[Cluster-devel] Re: [NFS] [PATCH 3/4 Revised] NLM - kernel lockd-statd changes

[Wendy Cheng <wcheng@redhat.com>]

Olaf Kirch wrote:
> On Tuesday 17 April 2007 15:24, Wendy Cheng wrote:
>   
>>> I think in term of correctness, it's better to send an SM_NOTIFY
>>> for each IP associated with such a set, anyway.
>>>   
>>>       
>> That's exactly what we have been proposing... :) .. We'll rely heavily 
>> on HA callout program to tell us which client uses which (server) 
>> floating IP. 
>>     
>
> Correct me if I'm wrong, but it seems your patch records every IP
> used by the client, rather than the *all* the IPs related to the set of
> file systems being moved. So if there are several IPs for this set, you
> will end up sending notifications only from those the client happened
> to have talked to.
>   

Yes. Then why should we SM_NOTIFY the clients that do not have the 
associated locks (and introducing more possible reclaiming delay) ? Be 
aware that failover normally has timing constraints - it needs to get 
finished within a sensible time interval.

-- Wendy
> My point was, you move a set of file systems A, B and C, with
> IPs X, Y, Z. You know what the addresses are, so from a
> robustness point of view your best bet is to send SM_NOTIFY
> messages from IPs X, Y, Z, regardless of whether the client has
> been talking to all of them, or just one.
>
> Olaf
>