Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <29dec29c4a776db82406f272557f97cebabca963.camel@kernel.org>
Subject: Re: [linus:master] [NFSD]  39d432fc76:  fsmark.files_per_sec
 -100.0% regression
From:   Jeff Layton <jlayton@kernel.org>
To:     Benjamin Coddington <bcodding@redhat.com>,
        Chuck Lever <chuck.lever@oracle.com>
Cc:     kernel test robot <oliver.sang@intel.com>, oe-lkp@lists.linux.dev,
        lkp@intel.com, linux-kernel@vger.kernel.org,
        linux-nfs@vger.kernel.org, ying.huang@intel.com,
        feng.tang@intel.com, fengwei.yin@intel.com
Date:   Thu, 24 Aug 2023 10:26:15 -0400
In-Reply-To: <D38094B9-CE6F-4BE1-9250-22073A1DD86E@redhat.com>
References: <202308241229.68396422-oliver.sang@intel.com>
         <ZOdZ7DiQ4jqhoj0i@tissot.1015granger.net>
         <D38094B9-CE6F-4BE1-9250-22073A1DD86E@redhat.com>
Content-Type: text/plain; charset="ISO-8859-15"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.48.4 (3.48.4-1.fc38) 
MIME-Version: 1.0
Precedence: bulk

On Thu, 2023-08-24 at 10:12 -0400, Benjamin Coddington wrote:
> On 24 Aug 2023, at 9:23, Chuck Lever wrote:
>=20
> > On Thu, Aug 24, 2023 at 01:59:06PM +0800, kernel test robot wrote:
> > >=20
> > >=20
> > > hi, Chuck Lever,
> > >=20
> > > Fengwei (CCed) helped us review this astonishing finding by fsmark te=
sts,
> > > and doubt below part:
> > > -			nfsd4_end_grace(nn);
> > > +			trace_nfsd_end_grace(netns(file));
> > >=20
> > > and confirmed if adding back:
> > > 			nfsd4_end_grace(nn);
> > >=20
> > > the regression is gone and files_per_sec restore to 61.93.
> >=20
> > As always, thanks for the report. However, this result is not
> > plausible. "end_grace" happens only once after a server reboot.
> >=20
> > Can you confirm that the NFS server kernel is not crashing
> > during the test?
>=20
> Does the removal of nfsd4_end_grace() here disable the ability of nfsdclt=
rack to
> terminate the grace period early on a first start of the server?
>=20
>=20

Yes. That should be the only effect (which is why we didn't notice it).
--=20
Jeff Layton <jlayton@kernel.org>