Question on IPv6 failover.

Question on IPv6 failover.

[Ben Greear]

One of our users is reporting that the CIFS failover case is not working
properly.  They have a cluster of CIFS servers that migrate an IPv6 addr
to the active system.

Evidently, our client-side CIFS connection follows the first migration,
but fails to go back to the original on a second failover.

It could be any number of things, but I wanted first to see if anyone
knows how long client-side CIFS will keep attempting to regain contact with it's
server?  Forever?

Thanks,
Ben

-- 
Ben Greear <greearb-my8/4N5VtI7c+919tysfdA@public.gmane.org>
Candela Technologies Inc  http://www.candelatech.com

Re: Question on IPv6 failover.

[Steve French]

On Thu, Oct 3, 2013 at 4:04 PM, Ben Greear <greearb-my8/4N5VtI7c+919tysfdA@public.gmane.org> wrote:
> One of our users is reporting that the CIFS failover case is not working
> properly.  They have a cluster of CIFS servers that migrate an IPv6 addr
> to the active system.
>
> Evidently, our client-side CIFS connection follows the first migration,
> but fails to go back to the original on a second failover.
>
> It could be any number of things, but I wanted first to see if anyone
> knows how long client-side CIFS will keep attempting to regain contact with
> it's
> server?  Forever?
>
> Thanks,
> Ben

Depends on kernel version and also hard vs. soft (default) mount options.

The "smb echo" approach that cifs moved to a few years ago allows
the cifs kernel client to notice more quickly that the server is down,
and significantly improved failover.

Which kernel version?


-- 
Thanks,

Steve

Re: Question on IPv6 failover.

[Ben Greear]

On 10/03/2013 03:30 PM, Steve French wrote:
> On Thu, Oct 3, 2013 at 4:04 PM, Ben Greear <greearb-my8/4N5VtI7c+919tysfdA@public.gmane.org> wrote:
>> One of our users is reporting that the CIFS failover case is not working
>> properly.  They have a cluster of CIFS servers that migrate an IPv6 addr
>> to the active system.
>>
>> Evidently, our client-side CIFS connection follows the first migration,
>> but fails to go back to the original on a second failover.
>>
>> It could be any number of things, but I wanted first to see if anyone
>> knows how long client-side CIFS will keep attempting to regain contact with
>> it's
>> server?  Forever?
>>
>> Thanks,
>> Ben
>
> Depends on kernel version and also hard vs. soft (default) mount options.
>
> The "smb echo" approach that cifs moved to a few years ago allows
> the cifs kernel client to notice more quickly that the server is down,
> and significantly improved failover.
>
> Which kernel version?

3.7.10

The hard v/s soft is not being specified when mounting unless the
customer did something interesting, so I guess it is soft.

I'm still waiting on a network dump, so at this time I have no
reason to believe that Linux CIFS is actually to blame...just
trying to understand the possibilities better.

Thanks,
Ben

-- 
Ben Greear <greearb-my8/4N5VtI7c+919tysfdA@public.gmane.org>
Candela Technologies Inc  http://www.candelatech.com

Re: Question on IPv6 failover.

[Jeff Layton]

On Thu, 03 Oct 2013 14:04:55 -0700
Ben Greear <greearb-my8/4N5VtI7c+919tysfdA@public.gmane.org> wrote:

> One of our users is reporting that the CIFS failover case is not working
> properly.  They have a cluster of CIFS servers that migrate an IPv6 addr
> to the active system.
> 
> Evidently, our client-side CIFS connection follows the first migration,
> but fails to go back to the original on a second failover.
> 
> It could be any number of things, but I wanted first to see if anyone
> knows how long client-side CIFS will keep attempting to regain contact with it's
> server?  Forever?
> 

Yep, forever. The address to connect to is determined by the mount
helper and passed down to the kernel at mount time. This is the same
for IPv4 and IPv6. If the address is floating between servers then
it'll just look like the server rebooted when you fail over, and the
client should just keep trying to reconnect indefinitely.

-- 
Jeff Layton <jlayton-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>