Le mercredi 17 avril 2013 à 20:52 +0200, Olivier Bonvalet a écrit : > What I didn't understand is why the OSD process crash, instead of > marking that PG "corrupted", and does that PG really "corrupted" are > is > this just an OSD bug ? Once again, a bit more informations : by searching informations about one of this faulty PG (3.d), I found that : -592> 2013-04-20 08:31:56.838280 7f0f41d1b700 0 log [ERR] : 3.d osd.25 inconsistent snapcolls on a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 found expected 12d7 -591> 2013-04-20 08:31:56.838284 7f0f41d1b700 0 log [ERR] : 3.d osd.4 inconsistent snapcolls on a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 found expected 12d7 -590> 2013-04-20 08:31:56.838290 7f0f41d1b700 0 log [ERR] : 3.d osd.4: soid a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 size 4194304 != known size 0 -589> 2013-04-20 08:31:56.838292 7f0f41d1b700 0 log [ERR] : 3.d osd.11 inconsistent snapcolls on a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 found expected 12d7 -588> 2013-04-20 08:31:56.838294 7f0f41d1b700 0 log [ERR] : 3.d osd.11: soid a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 size 4194304 != known size 0 -587> 2013-04-20 08:31:56.838395 7f0f41d1b700 0 log [ERR] : scrub 3.d a8620b0d/rb.0.15c26.238e1f29.000000004603/12d7//3 on disk size (0) does not match object info size (4194304) I prefered to verify, so I found that : # md5sum /var/lib/ceph/osd/ceph-*/current/3.d_head/DIR_D/DIR_0/DIR_B/DIR_0/rb.0.15c26.238e1f29.000000004603__12d7_A8620B0D__3 217ac2518dfe9e1502e5bfedb8be29b8 /var/lib/ceph/osd/ceph-4/current/3.d_head/DIR_D/DIR_0/DIR_B/DIR_0/rb.0.15c26.238e1f29.000000004603__12d7_A8620B0D__3 (4MB) 217ac2518dfe9e1502e5bfedb8be29b8 /var/lib/ceph/osd/ceph-11/current/3.d_head/DIR_D/DIR_0/DIR_B/DIR_0/rb.0.15c26.238e1f29.000000004603__12d7_A8620B0D__3 (4MB) d41d8cd98f00b204e9800998ecf8427e /var/lib/ceph/osd/ceph-25/current/3.d_head/DIR_D/DIR_0/DIR_B/DIR_0/rb.0.15c26.238e1f29.000000004603__12d7_A8620B0D__3 (0B) So this object is identical on OSD 4 and 11, but is empty on OSD 25. Since 4 is the master, this should not be a problem, so I try a repair, without any success : ceph pg repair 3.d Is there a way to force rewrite of this replica ? _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com