_t&, int)+0x1cb) [0x68a33b]
 13: (Context::complete(int)+0x9) [0x56ce49]
 14: (MDS::_dispatch(Message*)+0x454) [0x58c5c4]
 15: (MDS::ms_dispatch(Message*)+0x1d3) [0x58dc13]
 16: (DispatchQueue::entry()+0x549) [0x952d49]
 17: (DispatchQueue::DispatchThread::entry()+0xd) [0x8760cd]
 18: (()+0x7e9a) [0x7f17bc0ede9a]
 19: (clone()+0x6d) [0x7f17ba68accd]
 NOTE: a copy of the executable, or `objdump -rdS <executable>` is needed to interpret this.

/a/teuthology-2013-09-17_23:01:16-fs-next-testing-basic-plana/1234

We've hit this assert before (#5458), but not with this backtrace. The MDS was in standby-reply, got moved to replay, and pretty quickly moved to rejoin. I'm guessing something about the sequence means we emptied out our segments while in standby-replay and then never put anything back in it once we switched to active.