On Sat, Aug 28, 2021 at 08:03:43AM +1000, Dave Chinner wrote:

On Fri, Aug 27, 2021 at 04:53:47PM +1000, Chris Dunlop wrote:

quoted

On 8/25/21 9:06 PM, Chris Dunlop wrote:

quoted

Background: I'm chasing a mysterious ENOSPC error on an XFS
filesystem with way more space than the app should be asking
for. There are no quotas on the fs. Unfortunately it's a third
party app and I can't tell what sequence is producing the error,
but this fallocate issue is a possibility.

Oh, another reference: this is extensive reflinking happening on this
filesystem.

Ah. Details that are likely extremely important. The workload,
layout problems and ephemeral ENOSPC symptoms match the description
of the problem that was fixed by the series of commits that went
into 5.13 that ended in this one:

commit fd43cf600cf61c66ae0a1021aca2f636115c7fcb
Author: Brian Foster [off-list ref]
Date:   Wed Apr 28 15:06:05 2021 -0700

   xfs: set aside allocation btree blocks from block reservation

Oh wow. Yes, sounds like a candidate. Is there same easy(-ish?) way of 
seeing if this fs is likely to be suffering from this particular issue or 
is it a matter of installing an appropriate kernel and seeing if the 
problem goes away?

The job getting this ENOSPC error is one of 45 similar jobs, and it's the 
only one getting the error. There doesn't seem to be anything special 
about this job, it's main file where the writes are going is the 9th 
largest (up to 1.8T), and it has a lot of extents (842G split into 750M 
extents) but not as many as some others (e.g. 809G split into 1G extents).  
That said, the app works in mysterious ways so this particular job may be 
a special snowflake in some unobvious manner.

Cheers,

Chris