Re: [PATCH] Cleanup old XFS_BTREE_* traces

[Carlos Maiolino <cmaiolino@redhat.com>]

On Wed, Feb 14, 2018 at 10:58:14AM +1100, Dave Chinner wrote:
> On Tue, Feb 13, 2018 at 03:17:31PM -0800, Darrick J. Wong wrote:
> > On Tue, Feb 13, 2018 at 08:21:06AM +1100, Dave Chinner wrote:
> > > On Mon, Feb 12, 2018 at 10:29:48AM -0800, Darrick J. Wong wrote:
> > > > On Mon, Feb 12, 2018 at 02:00:05PM +0100, Carlos Maiolino wrote:
> > > > > Remove unused legacy btree traces from IRIX era.
> > > > > 
> > > > > Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
> > > > > ---
> > > > > 
> > > > > Talking to Dave about it, he mentioned XFS_BTREE_TRACE_CURSOR might be worth to
> > > > > turn into a proper ftrace trace point, so I didn't touch _CURSOR traces in this
> > > > > patchset, and a proper conversion will be sent later, unless it's not worth at
> > > > > all, and I should send a V2 also removing TRACE_CURSOR.
> > > > 
> > > > TBH I wonder the opposite -- why not turn all of these into tracepoints?
> > > 
> > > TBH, we haven't used them in at least 15 years. What value do they
> > > provide apart from making the traces even noisier (and potentially
> > > more lossy) than they already are?
> > 
> > FWIW adding trace_printks to some of those functions was rather useful
> > for checking that the unusual refcount and rmap btree semantics actually
> > resulted in calls to the desired btree functions.  I wish I'd cleaned up
> > that debugging patch and sent it, but it's lost now.
> 

I see your point, although, for me, it sounds like a very specific debug case,
and not something which would add enough benefit to have these extra debug
traces lying around.

I mean, if we keep adding trace points for many single debug use-cases, we would
end up with tons of trace points in xfs, which are not used in 99% of the debugging
processes.

But well, I don't have a decent knowledge in the Btrees infra-structure yet to
give a more detailed reason if such extra trace points would be useful or not in
several situations. So, this is just my $0.02, based on the amount of trace
points we already have, where, most of time I use one or another, and yet, I
need to add my own trace_printks, so, particularly, I don't think adding such
extra traces in Btree code would be that useful in a long-term period, but well,
as I said, just my $0.02.

> Ok, so the non-cursor traces would have been useful to you.
> That's fine - I just don't want to add stuff that doesn't have any
> specific use because it's already hard enough to filter traces down
> to just the things we need to see....
> 

> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com

-- 
Carlos