Hi Andreas,
Do you think I need to do a whole lfsck?
I got an error when I was generating mdsdb on our test file system (with OST
pool). Errors
are same as discussed at this thread:
http://comments.gmane.org/gmane.comp.file-systems.lustre.user/10111
Do you have any suggestion?
Lu Wang
Computing Center
IHEP
-----????-----
???: "WANG Lu" <wanglu at ihep.ac.cn>
????: 2011?8?2? ???
???: "WANG Lu" <wanglu at ihep.ac.cn>
??: "lustre-discuss at lists.lustre.org" <lustre-discuss at
lists.lustre.org>
??: Re: [Lustre-discuss] lvbo_init failed after e2fsck
Update some information:
1. After running "ll_recover_lost_found_objs", we still have the
"lvbo_init faild" error.
2. There is no files under "lost+found", in some of related OSTs. Here
is the result of debugf:
# debugfs -c /dev/sdb1
debugfs 1.41.10.sun2 (24-Feb-2010)
/dev/sdb1: catastrophic mode - not reading inode or group bitmaps
debugfs: ls
2 (12) . 2 (12) .. 11 (20) lost+found 103784449 (16) CONFIGS
12 (20) last_rcvd 13 (20) health_check 222863361 (3996) O
debugfs: cd lost+found
debugfs: ls
11 (12) . 2 (4084) .. 0 (4096) 0 (4096) 0 (4096)
3. We are currently running Lustre 1.8.5.
Thank you in advance for your help!
Lu Wang
CC-IHEP
> -----????-----
> ???: "WANG Lu" <wanglu at ihep.ac.cn>
> ????: 2011?8?1? ???
> ???: "lustre-discuss at lists.lustre.org" <lustre-discuss at
lists.lustre.org>
> ??:
> ??: [Lustre-discuss] lvbo_init failed after e2fsck
>
> Dear all,
> After an annual e2fsck of all OSTs, two of our OSTs have become read
only with error:
> Jul 25 08:37:34 com04 kernel: LDISKFS-fs error (device sdb1):
ldiskfs_dx_find_entry: bad entry in directory #222863370: inode out of bounds -
offset=3280896, inode=656179638, rec_len=4096, name_len=0
> Jul 25 08:37:34 com04 kernel: Aborting journal on device sdb1-8.
> Jul 25 08:37:34 com04 kernel: LDISKFS-fs (sdb1): Remounting filesystem
read-only
> tune2fs shows the OSTs are at stat "clean with error", after
umount and e2fsck again, the two OSTs could be mount normally(and the stat
changed to "clean").
>
> However, we began to meet hundreds of "lvbo_init failed" on
serveral OSTs, not limited on the two OSTs which have been read-only.
>
> Three of our OSTs have met hundreds of lvbo_init faild after an annual
e2fsck examination.
>
> Aug 1 17:48:26 com04 kernel: LustreError:
5493:0:(ldlm_resource.c:862:ldlm_resource_add()) Skipped 1 previous similar
message
> Aug 1 17:59:02 com04 kernel: LustreError:
5632:0:(ldlm_resource.c:862:ldlm_resource_add()) filter-publicfs-OST001d_UUID:
lvbo_init failed for resource 2997406: rc -2
> Aug 1 17:59:02 com04 kernel: LustreError:
5632:0:(ldlm_resource.c:862:ldlm_resource_add()) Skipped 1 previous similar
message
> Aug 1 18:10:51 com04 kernel: LustreError:
5602:0:(ldlm_resource.c:862:ldlm_resource_add()) filter-publicfs-OST001d_UUID:
lvbo_init failed for resource 3240254: rc -2
> Aug 1 18:10:51 com04 kernel: LustreError:
5602:0:(ldlm_resource.c:862:ldlm_resource_add()) Skipped 2 previous similar
messages
> Aug 1 18:21:49 com04 kernel: LustreError:
5642:0:(ldlm_resource.c:862:ldlm_resource_add()) filter-publicfs-OST001f_UUID:
lvbo_init failed for resource 3204200: rc -2
> Aug 1 18:21:49 com04 kernel: LustreError:
5642:0:(ldlm_resource.c:862:ldlm_resource_add()) Skipped 6 previous similar
messages
> Aug 1 18:53:18 com04 kernel: LustreError:
5324:0:(ldlm_resource.c:862:ldlm_resource_add()) filter-publicfs-OST001f_UUID:
lvbo_init failed for resource 12856264: rc -2
> Aug 1 18:53:18 com04 kernel: LustreError:
5324:0:(ldlm_resource.c:862:ldlm_resource_add()) Skipped 1 previous similar
message
>
> According to previous discussions, it seems that the related Objects
have been deleted or moved to lost+found. I am not sure:
> 1. if the commmand " ll_recover_lost_found_objs" can get back
all the lost objects
> 2. if not, how can I get a list of demaged files?
> 3. as users continuely writing new data to the OSTs, the number of
demaged Objects will increase?
>
> do you have any suggestion? Thank you very much!
>
>
> Lu Wang
> Computing Center
> IHEP,China
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss