Re: xfs corruption issue

[Danny Shavit <danny@xxxxxxxxxxxxxxxxx>]

Thanks guys.

So far we did not figure out the bit fllip.

Will update if there is interesting information.

Best regards,

Danny

On Wed, Apr 1, 2015 at 8:12 PM, Eric Sandeen <sandeen@xxxxxxxxxxx> wrote:

On 4/1/15 10:09 AM, Danny Shavit wrote:
> Hello Dave,
> My name is Danny Shavit and I am with Zadara storage.
> We will appreciate your feedback reagrding an xfs_corruption and xfs_reapir issue.
>
> We found a corrupted xfs volume in one of our systems. It is around 1 TB size and about 12 M files.
> We run xfs_repair on the volume which succeeded after 42 minutes.
> We noticed that memory consumption raised to about 7.5 GB.
> Since some customers are using only 4GB (and sometimes even 2 GB) we tried running "xfs_repair -m 3200" on a 4GB RAM machine.
> However, this time an OOM event happened during handling of AG 26 during step 3.
> The log of xfs_repair is enclosed below.
> We will appreciate your feedback on the amount of memory needed for xfs_repair in general and when using "-m" option specifically.
> The xfs metadata dump (prior to xfs_repair) can be found here:
> https://zadarastorage-public.s3.amazonaws.com/xfs/xfsdump-prod-ebs_2015-03-30_23-00-38.tgz
> It is a 1.2 GB file (and 5.7 GB uncompressed).
>
> We will appreciate your feedback on the corruption pattern as well.
> --
> Thank you,
> Danny Shavit
> Zadarastorage
>
> ---------- xfs_repair log ----------------

Just a note ...

> bad . entry in directory inode 5691013154, was 5691013170: correcting

101010011001101011111100000100100
101010011001101011111100000110100
              ^ bit flip

> bad . entry in directory inode 5691013156, was 5691013172: correcting

101010011001101011111100000100100
101010011001101011111100000110100
              ^ bit flip

etc ...

> bad . entry in directory inode 5691013157, was 5691013173: correcting
> bad . entry in directory inode 5691013163, was 5691013179: correcting

--

Regards,

Danny