[Gluster-devel] The return of the all-null pending matrix

[Vijay Bellur]

On 07/14/2013 06:31 AM, Emmanuel Dreyfus wrote:
> Hi
>
> No idea on that one? This is a huge regression from 3.3.x, since 3.4.0
> will not work more than a few hours at mine before corrupting files.


I checked the logs and did not notice anything unusual in the logs.

Do you have multiple clients involved? Does turning off eager-locking 
through:

volume set <volname> eager-lock off

help?

Thanks,
Vijay

>
> Emmanuel Dreyfus <manu at netbsd.org> wrote:
>
>> Vijay Bellur <vbellur at redhat.com> wrote:
>>
>>> Are there any disconnections between the client and bricks?
>>
>> Not any. See complete log:
>> http://ftp.espci.fr/shadow/manu/nullmatrix340b4.log
>>
>>> Is this seen with a striped-replicate setup? Do we observe the same
>>> behavior when stripe xlator is not involved, say in a
>>> distributed-replicate setup?
>>
>> It is distributed-replicate. No stripe xlator involved.
>
>