Gluster users - Dec 2014 - nfs bug when do dd write

hi all,

       In my environment, I meet a nfs problem.When I do dd write for a while,
nfs client will report "nfs server no responding".I think this maybe a
bug of nfs server, Is anybody meet similar problem?

The details of my environment and test  are as follow:

glusterfs version:
    3.6.1
3 nodes:
  cpu with 8 core 3.20 GHz, 16 GB memory, 10000baseT/Full network card
 
I use two nodes create a replica volume with two bricks, another nodes act as
client, client use protocol nfs3 to mount volume through the 10000baseT/Full
network card, then use command "dd if=/dev/zero of=/mnt/ida/testdata
bs=128k" to write data.

Generally, onlydataabout10G can be write successfully. Then client will say
"nfs server ip no responding".At the same time, I use gdb to trace nfs
process in nfs server, the result is that nfs server keeps staying in
epoll_wait, why nfs server  can not listen requests from nfs client ?

In addition, I noticed the difference between fuse and nfs. When access through
fuse, two threads will work together, one responsible for fuse-bridge, another
responsible for listening socket. But in nfs, only one thread work in
background, both responsible for nfs server and listening socket. Will the cause
process bottleneck?


Best wishes??

Looking forward to yourreply?
 
   
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20141209/88207a8f/attachment.html>

On Tuesday 09 December 2014 02:47 PM, shuau li wrote:
> hi all,
>
>        In my environment, I meet a nfs problem.When I do dd write for 
> a while, nfs client will report "nfs server no responding".I
think
> this maybe a bug of nfs server, Is anybody meet similar problem?
>
> The details of my environment and test  are as follow:
>
> glusterfs version:
>     3.6.1
> 3 nodes:
>   cpu with 8 core 3.20 GHz, 16 GB memory, 10000baseT/Full network card
>
> I use two nodes create a replica volume with two bricks, another nodes 
> act as client, client use protocol nfs3 to mount volume through the 
> 10000baseT/Full network card, then use command "dd if=/dev/zero 
> of=/mnt/ida/testdata bs=128k" to write data.
>
> Generally, only data about 10G can be write successfully. Then client 
> will say "nfs server ip no responding".At the same time, I use
gdb to
> trace nfs process in nfs server, the result is that nfs server keeps 
> staying in epoll_wait, why nfs server  can not listen requests from 
> nfs client ?
>
> In addition, I noticed the difference between fuse and nfs. When 
> access through fuse, two threads will work together, one responsible 
> for fuse-bridge, another responsible for listening socket. But in nfs, 
> only one thread work in background, both responsible for nfs server 
> and listening socket. Will the cause process bottleneck?
>
> Best wishes!!
>
> Looking forward to your reply!
>
>
>
Hi,

Can you please attach the nfs server and brick log files? You can find 
the log files in /var/log/glusterfs.

Regards,
Raghavendra Bhat
>
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20141209/e75a135e/attachment.html>