Re: Poll: time to switch skinny-metadata on by default?

[Josef Bacik <jbacik@fb.com>]

On 10/18/2014 07:21 AM, Petr Janecek wrote:
> Hello,
>
>>>    so far I haven't succeeded running btrfs balance on a large
>>> skinny-metadata fs -- segfault, kernel bug, reproducible.  No such
>>> problems on ^skinny-metadata fs (same disks, same data).  Tried both
>>> several times on 3.17.  More info in comments 10,14 in
>>> https://bugzilla.kernel.org/show_bug.cgi?id=64961
>>
>> I can't reproduce this, how big is your home directory, and are you
>> still seeing corruptions after just rsyncing to a clean fs?  Thanks,
>
>    as I wrote in comment 10, it has improved since year ago when I
> reported it: I see no corruption at all, neither after rsync, nor after
> balance crash: btrfs check doesn't find anything wrong, files look ok.
> The only problem is that after adding a disk the balance segfaults on a
> kernel bug and the fs gets stuck.  When I run balance again after
> reboot, it makes only a very small progress and crashes again the same
> way.
>
>    There are some 2.5TB of data in 7.5M files on that fs.  And couple
> dozen ro snapshots -- I'm testing 3.17 + revert of 9c3b306e1c9e right
> now, but it takes more than day to copy the data and recreate all the
> snapshots.  But a test with ^skinny-metadata showed no problems, so I
> don't thing I got bitten by that bug.
>
>    I have btrfs-image of one of previous runs after crashed balance.
> It's 15GB. I can place it somewhere with fast link, are you interested?
>
>

Yup, send me the link and I'll pull it down.  Thanks,

Josef