[Gluster-devel] getting Input/output error on some files when using tla greater than patch-634

Mickey Mazarick mic at digitaltadpole.com
Fri Nov 21 07:57:10 UTC 2008

None of the copies were deleted from the backend (we even did a search 
and verify) the strange thing is this doesn't occur on patch-634. We are 
also seeing the same thing on a a afr/stripe/unify (so I'm thinking afr 
is the culprit). once again with the strip/afr/unify example it's only 
some of the files,.

We did try to delete all but one copy directly from the backend servers 
to repair some of the files but the error persisted.

-Mickey Mazarick

Vikas Gorur wrote:
> 2008/11/21 Mickey Mazarick <mic at digitaltadpole.com>:
>> This only occurs on about 10% of the files on an afr/unify over ibverbs
>> mount
>> I'm seeing the following:
>> # cat /scripts/setupMachine.sh
>> cat: /scripts/setupMachine.sh: Input/output error
>> glusterfs log says:
>> 2008-11-20 20:03:12 W [afr-self-heal-common.c:943:afr_self_heal] afr1:
>> performing self heal on /scripts/setupMachine.sh (metadata=0 data=1 entry=0)
>> 2008-11-20 20:03:12 E [afr-self-heal-data.c:767:afr_sh_data_fix] afr1:
>> Unable to resolve conflicting data of /scripts/setupMachine.sh. Please
>> resolve manually by deleting the file /scripts/setupMachine.sh from all but
>> the preferred subvolume
>> 2008-11-20 20:03:12 W [afr-self-heal-data.c:70:afr_sh_data_done] afr1: self
>> heal of /scripts/setupMachine.sh completed
>> 2008-11-20 20:03:12 E [unify.c:928:unify_open_cbk] unify: Open success on
>> namespace, failed on child node
>> 2008-11-20 20:03:12 E [fuse-bridge.c:605:fuse_fd_cbk] glusterfs-fuse: 297:
>> OPEN() /scripts/setupMachine.sh => -1 (Input/output error)
>> There are only 2 files on 2 of the 6 storage systems and they are look
>> identical. I have the same results if I delete one of them. Is there an easy
>> fix?
> Did you delete one of the copies on the backend? If yes, then it looks
> like self heal is being spuriously triggered. Thanks for reporting, I
> will try to reproduce this behavior.
> Vikas


More information about the Gluster-devel mailing list