2011-01-13 20:26:02

by Mkrtchyan, Tigran

[permalink] [raw]
Subject: nfsv4 + readdir == too many requests



By running intensive tests I notice that readdir oration over nfsv4.1
asks for
file attributes in readdir (attr_request) and later on for each file. To
me it looks like
broken.

This 2.6.37 + pnfs-all-latests :

git://linux-nfs.org/~bhalevy/linux-pnfs.git
378978453b64bf895de8cd0e68dbcd9e9c0155ce

100% reproducible.

Regards,
Tigran.


2011-01-17 15:38:42

by Mkrtchyan, Tigran

[permalink] [raw]
Subject: Re: nfsv4 + readdir == too many requests

Still didn't got a time to test with vanilla kernel. But got some details.
It's works as expected with sec=sys. With sec=krb5 I see this effect.
Unfortunately kernel crashes as soon as I try to re-run tests. It simply
freezes with no stack traces....

Regards,
Tigran

On 01/13/2011 10:14 PM, Tigran Mkrtchyan wrote:
> On 01/13/2011 10:01 PM, Trond Myklebust wrote:
>> On Thu, 2011-01-13 at 21:26 +0100, Tigran Mkrtchyan wrote:
>>
>>> By running intensive tests I notice that readdir oration over nfsv4.1
>>> asks for
>>> file attributes in readdir (attr_request) and later on for each file. To
>>> me it looks like
>>> broken.
>>>
>>> This 2.6.37 + pnfs-all-latests :
>>>
>>> git://linux-nfs.org/~bhalevy/linux-pnfs.git
>>> 378978453b64bf895de8cd0e68dbcd9e9c0155ce
>>>
>>> 100% reproducible.
>>>
>> 1) What makes you think this is broken, and not just a case of the
>> attribute cache timing out and/or the directory changing?
>>
> The directory does not changed. I issue 'ls -l' and with wireshak can
> see READDIR requests
> followed by GETATTR. looks like cache is ignored or too small.
>> 2) Can you produce a testcase that proves brokenness?
>>
> I will try to come up with something.
>> 3) Can you reproduce on a non-pnfs-all-latests kernel?
>>
>> Readdir shouldn't be affected by the stuff in Benny's tree, but I'm
>> always wary of tests on out-of-mainline trees.
>>
> NP. with vanilla 2.6.37 ?
>
> Regards,
> Tigran.
>> Cheers
>> Trond
>>
> --
> 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


2011-01-13 21:01:42

by Myklebust, Trond

[permalink] [raw]
Subject: Re: nfsv4 + readdir == too many requests

On Thu, 2011-01-13 at 21:26 +0100, Tigran Mkrtchyan wrote:
>
> By running intensive tests I notice that readdir oration over nfsv4.1
> asks for
> file attributes in readdir (attr_request) and later on for each file. To
> me it looks like
> broken.
>
> This 2.6.37 + pnfs-all-latests :
>
> git://linux-nfs.org/~bhalevy/linux-pnfs.git
> 378978453b64bf895de8cd0e68dbcd9e9c0155ce
>
> 100% reproducible.

1) What makes you think this is broken, and not just a case of the
attribute cache timing out and/or the directory changing?
2) Can you produce a testcase that proves brokenness?
3) Can you reproduce on a non-pnfs-all-latests kernel?

Readdir shouldn't be affected by the stuff in Benny's tree, but I'm
always wary of tests on out-of-mainline trees.

Cheers
Trond
--
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
http://www.netapp.com


2011-01-13 21:14:30

by Mkrtchyan, Tigran

[permalink] [raw]
Subject: Re: nfsv4 + readdir == too many requests

On 01/13/2011 10:01 PM, Trond Myklebust wrote:
> On Thu, 2011-01-13 at 21:26 +0100, Tigran Mkrtchyan wrote:
>
>> By running intensive tests I notice that readdir oration over nfsv4.1
>> asks for
>> file attributes in readdir (attr_request) and later on for each file. To
>> me it looks like
>> broken.
>>
>> This 2.6.37 + pnfs-all-latests :
>>
>> git://linux-nfs.org/~bhalevy/linux-pnfs.git
>> 378978453b64bf895de8cd0e68dbcd9e9c0155ce
>>
>> 100% reproducible.
>>
> 1) What makes you think this is broken, and not just a case of the
> attribute cache timing out and/or the directory changing?
>

The directory does not changed. I issue 'ls -l' and with wireshak can
see READDIR requests
followed by GETATTR. looks like cache is ignored or too small.
> 2) Can you produce a testcase that proves brokenness?
>
I will try to come up with something.
> 3) Can you reproduce on a non-pnfs-all-latests kernel?
>
> Readdir shouldn't be affected by the stuff in Benny's tree, but I'm
> always wary of tests on out-of-mainline trees.
>
NP. with vanilla 2.6.37 ?

Regards,
Tigran.
> Cheers
> Trond
>