Re: [PATCH V2] Btrfs: keep dropped roots in cache until transaction commit

["Holger Hoffstätte" <holger.hoffstaette@googlemail.com>]

On 09/16/15 15:50, Josef Bacik wrote:
> On 09/16/2015 04:58 AM, Holger Hoffstätte wrote:
>> On 09/15/15 21:15, Josef Bacik wrote:
>>> On 09/15/2015 03:08 PM, Holger Hoffstätte wrote:
>>>> On 09/15/15 17:50, Holger Hoffstätte wrote:
>>>>> This V2 does indeed seem to fix the issues I reported with snapshot
>>>>> deletion & concurrent sync. I've now created/filled/deleted countless
>>>>> snapshots while issuing sync(s) in parallel, and the problem that I
>>>>> saw fairly frequently with V1 no longer seems to occur here.
>>>>
>>>> Well..I may have spoken too soon:
>>>
>>> Huh this doesn't seem related to my stuff.  Can I have your script
>>> you are running to reproduce this?  Thanks,
>>>
>>> Josef
>>
>> It seemed to be related to Omar's free-space-tree - I saw the same bug
>> a bit later on a different device without snapshot deletion, just
>> deleting two large files. Sorry for the false positive.
>>
> 
> Omar, do you have time to look into this? Were they highly fragmented
> files or just giant contiguous files? Thanks,

These were very contiguous (2* ~1.3 GB, 1-3 extents each). I already sent
Omar a long klog dump with more background info. The biggest issue is that
after this bug the volume is hosed since the fst sets an incompat flag,
and after mounting rw just stumbles again trying to replay.
I'll keep trying to isolate exactly how and when this happens.

-h