[OpenAFS-devel] Re: The ihandle sync thing

[Brandon Allbery]

There's some kind of background sync in almost every OS, certainly in all t=
he Unix-like OSes OpenAFS runs on. But it's stupid and not tunable to be ap=
propriate for things like database servers or file servers (by necessity; t=
he service itself knows best what sync strategy is appropriate); it's best =
considered a last resort.=0A=
=0A=
--=0A=
brandon s allbery kf8nh                               sine nomine associate=
s=0A=
allbery.b@gmail.com                                  ballbery@sinenomine.ne=
t=0A=
unix, openafs, kerberos, infrastructure, xmonad        http://sinenomine.ne=
t=0A=
=0A=
________________________________________=0A=
From: openafs-devel-admin@openafs.org [openafs-devel-admin@openafs.org] on =
behalf of chas williams - CONTRACTOR [chas@cmf.nrl.navy.mil]=0A=
Sent: Thursday, March 28, 2013 11:46=0A=
To: Andrew Deason=0A=
Cc: openafs-devel@openafs.org=0A=
Subject: Re: [OpenAFS-devel] Re: The ihandle sync thing=0A=
=0A=
On Thu, 28 Mar 2013 10:14:09 -0500=0A=
Andrew Deason <adeason@sinenomine.net> wrote:=0A=
=0A=
> On Thu, 28 Mar 2013 08:04:45 -0400=0A=
> chas williams - CONTRACTOR <chas@cmf.nrl.navy.mil> wrote:=0A=
>=0A=
> > On Wed, 27 Mar 2013 17:19:07 -0400=0A=
> > Chaskiel Grundman <cg2v@andrew.cmu.edu> wrote:=0A=
> >=0A=
> > > Just because ext3 (in ordered mode) does something doesn't mean it's=
=0A=
> > > correct or something you should count on. If you don't=0A=
> > > fsync/fdatasync, there is no guarantee your data is on the media or=
=0A=
> > > ever will be on the media.=0A=
>=0A=
> Doesn't pdflush guarantee this? That something doesn't stay around in=0A=
> memory forever.=0A=
=0A=
On a modern Linux, yes.  On other operating systems, I don't know.  I=0A=
imagine Solaris has something but off the top of my head I have no idea=0A=
its behavior with regard to ufs and timely flushing of dirty pages.=0A=