Re: [PATCH 2/2] xfs: Properly retry failed inode items in case of error during buffer writeback

["Luis R. Rodriguez" <mcgrof@kernel.org>]

On Wed, Jun 21, 2017 at 06:10:52AM -0400, Brian Foster wrote:
> On Tue, Jun 20, 2017 at 08:05:05PM +0200, Luis R. Rodriguez wrote:
> > On Tue, Jun 20, 2017 at 01:20:41PM -0400, Brian Foster wrote:
> > > On Tue, Jun 20, 2017 at 06:52:04PM +0200, Luis R. Rodriguez wrote:
> > > > On Mon, Jun 19, 2017 at 06:59:05AM -0400, Brian Foster wrote:
> > > > > It hasn't seemed necessary to me to take that approach given the lower
> > > > > prevalence of the issue 
> > > > 
> > > > Of this issue? I suppose its why I asked about examples of issues, I seem
> > > > to have found it likely much more possible out in the wild than expected.
> > > > It would seem folks might be working around it somehow.
> > > > 
> > > 
> > > If we're talking about the thin provisioning case, I suspect most people
> > > work around it by properly configuring their storage. ;)
> > 
> > The fact that we *hang* makes it more serious, so even if folks misconfigured
> > storage with less space it should be no reason to consider hangs any less
> > severe. Specially if it seems to be a common issue, and I'm alluding to the
> > fact that this might be more common than the patch describes.
> > 
> 
> My point is simply that a hang was a likely outcome before the patch
> that introduced the regression as well, so the benefit of doing a proper
> revert doesn't clearly outweigh the cost.

Sure agreed.

> Despite what the side effect
> is, the fact that this tends to primarily affect users who have thin
> volumes going inactive also suggests that this can be worked around
> reasonably well enough via storage configuration. This all suggests to
> me that Carlos' current approach is the most reasonable one. 

OK thanks.

> I'm not following what the line of argument is here. Are you suggesting
> a different approach? If so, based on what use case/reasoning?

No, it just seemed to me you were indicating that the hang was not that serious
of an issue given you could work around it with proper storage configuration.
I see now you were using that analogy just to indicate it was also an issue
before so the revert is with merit.

  Luis