Is there some way (mount option?) I can ask the client to attempt a 4.1
(sessions) mount but do not use pnfs?
On Tue, 2011-01-18 at 12:44 -0500, [email protected] wrote:
>
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]] On Behalf Of Trond Myklebust
> > Sent: Friday, January 14, 2011 7:42 AM
> > To: Jim Rees
> > Cc: William A. (Andy) Adamson; [email protected]
> > Subject: Re: 4.1 no-pnfs mount option?
> >
> > On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
> > > William A. (Andy) Adamson wrote:
> > >
> > > No mount option - just configure your machine to not load any pnfs
> > > layoutdriver modules.
> > >
> > > That works, thanks, but I was hoping for a way to have
> > layout and non-layout
> > > mounts going on at the same time.
> >
> > Different VMs?
> >
> > Trond
>
> Would there be any objections to adding a "nopnfs" mount option to force this behavior? It could be useful at least for testing, possibly for working around server problems, or if an admin knows that certain clients' usage patterns would be better served by disabling pnfs.
Yes. Why should we be adding debugging mount options to the upstream
code? Just test the damned pnfs code properly before it goes upstream...
Trond
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
>
> > -----Original Message-----
> > From: Trond Myklebust [mailto:[email protected]]
> > Sent: Tuesday, January 18, 2011 11:15 AM
> > To: Matt W. Benjamin
> > Cc: Muntz, Daniel; [email protected]; [email protected];
> > [email protected]; Benny Halevy
> > Subject: Re: 4.1 no-pnfs mount option?
> >
> > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > Hi,
> > >
> > > Isn't by mount a plausible way to select for pnfs
> > independent of debugging? Is it assured that a client
> > administrator would never reasonably wish to do this?
> >
> > "Why would an administrator never want to do this?" is not a helpful
> > question.
> >
> > A more useful question is "what reason would you possibly have for
> > overriding the server's request that you do pNFS when your client has
> > pNFS support?" What makes pNFS so special that we must allow
> > administrators to do this on a per-mount basis?
>
> By the same logic, why should a user be allowed to select which version of NFS they use for mounting when the server has a perfectly reasonable way of negotiating it? Getting to choose v2 vs. v3 vs. v4 seems like much less of a distinction than choosing between pNFS and no pNFS. Frankly, it never even occurred to me that there wouldn't be a mount option to make this choice. Enabling/disabling the layout driver doesn't fit the existing model of choosing mount behavior, and is a big hammer--it's all or nothing.
>
> Anyway, here's a use case: I'm working at an HPC/gas+oil/satellite data site. We have an awesome pNFS server for our big data and I want to access my big data with pNFS. We have another server for homedirs, some big data, and other stuff. Some mounts are fine with pNFS, others are abysmal. So, I want to mount some directories with pNFS, and some without pNFS, on the same client, independent of the server configuration.
mount -t nfs -overs=4,minorversion=0 foo:/ /bar
Done... Any more questions?
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
On 2011-01-18 20:38, Trond Myklebust wrote:
> On Tue, 2011-01-18 at 20:35 +0200, Benny Halevy wrote:
>> On 2011-01-18 20:28, Trond Myklebust wrote:
>>> On Tue, 2011-01-18 at 12:44 -0500, [email protected] wrote:
>>>>
>>>>> -----Original Message-----
>>>>> From: [email protected]
>>>>> [mailto:[email protected]] On Behalf Of Trond Myklebust
>>>>> Sent: Friday, January 14, 2011 7:42 AM
>>>>> To: Jim Rees
>>>>> Cc: William A. (Andy) Adamson; [email protected]
>>>>> Subject: Re: 4.1 no-pnfs mount option?
>>>>>
>>>>> On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
>>>>>> William A. (Andy) Adamson wrote:
>>>>>>
>>>>>> No mount option - just configure your machine to not load any pnfs
>>>>>> layoutdriver modules.
>>>>>>
>>>>>> That works, thanks, but I was hoping for a way to have
>>>>> layout and non-layout
>>>>>> mounts going on at the same time.
>>>>>
>>>>> Different VMs?
>>>>>
>>>>> Trond
>>>>
>>>> Would there be any objections to adding a "nopnfs" mount option to force this behavior? It could be useful at least for testing, possibly for working around server problems, or if an admin knows that certain clients' usage patterns would be better served by disabling pnfs.
>>>
>>> Yes. Why should we be adding debugging mount options to the upstream
>>> code? Just test the damned pnfs code properly before it goes upstream...
>>>
>>> Trond
>>
>> Such a mount option could be useful for dealing with buggy servers (pnfs-wise) so you
>> could mount one server with pnfs and another without.
>
> You can find ways around that. Just use 2 clients: one with pnfs
> switched on, and one with it off.
>
> I really don't want to introduce mount options upstream unless they are
> useful in the long term. One off usefulness does not pass that test.
>
I guess we can live with that for the short term.
On Tue, 2011-01-18 at 22:30 -0500, Matt W. Benjamin wrote:
> The server has indicated that it supports pnfs. Why is the client obligated to request it?
It is not required by the protocol, but the client had better have a
damned good reason for not doing so. As stated several times previously,
pNFS is all about increasing _server_ scalability. If the server doesn't
need to scale out, why would it tell us to use pNFS for those files in
the first place?
Trond
> ----- "Trond Myklebust" <[email protected]> wrote:
>
> > On Tue, 2011-01-18 at 21:29 -0500, [email protected] wrote:
> > >
> > > > -----Original Message-----
> > > > From: Trond Myklebust [mailto:[email protected]]
> > > > Sent: Tuesday, January 18, 2011 5:45 PM
> > > > To: Muntz, Daniel
> > > > Cc: [email protected]; [email protected];
> > > > [email protected]; [email protected];
> > > > [email protected]
> > > > Subject: RE: 4.1 no-pnfs mount option?
> > > >
> > > > On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Trond Myklebust [mailto:[email protected]]
> > > > > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > > > > To: Matt W. Benjamin
> > > > > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > > > > [email protected]; Benny Halevy
> > > > > > Subject: Re: 4.1 no-pnfs mount option?
> > > > > >
> > > > > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > Isn't by mount a plausible way to select for pnfs
> > > > > > independent of debugging? Is it assured that a client
> > > > > > administrator would never reasonably wish to do this?
> > > > > >
> > > > > > "Why would an administrator never want to do this?" is
> > > > not a helpful
> > > > > > question.
> > > > > >
> > > > > > A more useful question is "what reason would you possibly have
> > for
> > > > > > overriding the server's request that you do pNFS when
> > > > your client has
> > > > > > pNFS support?" What makes pNFS so special that we must allow
> > > > > > administrators to do this on a per-mount basis?
> > > > >
> > > > > By the same logic, why should a user be allowed to select
> > > > which version of NFS they use for mounting when the server
> > > > has a perfectly reasonable way of negotiating it? Getting to
> > > > choose v2 vs. v3 vs. v4 seems like much less of a distinction
> > > > than choosing between pNFS and no pNFS. Frankly, it never
> > > > even occurred to me that there wouldn't be a mount option to
> > > > make this choice. Enabling/disabling the layout driver
> > > > doesn't fit the existing model of choosing mount behavior,
> > > > and is a big hammer--it's all or nothing.
> > > > >
> > > > > Anyway, here's a use case: I'm working at an
> > > > HPC/gas+oil/satellite data site. We have an awesome pNFS
> > > > server for our big data and I want to access my big data with
> > > > pNFS. We have another server for homedirs, some big data,
> > > > and other stuff. Some mounts are fine with pNFS, others are
> > > > abysmal. So, I want to mount some directories with pNFS, and
> > > > some without pNFS, on the same client, independent of the
> > > > server configuration.
> > > >
> > > > mount -t nfs -overs=4,minorversion=0 foo:/ /bar
> > > >
> > > > Done... Any more questions?
> > >
> > > Several, but I'll stick to one rhetorical. Does NFSv4.1 have any
> > features, other than pNFS, that are not in 4.0?
> >
> > Why stop now, when you were batting 100? I told you what the criteria
> > were for adding more mount options, and you start whining about not
> > being able to conceive of a world without mount options.
> >
> > The point is that NFSv4.1 was supposed to let the _server_ tell the
> > client when to use pNFS. The reason why you let the _server_ do this,
> > is
> > because pNFS is about enabling _server_ scalability. It is not about
> > faster clients...
> > If you don't want the client to use pNFS, then fix the _server_
> > settings...
> >
> > --
> > Trond Myklebust
> > Linux NFS client maintainer
> >
> > NetApp
> > [email protected]
> > http://www.netapp.com
>
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
The server has indicated that it supports pnfs. Why is the client obligated to request it?
Matt
----- "Trond Myklebust" <[email protected]> wrote:
> On Tue, 2011-01-18 at 21:29 -0500, [email protected] wrote:
> >
> > > -----Original Message-----
> > > From: Trond Myklebust [mailto:[email protected]]
> > > Sent: Tuesday, January 18, 2011 5:45 PM
> > > To: Muntz, Daniel
> > > Cc: [email protected]; [email protected];
> > > [email protected]; [email protected];
> > > [email protected]
> > > Subject: RE: 4.1 no-pnfs mount option?
> > >
> > > On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> > > >
> > > > > -----Original Message-----
> > > > > From: Trond Myklebust [mailto:[email protected]]
> > > > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > > > To: Matt W. Benjamin
> > > > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > > > [email protected]; Benny Halevy
> > > > > Subject: Re: 4.1 no-pnfs mount option?
> > > > >
> > > > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > > > Hi,
> > > > > >
> > > > > > Isn't by mount a plausible way to select for pnfs
> > > > > independent of debugging? Is it assured that a client
> > > > > administrator would never reasonably wish to do this?
> > > > >
> > > > > "Why would an administrator never want to do this?" is
> > > not a helpful
> > > > > question.
> > > > >
> > > > > A more useful question is "what reason would you possibly have
> for
> > > > > overriding the server's request that you do pNFS when
> > > your client has
> > > > > pNFS support?" What makes pNFS so special that we must allow
> > > > > administrators to do this on a per-mount basis?
> > > >
> > > > By the same logic, why should a user be allowed to select
> > > which version of NFS they use for mounting when the server
> > > has a perfectly reasonable way of negotiating it? Getting to
> > > choose v2 vs. v3 vs. v4 seems like much less of a distinction
> > > than choosing between pNFS and no pNFS. Frankly, it never
> > > even occurred to me that there wouldn't be a mount option to
> > > make this choice. Enabling/disabling the layout driver
> > > doesn't fit the existing model of choosing mount behavior,
> > > and is a big hammer--it's all or nothing.
> > > >
> > > > Anyway, here's a use case: I'm working at an
> > > HPC/gas+oil/satellite data site. We have an awesome pNFS
> > > server for our big data and I want to access my big data with
> > > pNFS. We have another server for homedirs, some big data,
> > > and other stuff. Some mounts are fine with pNFS, others are
> > > abysmal. So, I want to mount some directories with pNFS, and
> > > some without pNFS, on the same client, independent of the
> > > server configuration.
> > >
> > > mount -t nfs -overs=4,minorversion=0 foo:/ /bar
> > >
> > > Done... Any more questions?
> >
> > Several, but I'll stick to one rhetorical. Does NFSv4.1 have any
> features, other than pNFS, that are not in 4.0?
>
> Why stop now, when you were batting 100? I told you what the criteria
> were for adding more mount options, and you start whining about not
> being able to conceive of a world without mount options.
>
> The point is that NFSv4.1 was supposed to let the _server_ tell the
> client when to use pNFS. The reason why you let the _server_ do this,
> is
> because pNFS is about enabling _server_ scalability. It is not about
> faster clients...
> If you don't want the client to use pNFS, then fix the _server_
> settings...
>
> --
> Trond Myklebust
> Linux NFS client maintainer
>
> NetApp
> [email protected]
> http://www.netapp.com
--
Matt Benjamin
The Linux Box
206 South Fifth Ave. Suite 150
Ann Arbor, MI 48104
http://linuxbox.com
tel. 734-761-4689
fax. 734-769-8938
cel. 734-216-5309
On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
> William A. (Andy) Adamson wrote:
>
> No mount option - just configure your machine to not load any pnfs
> layoutdriver modules.
>
> That works, thanks, but I was hoping for a way to have layout and non-layout
> mounts going on at the same time.
Different VMs?
Trond
> -----Original Message-----
> From: Trond Myklebust [mailto:[email protected]]
> Sent: Tuesday, January 18, 2011 11:15 AM
> To: Matt W. Benjamin
> Cc: Muntz, Daniel; [email protected]; [email protected];
> [email protected]; Benny Halevy
> Subject: Re: 4.1 no-pnfs mount option?
>
> On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > Hi,
> >
> > Isn't by mount a plausible way to select for pnfs
> independent of debugging? Is it assured that a client
> administrator would never reasonably wish to do this?
>
> "Why would an administrator never want to do this?" is not a helpful
> question.
>
> A more useful question is "what reason would you possibly have for
> overriding the server's request that you do pNFS when your client has
> pNFS support?" What makes pNFS so special that we must allow
> administrators to do this on a per-mount basis?
By the same logic, why should a user be allowed to select which version of NFS they use for mounting when the server has a perfectly reasonable way of negotiating it? Getting to choose v2 vs. v3 vs. v4 seems like much less of a distinction than choosing between pNFS and no pNFS. Frankly, it never even occurred to me that there wouldn't be a mount option to make this choice. Enabling/disabling the layout driver doesn't fit the existing model of choosing mount behavior, and is a big hammer--it's all or nothing.
Anyway, here's a use case: I'm working at an HPC/gas+oil/satellite data site. We have an awesome pNFS server for our big data and I want to access my big data with pNFS. We have another server for homedirs, some big data, and other stuff. Some mounts are fine with pNFS, others are abysmal. So, I want to mount some directories with pNFS, and some without pNFS, on the same client, independent of the server configuration.
>
> Throwing more and more knobs into the kernel is easy. The
> difficult bit
> is to figure out which are useful knobs, and that is why I
> want real use
> cases...
>
> Trond
>
>
> > ----- "Trond Myklebust" <[email protected]> wrote:
> >
> > > >
> > > > Such a mount option could be useful for dealing with
> buggy servers
> > > (pnfs-wise) so you
> > > > could mount one server with pnfs and another without.
> > >
> > > You can find ways around that. Just use 2 clients: one with pnfs
> > > switched on, and one with it off.
> > >
> > > I really don't want to introduce mount options upstream
> unless they
> > > are
> > > useful in the long term. One off usefulness does not pass
> that test.
> > >
> > > --
> > > Trond Myklebust
> > > Linux NFS client maintainer
> > >
> > > NetApp
> > > [email protected]
> > > http://www.netapp.com
> > >
> > > --
> > > To unsubscribe from this list: send the line "unsubscribe
> linux-nfs"
> > > in
> > > the body of a message to [email protected]
> > > More majordomo info at http://vger.kernel.org/majordomo-info.html
> >
>
> --
> Trond Myklebust
> Linux NFS client maintainer
>
> NetApp
> [email protected]
> http://www.netapp.com
>
>
>
> -----Original Message-----
> From: Trond Myklebust [mailto:[email protected]]
> Sent: Tuesday, January 18, 2011 6:57 PM
> To: Muntz, Daniel
> Cc: [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]
> Subject: RE: 4.1 no-pnfs mount option?
>
> On Tue, 2011-01-18 at 21:29 -0500, [email protected] wrote:
> >
> > > -----Original Message-----
> > > From: Trond Myklebust [mailto:[email protected]]
> > > Sent: Tuesday, January 18, 2011 5:45 PM
> > > To: Muntz, Daniel
> > > Cc: [email protected]; [email protected];
> > > [email protected]; [email protected];
> > > [email protected]
> > > Subject: RE: 4.1 no-pnfs mount option?
> > >
> > > On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> > > >
> > > > > -----Original Message-----
> > > > > From: Trond Myklebust [mailto:[email protected]]
> > > > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > > > To: Matt W. Benjamin
> > > > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > > > [email protected]; Benny Halevy
> > > > > Subject: Re: 4.1 no-pnfs mount option?
> > > > >
> > > > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > > > Hi,
> > > > > >
> > > > > > Isn't by mount a plausible way to select for pnfs
> > > > > independent of debugging? Is it assured that a client
> > > > > administrator would never reasonably wish to do this?
> > > > >
> > > > > "Why would an administrator never want to do this?" is
> > > not a helpful
> > > > > question.
> > > > >
> > > > > A more useful question is "what reason would you
> possibly have for
> > > > > overriding the server's request that you do pNFS when
> > > your client has
> > > > > pNFS support?" What makes pNFS so special that we must allow
> > > > > administrators to do this on a per-mount basis?
> > > >
> > > > By the same logic, why should a user be allowed to select
> > > which version of NFS they use for mounting when the server
> > > has a perfectly reasonable way of negotiating it? Getting to
> > > choose v2 vs. v3 vs. v4 seems like much less of a distinction
> > > than choosing between pNFS and no pNFS. Frankly, it never
> > > even occurred to me that there wouldn't be a mount option to
> > > make this choice. Enabling/disabling the layout driver
> > > doesn't fit the existing model of choosing mount behavior,
> > > and is a big hammer--it's all or nothing.
> > > >
> > > > Anyway, here's a use case: I'm working at an
> > > HPC/gas+oil/satellite data site. We have an awesome pNFS
> > > server for our big data and I want to access my big data with
> > > pNFS. We have another server for homedirs, some big data,
> > > and other stuff. Some mounts are fine with pNFS, others are
> > > abysmal. So, I want to mount some directories with pNFS, and
> > > some without pNFS, on the same client, independent of the
> > > server configuration.
> > >
> > > mount -t nfs -overs=4,minorversion=0 foo:/ /bar
> > >
> > > Done... Any more questions?
> >
> > Several, but I'll stick to one rhetorical. Does NFSv4.1
> have any features, other than pNFS, that are not in 4.0?
>
> Why stop now, when you were batting 100? I told you what the criteria
> were for adding more mount options, and you start whining about not
> being able to conceive of a world without mount options.
>
> The point is that NFSv4.1 was supposed to let the _server_ tell the
> client when to use pNFS. The reason why you let the _server_
> do this, is
> because pNFS is about enabling _server_ scalability. It is not about
> faster clients...
> If you don't want the client to use pNFS, then fix the _server_
> settings...
I don't miss the kiss-up-kick-down model of software development...
The criteria was a use case, so I gave a use case. I wouldn't rule out the "testing use case" either, as a mount option makes it possible to test mixed pnfs/non-pnfs 4.1 traffic without multiple servers.
The point is that pNFS, like everything else, is about the users.
-Dan
>
> --
> Trond Myklebust
> Linux NFS client maintainer
>
> NetApp
> [email protected]
> http://www.netapp.com
>
>
>
On Tue, 2011-01-18 at 14:35 -0500, Matt W. Benjamin wrote:
> Hi,
>
> ----- "Trond Myklebust" <[email protected]> wrote:
>
> > "Why would an administrator never want to do this?" is not a helpful
> > question.
> >
> > A more useful question is "what reason would you possibly have for
> > overriding the server's request that you do pNFS when your client has
> > pNFS support?" What makes pNFS so special that we must allow
> > administrators to do this on a per-mount basis?
>
> Well, I phrased my question the other way because I suspect such cases will be found, but I may not have found all of them.
>
> Some thoughts on why I might wish to take a hand in the decision:
>
> 1. the client doing pnfs might behave badly due to a misconfiguration or outage, yet behave acceptably using ordinary nfsv4?
The client should be automatically falling back to non-pNFS mode when
this sort of thing happens (and, yes, we are already testing that kind
of scenario).
In what cases do we expect the administrator to be able to detect that
the client is misbehaving without the client itself being able to detect
it and thus take action?
> 2. restricting the client to ordinary nfsv4 might be desirable for non-developer troubleshooting or other configuration work?
That is the one-off testing case. You already have the option of
removing and disabling the pNFS module to do your testing.
> I apologize if neither is compelling.
>
> >
> > Throwing more and more knobs into the kernel is easy. The difficult
> > bit
> > is to figure out which are useful knobs, and that is why I want real
> > use
> > cases...
> >
> > Trond
> >
>
> Matt
>
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
> -----Original Message-----
> From: Trond Myklebust [mailto:[email protected]]
> Sent: Tuesday, January 18, 2011 5:45 PM
> To: Muntz, Daniel
> Cc: [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]
> Subject: RE: 4.1 no-pnfs mount option?
>
> On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> >
> > > -----Original Message-----
> > > From: Trond Myklebust [mailto:[email protected]]
> > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > To: Matt W. Benjamin
> > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > [email protected]; Benny Halevy
> > > Subject: Re: 4.1 no-pnfs mount option?
> > >
> > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > Hi,
> > > >
> > > > Isn't by mount a plausible way to select for pnfs
> > > independent of debugging? Is it assured that a client
> > > administrator would never reasonably wish to do this?
> > >
> > > "Why would an administrator never want to do this?" is
> not a helpful
> > > question.
> > >
> > > A more useful question is "what reason would you possibly have for
> > > overriding the server's request that you do pNFS when
> your client has
> > > pNFS support?" What makes pNFS so special that we must allow
> > > administrators to do this on a per-mount basis?
> >
> > By the same logic, why should a user be allowed to select
> which version of NFS they use for mounting when the server
> has a perfectly reasonable way of negotiating it? Getting to
> choose v2 vs. v3 vs. v4 seems like much less of a distinction
> than choosing between pNFS and no pNFS. Frankly, it never
> even occurred to me that there wouldn't be a mount option to
> make this choice. Enabling/disabling the layout driver
> doesn't fit the existing model of choosing mount behavior,
> and is a big hammer--it's all or nothing.
> >
> > Anyway, here's a use case: I'm working at an
> HPC/gas+oil/satellite data site. We have an awesome pNFS
> server for our big data and I want to access my big data with
> pNFS. We have another server for homedirs, some big data,
> and other stuff. Some mounts are fine with pNFS, others are
> abysmal. So, I want to mount some directories with pNFS, and
> some without pNFS, on the same client, independent of the
> server configuration.
>
> mount -t nfs -overs=4,minorversion=0 foo:/ /bar
>
> Done... Any more questions?
Several, but I'll stick to one rhetorical. Does NFSv4.1 have any features, other than pNFS, that are not in 4.0?
>
> --
> Trond Myklebust
> Linux NFS client maintainer
>
> NetApp
> [email protected]
> http://www.netapp.com
>
>
>
On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> Hi,
>
> Isn't by mount a plausible way to select for pnfs independent of debugging? Is it assured that a client administrator would never reasonably wish to do this?
"Why would an administrator never want to do this?" is not a helpful
question.
A more useful question is "what reason would you possibly have for
overriding the server's request that you do pNFS when your client has
pNFS support?" What makes pNFS so special that we must allow
administrators to do this on a per-mount basis?
Throwing more and more knobs into the kernel is easy. The difficult bit
is to figure out which are useful knobs, and that is why I want real use
cases...
Trond
> ----- "Trond Myklebust" <[email protected]> wrote:
>
> > >
> > > Such a mount option could be useful for dealing with buggy servers
> > (pnfs-wise) so you
> > > could mount one server with pnfs and another without.
> >
> > You can find ways around that. Just use 2 clients: one with pnfs
> > switched on, and one with it off.
> >
> > I really don't want to introduce mount options upstream unless they
> > are
> > useful in the long term. One off usefulness does not pass that test.
> >
> > --
> > Trond Myklebust
> > Linux NFS client maintainer
> >
> > NetApp
> > [email protected]
> > http://www.netapp.com
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-nfs"
> > in
> > the body of a message to [email protected]
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
>
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
On Jan 14, 2011, at 10:38 AM, Jim Rees wrote:
> William A. (Andy) Adamson wrote:
>
> No mount option - just configure your machine to not load any pnfs
> layoutdriver modules.
>
> That works, thanks, but I was hoping for a way to have layout and non-layout
> mounts going on at the same time.
I guess that means two clients.
-->Andy
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
William A. (Andy) Adamson wrote:
No mount option - just configure your machine to not load any pnfs
layoutdriver modules.
That works, thanks, but I was hoping for a way to have layout and non-layout
mounts going on at the same time.
On Wed, 2011-01-19 at 00:54 -0500, [email protected] wrote:
>
> > -----Original Message-----
> > From: Trond Myklebust [mailto:[email protected]]
> > Sent: Tuesday, January 18, 2011 6:57 PM
> > To: Muntz, Daniel
> > Cc: [email protected]; [email protected];
> > [email protected]; [email protected];
> > [email protected]
> > Subject: RE: 4.1 no-pnfs mount option?
> >
> > On Tue, 2011-01-18 at 21:29 -0500, [email protected] wrote:
> > >
> > > > -----Original Message-----
> > > > From: Trond Myklebust [mailto:[email protected]]
> > > > Sent: Tuesday, January 18, 2011 5:45 PM
> > > > To: Muntz, Daniel
> > > > Cc: [email protected]; [email protected];
> > > > [email protected]; [email protected];
> > > > [email protected]
> > > > Subject: RE: 4.1 no-pnfs mount option?
> > > >
> > > > On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Trond Myklebust [mailto:[email protected]]
> > > > > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > > > > To: Matt W. Benjamin
> > > > > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > > > > [email protected]; Benny Halevy
> > > > > > Subject: Re: 4.1 no-pnfs mount option?
> > > > > >
> > > > > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > Isn't by mount a plausible way to select for pnfs
> > > > > > independent of debugging? Is it assured that a client
> > > > > > administrator would never reasonably wish to do this?
> > > > > >
> > > > > > "Why would an administrator never want to do this?" is
> > > > not a helpful
> > > > > > question.
> > > > > >
> > > > > > A more useful question is "what reason would you
> > possibly have for
> > > > > > overriding the server's request that you do pNFS when
> > > > your client has
> > > > > > pNFS support?" What makes pNFS so special that we must allow
> > > > > > administrators to do this on a per-mount basis?
> > > > >
> > > > > By the same logic, why should a user be allowed to select
> > > > which version of NFS they use for mounting when the server
> > > > has a perfectly reasonable way of negotiating it? Getting to
> > > > choose v2 vs. v3 vs. v4 seems like much less of a distinction
> > > > than choosing between pNFS and no pNFS. Frankly, it never
> > > > even occurred to me that there wouldn't be a mount option to
> > > > make this choice. Enabling/disabling the layout driver
> > > > doesn't fit the existing model of choosing mount behavior,
> > > > and is a big hammer--it's all or nothing.
> > > > >
> > > > > Anyway, here's a use case: I'm working at an
> > > > HPC/gas+oil/satellite data site. We have an awesome pNFS
> > > > server for our big data and I want to access my big data with
> > > > pNFS. We have another server for homedirs, some big data,
> > > > and other stuff. Some mounts are fine with pNFS, others are
> > > > abysmal. So, I want to mount some directories with pNFS, and
> > > > some without pNFS, on the same client, independent of the
> > > > server configuration.
> > > >
> > > > mount -t nfs -overs=4,minorversion=0 foo:/ /bar
> > > >
> > > > Done... Any more questions?
> > >
> > > Several, but I'll stick to one rhetorical. Does NFSv4.1
> > have any features, other than pNFS, that are not in 4.0?
> >
> > Why stop now, when you were batting 100? I told you what the criteria
> > were for adding more mount options, and you start whining about not
> > being able to conceive of a world without mount options.
> >
> > The point is that NFSv4.1 was supposed to let the _server_ tell the
> > client when to use pNFS. The reason why you let the _server_
> > do this, is
> > because pNFS is about enabling _server_ scalability. It is not about
> > faster clients...
> > If you don't want the client to use pNFS, then fix the _server_
> > settings...
>
> I don't miss the kiss-up-kick-down model of software development...
>
> The criteria was a use case, so I gave a use case. I wouldn't rule out the "testing use case" either, as a mount option makes it possible to test mixed pnfs/non-pnfs 4.1 traffic without multiple servers.
>
> The point is that pNFS, like everything else, is about the users.
The criteria was a use case that explains adequately why the client
needs to be able to override the server settings, not just random 'I
might want to do this' moments.
Rhetorical crap like 'it's about the users' isn't helping much either.
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
No mount option - just configure your machine to not load any pnfs
layoutdriver modules.
-->Andy
On Fri, Jan 14, 2011 at 10:19 AM, Jim Rees <[email protected]> wrote:
> Is there some way (mount option?) I can ask the client to attempt a 4.1
> (sessions) mount but do not use pnfs?
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to [email protected]
> More majordomo info at ?http://vger.kernel.org/majordomo-info.html
>
Hi,
----- "Trond Myklebust" <[email protected]> wrote:
> "Why would an administrator never want to do this?" is not a helpful
> question.
>
> A more useful question is "what reason would you possibly have for
> overriding the server's request that you do pNFS when your client has
> pNFS support?" What makes pNFS so special that we must allow
> administrators to do this on a per-mount basis?
Well, I phrased my question the other way because I suspect such cases will be found, but I may not have found all of them.
Some thoughts on why I might wish to take a hand in the decision:
1. the client doing pnfs might behave badly due to a misconfiguration or outage, yet behave acceptably using ordinary nfsv4?
2. restricting the client to ordinary nfsv4 might be desirable for non-developer troubleshooting or other configuration work?
I apologize if neither is compelling.
>
> Throwing more and more knobs into the kernel is easy. The difficult
> bit
> is to figure out which are useful knobs, and that is why I want real
> use
> cases...
>
> Trond
>
Matt
--
Matt Benjamin
The Linux Box
206 South Fifth Ave. Suite 150
Ann Arbor, MI 48104
http://linuxbox.com
tel. 734-761-4689
fax. 734-769-8938
cel. 734-216-5309
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Trond Myklebust
> Sent: Friday, January 14, 2011 7:42 AM
> To: Jim Rees
> Cc: William A. (Andy) Adamson; [email protected]
> Subject: Re: 4.1 no-pnfs mount option?
>
> On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
> > William A. (Andy) Adamson wrote:
> >
> > No mount option - just configure your machine to not load any pnfs
> > layoutdriver modules.
> >
> > That works, thanks, but I was hoping for a way to have
> layout and non-layout
> > mounts going on at the same time.
>
> Different VMs?
>
> Trond
Would there be any objections to adding a "nopnfs" mount option to force this behavior? It could be useful at least for testing, possibly for working around server problems, or if an admin knows that certain clients' usage patterns would be better served by disabling pnfs.
-Dan
>
> --
> To unsubscribe from this list: send the line "unsubscribe
> linux-nfs" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
On Tue, 2011-01-18 at 21:29 -0500, [email protected] wrote:
>
> > -----Original Message-----
> > From: Trond Myklebust [mailto:[email protected]]
> > Sent: Tuesday, January 18, 2011 5:45 PM
> > To: Muntz, Daniel
> > Cc: [email protected]; [email protected];
> > [email protected]; [email protected];
> > [email protected]
> > Subject: RE: 4.1 no-pnfs mount option?
> >
> > On Tue, 2011-01-18 at 19:53 -0500, [email protected] wrote:
> > >
> > > > -----Original Message-----
> > > > From: Trond Myklebust [mailto:[email protected]]
> > > > Sent: Tuesday, January 18, 2011 11:15 AM
> > > > To: Matt W. Benjamin
> > > > Cc: Muntz, Daniel; [email protected]; [email protected];
> > > > [email protected]; Benny Halevy
> > > > Subject: Re: 4.1 no-pnfs mount option?
> > > >
> > > > On Tue, 2011-01-18 at 13:46 -0500, Matt W. Benjamin wrote:
> > > > > Hi,
> > > > >
> > > > > Isn't by mount a plausible way to select for pnfs
> > > > independent of debugging? Is it assured that a client
> > > > administrator would never reasonably wish to do this?
> > > >
> > > > "Why would an administrator never want to do this?" is
> > not a helpful
> > > > question.
> > > >
> > > > A more useful question is "what reason would you possibly have for
> > > > overriding the server's request that you do pNFS when
> > your client has
> > > > pNFS support?" What makes pNFS so special that we must allow
> > > > administrators to do this on a per-mount basis?
> > >
> > > By the same logic, why should a user be allowed to select
> > which version of NFS they use for mounting when the server
> > has a perfectly reasonable way of negotiating it? Getting to
> > choose v2 vs. v3 vs. v4 seems like much less of a distinction
> > than choosing between pNFS and no pNFS. Frankly, it never
> > even occurred to me that there wouldn't be a mount option to
> > make this choice. Enabling/disabling the layout driver
> > doesn't fit the existing model of choosing mount behavior,
> > and is a big hammer--it's all or nothing.
> > >
> > > Anyway, here's a use case: I'm working at an
> > HPC/gas+oil/satellite data site. We have an awesome pNFS
> > server for our big data and I want to access my big data with
> > pNFS. We have another server for homedirs, some big data,
> > and other stuff. Some mounts are fine with pNFS, others are
> > abysmal. So, I want to mount some directories with pNFS, and
> > some without pNFS, on the same client, independent of the
> > server configuration.
> >
> > mount -t nfs -overs=4,minorversion=0 foo:/ /bar
> >
> > Done... Any more questions?
>
> Several, but I'll stick to one rhetorical. Does NFSv4.1 have any features, other than pNFS, that are not in 4.0?
Why stop now, when you were batting 100? I told you what the criteria
were for adding more mount options, and you start whining about not
being able to conceive of a world without mount options.
The point is that NFSv4.1 was supposed to let the _server_ tell the
client when to use pNFS. The reason why you let the _server_ do this, is
because pNFS is about enabling _server_ scalability. It is not about
faster clients...
If you don't want the client to use pNFS, then fix the _server_
settings...
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
Hi,
Isn't by mount a plausible way to select for pnfs independent of debugging? Is it assured that a client administrator would never reasonably wish to do this?
Matt
----- "Trond Myklebust" <[email protected]> wrote:
> >
> > Such a mount option could be useful for dealing with buggy servers
> (pnfs-wise) so you
> > could mount one server with pnfs and another without.
>
> You can find ways around that. Just use 2 clients: one with pnfs
> switched on, and one with it off.
>
> I really don't want to introduce mount options upstream unless they
> are
> useful in the long term. One off usefulness does not pass that test.
>
> --
> Trond Myklebust
> Linux NFS client maintainer
>
> NetApp
> [email protected]
> http://www.netapp.com
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs"
> in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Matt Benjamin
The Linux Box
206 South Fifth Ave. Suite 150
Ann Arbor, MI 48104
http://linuxbox.com
tel. 734-761-4689
fax. 734-769-8938
cel. 734-216-5309
On 2011-01-18 20:28, Trond Myklebust wrote:
> On Tue, 2011-01-18 at 12:44 -0500, [email protected] wrote:
>>
>>> -----Original Message-----
>>> From: [email protected]
>>> [mailto:[email protected]] On Behalf Of Trond Myklebust
>>> Sent: Friday, January 14, 2011 7:42 AM
>>> To: Jim Rees
>>> Cc: William A. (Andy) Adamson; [email protected]
>>> Subject: Re: 4.1 no-pnfs mount option?
>>>
>>> On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
>>>> William A. (Andy) Adamson wrote:
>>>>
>>>> No mount option - just configure your machine to not load any pnfs
>>>> layoutdriver modules.
>>>>
>>>> That works, thanks, but I was hoping for a way to have
>>> layout and non-layout
>>>> mounts going on at the same time.
>>>
>>> Different VMs?
>>>
>>> Trond
>>
>> Would there be any objections to adding a "nopnfs" mount option to force this behavior? It could be useful at least for testing, possibly for working around server problems, or if an admin knows that certain clients' usage patterns would be better served by disabling pnfs.
>
> Yes. Why should we be adding debugging mount options to the upstream
> code? Just test the damned pnfs code properly before it goes upstream...
>
> Trond
Such a mount option could be useful for dealing with buggy servers (pnfs-wise) so you
could mount one server with pnfs and another without.
Benny
On Tue, 2011-01-18 at 20:35 +0200, Benny Halevy wrote:
> On 2011-01-18 20:28, Trond Myklebust wrote:
> > On Tue, 2011-01-18 at 12:44 -0500, [email protected] wrote:
> >>
> >>> -----Original Message-----
> >>> From: [email protected]
> >>> [mailto:[email protected]] On Behalf Of Trond Myklebust
> >>> Sent: Friday, January 14, 2011 7:42 AM
> >>> To: Jim Rees
> >>> Cc: William A. (Andy) Adamson; [email protected]
> >>> Subject: Re: 4.1 no-pnfs mount option?
> >>>
> >>> On Fri, 2011-01-14 at 10:38 -0500, Jim Rees wrote:
> >>>> William A. (Andy) Adamson wrote:
> >>>>
> >>>> No mount option - just configure your machine to not load any pnfs
> >>>> layoutdriver modules.
> >>>>
> >>>> That works, thanks, but I was hoping for a way to have
> >>> layout and non-layout
> >>>> mounts going on at the same time.
> >>>
> >>> Different VMs?
> >>>
> >>> Trond
> >>
> >> Would there be any objections to adding a "nopnfs" mount option to force this behavior? It could be useful at least for testing, possibly for working around server problems, or if an admin knows that certain clients' usage patterns would be better served by disabling pnfs.
> >
> > Yes. Why should we be adding debugging mount options to the upstream
> > code? Just test the damned pnfs code properly before it goes upstream...
> >
> > Trond
>
> Such a mount option could be useful for dealing with buggy servers (pnfs-wise) so you
> could mount one server with pnfs and another without.
You can find ways around that. Just use 2 clients: one with pnfs
switched on, and one with it off.
I really don't want to introduce mount options upstream unless they are
useful in the long term. One off usefulness does not pass that test.
--
Trond Myklebust
Linux NFS client maintainer
NetApp
[email protected]
http://www.netapp.com
Have we beaten this dead horse to a pulp yet? I was just asking if there
was such an option, not advocating for adding one. It would have been
useful at the time for my own testing but I can see the argument against
adding unwanted knobs.
The discussion has been useful in one way. I will now be spending some time
making sure the block layout driver falls back gracefully to no-pnfs if, for
example, an iscsi target can't be contacted.