On Mon, 4 Oct 2010, Sage Weil wrote: > On Tue, 5 Oct 2010, Leander Yu wrote: > > Hi, > > I have a 46 machines cluster(44 osd/mon + 2 mds) running ceph now. MDS > > is running in active/standby mode. > > This morning one of the MDS suicide the log shows: > > > > ------------------------------------------- > > 2010-10-04 22:24:19.450022 7f2e5a1ee710 mds0.cache.ino(10000002b87) > > pop_projected_snaprealm 0x7f2e50cd9f70 seq1 > > 2010-10-04 22:26:12.180854 7f2debbfb710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.106:0/2453428678 pipe(0x7f2e380013d0 sd=-1 pgs=2 cs=1 > > l=0).fault with nothing to send, going to standby > > 2010-10-04 22:26:12.181019 7f2e481dc710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.111:0/18905730 pipe(0x7f2e38002250 sd=-1 pgs=2 cs=1 > > l=0).fault with nothing to send, going to standby > > 2010-10-04 22:26:12.181041 7f2dc3fff710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.114:0/1945631186 pipe(0x7f2e38000f00 sd=-1 pgs=2 cs=1 > > l=0).fault with nothing to send, going to standby > > 2010-10-04 22:26:12.181149 7f2deaef6710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.113:0/521184914 pipe(0x7f2e38002f90 sd=-1 pgs=2 cs=1 > > l=0).fault with nothing to send, going to standby > > 2010-10-04 22:26:12.181563 7f2deb5f5710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.112:0/4272114728 pipe(0x7f2e38002ac0 sd=-1 pgs=2 cs=1 > > l=0).fault with nothing to send, going to standby > > 2010-10-04 22:26:13.777624 7f2e5a1ee710 mds-1.3 handle_mds_map i > > (192.168.1.103:6800/2081) dne in the mdsmap, killing myself > > 2010-10-04 22:26:13.777649 7f2e5a1ee710 mds-1.3 suicide. wanted > > up:active, now down:dne > > 2010-10-04 22:26:13.777769 7f2e489e4710 -- 192.168.1.103:6800/2081 >> > > 192.168.1.101:0/15702 pipe(0x7f2e380008c0 sd=-1 pgs=1847 cs=1 > > l=0).fault with nothing to send, going to standby > > ------------------------------------------------------------------------------ > > Would you suggest how do I trouble shooting this issue? or should I > > just restart the mds to recover it? > > The MDS killed itself because it was removed from the mdsmap. The > monitor log will tell you why if you had logging turned up. If not, you > might find some clue by looking at each mdsmap iteration. If you do > > $ ceph mds stat > > it will tell you the map epoch (e###). You can then dump any map > iteration with > > $ ceph mds dump 123 -o - > > Work backward a few iterations until you find which epoch removed that mds > instance. The one prior to that might have some clue (maybe it was > laggy?)... Okay, looking at the maps on your cluster, it looks like there was a standby mds, and the live one was marked down. Probably some intermittent network issue preventing it from sending the monitor beacon on time, and the monitor decided it was dead/unresponsive. The standby cmds took over successfully. The recovery looks like it took about 20 seconds. 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