Printable View
Hello everybody,
i'm just playing around and wan't to build an active/active NFS Server.
But when switching the service, clients get stale NFS handles.
I figured out, that it has something to do with /var/lib/nfs.
I'm using a Linux 2.4.21 Kernel system.....
What options have to be set, in order to get an active/active
cluster running.
Regards,
D.Laurenz
On Thu, 14 Oct 2004 09:12:39 +0200, "Dirk Laurenz"
<dirk.laurenz@fujitsu-siemens.com> wrote:
[color=blue]
>Hello everybody,
>
>i'm just playing around and wan't to build an active/active NFS Server.
>But when switching the service, clients get stale NFS handles.
>I figured out, that it has something to do with /var/lib/nfs.
>I'm using a Linux 2.4.21 Kernel system.....
>What options have to be set, in order to get an active/active
>cluster running.
>
>Regards,
>
>D.Laurenz
>[/color]
No clue what the specific settings are but the IP address has to be
picked up by the partner. In appliance level failover the mac and IP
are assumed by the partner. Only in this way are mounts and locks
kept on the client. At least afaik.
~F
Hi,
thanks, in advance, but i already failover a virtual ip but not the mac address. Do you have any firther suggestion?
"Faeandar" <mr_castalot@yahoo.com> schrieb im Newsbeitrag news:btr8n0dknh7rehsa9bhkth1e3fvluu7fs2@4ax.com...[color=blue]
> On Thu, 14 Oct 2004 09:12:39 +0200, "Dirk Laurenz"
> <dirk.laurenz@fujitsu-siemens.com> wrote:
>[color=green]
> >Hello everybody,
> >
> >i'm just playing around and wan't to build an active/active NFS Server.
> >But when switching the service, clients get stale NFS handles.
> >I figured out, that it has something to do with /var/lib/nfs.
> >I'm using a Linux 2.4.21 Kernel system.....
> >What options have to be set, in order to get an active/active
> >cluster running.
> >
> >Regards,
> >
> >D.Laurenz
> >[/color]
>
> No clue what the specific settings are but the IP address has to be
> picked up by the partner. In appliance level failover the mac and IP
> are assumed by the partner. Only in this way are mounts and locks
> kept on the client. At least afaik.
>
> ~F[/color]
On Thu, 21 Oct 2004 20:21:23 +0200, "Dirk Laurenz"
<dirk.laurenz@fujitsu-siemens.com> wrote:
[color=blue]
>Hi,
>
>thanks, in advance, but i already failover a virtual ip but not the mac address. Do you have any firther suggestion?
>"Faeandar" <mr_castalot@yahoo.com> schrieb im Newsbeitrag news:btr8n0dknh7rehsa9bhkth1e3fvluu7fs2@4ax.com...[color=green]
>> On Thu, 14 Oct 2004 09:12:39 +0200, "Dirk Laurenz"
>> <dirk.laurenz@fujitsu-siemens.com> wrote:
>>[color=darkred]
>> >Hello everybody,
>> >
>> >i'm just playing around and wan't to build an active/active NFS Server.
>> >But when switching the service, clients get stale NFS handles.
>> >I figured out, that it has something to do with /var/lib/nfs.
>> >I'm using a Linux 2.4.21 Kernel system.....
>> >What options have to be set, in order to get an active/active
>> >cluster running.
>> >
>> >Regards,
>> >
>> >D.Laurenz
>> >[/color]
>>
>> No clue what the specific settings are but the IP address has to be
>> picked up by the partner. In appliance level failover the mac and IP
>> are assumed by the partner. Only in this way are mounts and locks
>> kept on the client. At least afaik.
>>
>> ~F[/color]
>[/color]
Hmm, only that there has to be coherency between whatever holds the
locks on each server. In the appliance world it's nvram, don';t know
what it is in the Linux world. But it has to exist otherwise there's
no migration of locks and this causes the stale mounts. Any new
connection would be fine, but existing ones on the failed partner will
stale and hang.
~F
Faeandar <mr_castalot@yahoo.com> wrote in message news:<aj7gn0lvgshqanri7d3u61t3aj3jr7di98@4ax.com>...[color=blue]
> On Thu, 21 Oct 2004 20:21:23 +0200, "Dirk Laurenz"
> <dirk.laurenz@fujitsu-siemens.com> wrote:
>[color=green]
> >Hi,
> >
> >thanks, in advance, but i already failover a virtual ip but not the mac address. Do you have any firther suggestion?
> >"Faeandar" <mr_castalot@yahoo.com> schrieb im Newsbeitrag news:btr8n0dknh7rehsa9bhkth1e3fvluu7fs2@4ax.com...[color=darkred]
> >> On Thu, 14 Oct 2004 09:12:39 +0200, "Dirk Laurenz"
> >> <dirk.laurenz@fujitsu-siemens.com> wrote:
> >>
> >> >Hello everybody,
> >> >
> >> >i'm just playing around and wan't to build an active/active NFS Server.
> >> >But when switching the service, clients get stale NFS handles.
> >> >I figured out, that it has something to do with /var/lib/nfs.
> >> >I'm using a Linux 2.4.21 Kernel system.....
> >> >What options have to be set, in order to get an active/active
> >> >cluster running.
> >> >
> >> >Regards,
> >> >
> >> >D.Laurenz
> >> >
> >>
> >> No clue what the specific settings are but the IP address has to be
> >> picked up by the partner. In appliance level failover the mac and IP
> >> are assumed by the partner. Only in this way are mounts and locks
> >> kept on the client. At least afaik.
> >>
> >> ~F[/color]
> >[/color]
>
> Hmm, only that there has to be coherency between whatever holds the
> locks on each server. In the appliance world it's nvram, don';t know
> what it is in the Linux world. But it has to exist otherwise there's
> no migration of locks and this causes the stale mounts. Any new
> connection would be fine, but existing ones on the failed partner will
> stale and hang.
>
> ~F[/color]
you're switching the disk that you're serving between servers when you
fail over right?
the major and minor device numbers need to be the same when you fail
over the disk. see [url]http://linuxha.trick.ca/HaNFS[/url]
active/active means that both servers in your cluster (assuming it's
just 2) are each serving NFS shares. it's not clear from your note if
that's what you're thinking.
Hi,
[color=blue]
> you're switching the disk that you're serving between servers when you
> fail over right?
>
> the major and minor device numbers need to be the same when you fail
> over the disk. see [url]http://linuxha.trick.ca/HaNFS[/url]
>
> active/active means that both servers in your cluster (assuming it's
> just 2) are each serving NFS shares. it's not clear from your note if
> that's what you're thinking.[/color]
That's right, it's a two node cluster with shared disks. But they're used with
lvm on top. Major / Minor Numbers are equal.