2007/416 NFSv4 Mirror-mounts

[Tom Haynes]

Tom Haynes wrote:
> Nicolas Williams wrote:
>> On Tue, Jul 17, 2007 at 10:29:06PM +0100, Calum Mackay wrote:
>>  
>>> Interaction with the automounter
>>>
>>> Where there is an existing automount trigger point setup for a
>>> particular server filesystem, it will take precedence over
>>> mirror-mounting, i.e. a mirror-mount will not occur for that 
>>> filesystem.
>>>     
>>
>> So, no mirror mounts in /net?!  Please say it ain't so :(
>>
>> The current /net -hosts automount map depends on the MOUNT protocol to
>> get hierarchical mount information and does honor hierarchical mounts,
>> but it does not refresh this ever, which is a big problem when the
>> server uses ZFS (and therefore likes to make/destroy datasets willy
>> nilly).  Mirror mounts would clearly solve this problem.
>>
>> Nico
>>   
>
>
> It ain't so.
>
> I think the intent here is that if you have a choice over an automount 
> or a mirror mount, we take the
> automount.
>
> But if there is no automount entry and there is a chance to mirror 
> mount, then we will of course
> mirror mount.
>
> Assume that a server has a zfs pool called tank and it is serving up 
> home dirs.
>
> So:
>
> cd /net/server
>
> No mirror mounts have occurred.
>
> No:
>
> cd tank
>
> Okay, this will cause a mirror mount to occur. And as we dive down 
> farther, more will occur.
>
>

I was wrong here. In this case, tank already exists and would thus have 
been gotten from the
MOUNT protocol doing a scan.

I was looking at the non-/net case.

In re-reading your email Nico, I see what you are really concerned with 
is the following scenario:

mount server:/ /mnt
cd /mnt
ls -la
...
.... tank
...
cd tank
ls -la
...
...
...

At which point those ZFS filesystems have been read in (actually occured 
way before this, right?).

And now if we go to the server and create a new zfs filesystem in tank, 
will a mirrormount occur?

I'm going to have to run this test and see what happens. I'll get back 
with that information.