[Spread-users] self-leave and trans

[Jonathan Stanton]

Hi,

On Fri, Feb 20, 2004 at 10:42:19AM -0000, Taj Khattra wrote:
> Hello,
[...]
> The man page for SP_receive() (and the User Guide) mention
> that a self-leave message includes the private group name
> of the leaving connection.  However, SP_receive() always
> seems to return a message size of 0 for such messages.
> Does this mean that a self-leave message contains only the
> group name (in the sender field) just like a transitional
> membership change message ?  

Wow. I checked the logs and this has been in the documentation since the 
man pages were written, and it has been wrong all that time. 

You are correct, the self-leave membership message does not include the 
private_group_name of the leaving process OR the group_id (as mentioned 
below) It only has the sender field with the group name the member left.

I wrote the man page and I can not remember any reason I would have 
documented it that way. The page even states that the private_group_name 
should always be identical to the private_group_name of the connection it 
is received on, so the information seems redundant. 

> The User Guide also mentions that a self-leave message
> 'contains the group_id of the new membership', but which new
> membership would that be - the connection is no longer part
> of the group ?

The group_id is NOT included in the self-leave message. This is a 
documentation mistake.

> 
> Also, how can I trigger transitional membership change messages ?
> I don't ever seem to get them in my simple setup - pointers to
> docs or papers would be welcome.

You will only trigger transitional memberships if you have multiple 
daemons running on different machines.  Even then it is sometimes hard to 
trigger them. You need to have messages flowing through the system and 
then pull the ethernet plug on one of the machines running a daemon (or 
use the spmonitor tool to create an artificial partition --which simulates 
pulling out the ethernet cable--) Then you should receive a transitional 
membership and some messages before receiving the new membership.  

Others may have more precise methods to trigger transitionals.

Thanks,

Jonathan

-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------