Ceph

09/15/2017

08:11 AM RADOS Bug #21388: inconsistent pg but repair does nothing reporting head data_digest != data_digest from auth oi / hopefully data seems ok

We did an scp of the OSD object file then a rados put and deep-scrub got it back to OK.
Thanks for your quick answer! Laurent GUERBY

09/14/2017

01:37 PM RADOS Bug #21388 (Duplicate): inconsistent pg but repair does nothing reporting head data_digest != data_digest from auth oi / hopefully data seems ok

ceph pg repair is currently not fixing three "inconsistent" objects
on one of our pg on a replica 3 pool.
For al... Laurent GUERBY

12/15/2015

04:15 PM Ceph Bug #14009: FAILED assert(old_size == total_chunk_size) in 0.94.5 / strange file size

The bug is about the OSDs that shouldn't crash in this case where only one shard in a EC 8+2 setting is corrupted in ... Laurent GUERBY

12/07/2015

05:11 PM Ceph Bug #14009: FAILED assert(old_size == total_chunk_size) in 0.94.5 / strange file size

There is a strange file on disk described here
http://tracker.ceph.com/issues/13972#note-32
http://tracker.ceph.c... Laurent GUERBY

05:09 PM Ceph Bug #14009: FAILED assert(old_size == total_chunk_size) in 0.94.5 / strange file size

OSD crashing the most often is 30. The other two are 9 and 33.
Various log files:
http://guerby.org/ftp/ooops-30.... Laurent GUERBY

05:06 PM Ceph Bug #14009: FAILED assert(old_size == total_chunk_size) in 0.94.5 / strange file size

... Laurent GUERBY

05:05 PM Ceph Bug #14009 (Can't reproduce): FAILED assert(old_size == total_chunk_size) in 0.94.5 / strange file size

Three of our OSD are failing with
2015-12-06 09:44:22.271873 7f47091fc700 -1 osd/ECUtil.cc: In function 'int ECUt... Laurent GUERBY

09:45 AM Ceph Bug #13972: osd/ECUtil.h: 117: FAILED assert(old_size == total_chunk_size) in 0.80.10

pg dump Laurent GUERBY

09:44 AM Ceph Bug #13972: osd/ECUtil.h: 117: FAILED assert(old_size == total_chunk_size) in 0.80.10

osd and pg dump Laurent GUERBY

09:33 AM Ceph Bug #13972: osd/ECUtil.h: 117: FAILED assert(old_size == total_chunk_size) in 0.80.10

crushmap per Loic request. Laurent GUERBY