RE: [RFC PATCH 3/4] ceph: introduce ceph_submit_write() method

[Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>]

On Wed, 2025-08-27 at 00:53 +0200, Max Kellermann wrote:
> On 2025/08/27 00:33, Viacheslav Dubeyko <Slava.Dubeyko@ibm.com> wrote:
> > Of course, we can revert any patch. This patchset has been sent not with the
> > goal of pure refactoring but it fixes several bugs. Reverting means returning
> > these bugs back.
> 
> You should have listened of Matthew and submit separate minimal
> bug-fixing patches instead of posting huge patches which move code
> around, change semantics and hidden somewhere deep within fix some bug
> (and then introduce new bugs).
> 
> > This patchset was available for review for a long time.
> 
> There was exactly one review, and no, you were not "happy to rework
> and to make any patch more better" - you openly rejected Matthew's
> review.
> 
> > From my point of view, reverting is not answer and it makes sense to
> > continue fix bugs and to make CephFS code more stable.
> 
> Your argument only appears to sound right, but it is detached from the
> reality I'm living in.
> 
> Your patches made Ceph less stable.  6.14 had one Ceph-related crash
> every other week, but 6.15 with your patches made all servers crash
> within hours.
> 
> The point is: the Linux kernel was better without your patches.  Your
> patches may have fixed a bug, but have introduced a dozen new bugs,
> including one that very quickly crashes the whole kernel, one that was
> really obvious enough, just nobody cared enough to read deeply enough
> after you rejected Matthew's review.  Too bad no maintainer stopped
> you!
> 
> Of course, the bug that was fixed by your patch set should be fixed -
> but not the way you did it.  Every aspect of your approach to fixing
> the bug was bad.
> 
> The best way forward for you would be to revert this patch set and
> write a minimal patch that only fixes the bug.  If you want to be
> helpful here, please give this a try.
> 
> 

This patchset had been tested by xfstests and no issue had been triggered. We
have not enough Ceph related test-cases. So, you are welcome to add a test-case
for the issue.

This issue had been reported more than a month ago. I tried to reproduce it but
I had no luck. So, if you are lucky enough, then simply share the patch or the
way to reproduce the issue. The patchset simply placed old code into dedicated
functions with the goal to manage complexity. So, the old code is still there
and the patch cannot introduce "dozen new bugs". Otherwise, xfstests can easily
reproduce these gazilion of bugs. :) Currently, we have only one issue reported.

The open-source is space for collaboration and respect of each other. It is not
space for offensive or bullying behavior.

Thanks,
Slava.