Re: lksctp question: Association redundancy

[Vlad Yasevich <vladislav.yasevich@hp.com>]

Hi Son

Please address your questions to the sctp list.

Son Le wrote:
> Hi all,
>  We have a situation where we have two nodes that have the same ip address and want to control sctp association redundancy. If the process that owns an association on one node crashes we want to do the following:
> 
> - We do not want to send the ABORT to the far-end by setting the SO_LINGER option. So the far-end at this point still think that this association is still active.
> - The stand by node (it learned some info such verification tag of the association of the other node) will send INIT message using the verification tag (and some other info from previous association) to the same far-end. The far-end can uses these info to reestablish the association on the new node. I assume that the far-end when receive the INIT message from the new node with, is not consider new association but will reestablish the current association
> 

It sounds like you want to trigger an association restart on the remote node.
All you have to do is attempt to initiate the association with the same source,
destination, and address parameters.

You do not need to know remotes verification tag.

> Do you think this strategy would work ? Or do we need to enhance the lksctp stack to support this functionality ?
> 

This strategy might work as long as the remote keeps it's association active.
That depends on how fast the remote detects path and association failures.

-vlad

> Thanks for your help.
> 
> Best regards,
> Son
> 
> -----Original Message-----
> From: Michael Tüxen [mailto:Michael.Tuexen@lurchi.franken.de] 
> Sent: Tuesday, December 02, 2008 11:39 AM
> To: Vlad Yasevich
> Cc: Le, Son (CAR:SI52)
> Subject: Re: lksctp question
> 
> On Dec 2, 2008, at 5:25 PM, Vlad Yasevich wrote:
> 
>> Michael Tüxen wrote:
>>> Hi Vlad,
>>>
>>> on question inside...
>>>
>>> Best regards
>>> Michael
>>>
>>> On Dec 2, 2008, at 3:11 PM, Vlad Yasevich wrote:
>>>
>>>> Son Le wrote:
>>>>> Hi all, Can you help to awnser the below question ?
>>>>>
>>>>>
>>>>> 1.    If an application process owns an SCTP association crashes
>>>>> (traps, or gets
>>>>> killed by other management processes), will the lksctp in the 
>>>>> kernel Shutdown the sctp link gracefully? Or will it just send an 
>>>>> abort to the remote side?
>>>> Depends on the setting of SO_LINGER.  By default, this will cause a 
>>>> graceful shutdown since a "crashed" process is no different from a 
>>>> gracefully exiting process in term of how file descriptors are 
>>>> closed.  If you set SO_LINGER, then any close() on the descriptor 
>>>> will trigger an ABORT.
>>> What if messages are still in the recv buffer? FreeBSD would then 
>>> ABORT the association no matter what SO_LINGER is set to....
>> So, if the application calls close() on the socket even if there is 
>> data there, you will ABORT?
> If there is still data in the recv buffer and the application calls
> close()
> the kernel knows that the application will not get everything. So it sends an ABORT...
>>
>> -vlad
>>
>