Discussion related to cado-nfs

[Emmanuel Thomé <Emmanuel.Thome@inria.fr>]

Hi,

This is entered as an issue here:

https://gitlab.inria.fr/cado-nfs/cado-nfs/-/issues/30031

A tentative fix is here:

https://gitlab.inria.fr/cado-nfs/cado-nfs/-/merge_requests/60

Caveats of my proposed fix are listed in the merge request messages.

Feel free to comment.

Best regards,

E.

On Tue, Feb 01, 2022 at 02:15:44PM +0000, Ed Hall wrote:
>  Dear Emmanuel,
> I see what you are describing as the trouble point.  The server would have 
> to stay running for the entire timeout or until the last client requests 
> its next WU.
> Alternately, could a separate much smaller process be provided, which could 
> be run on the server machine to take care of the graceful exit for the 
> tardy clients?  That would provide users an option.
> Another alternative in my case, would be to provide a manner to tell the 
> clients to shut down if there hasn't been a new WU for <time>.
> 
> Is there a section in the code I could study to see if I can provide 
> something for my own use?
> Sincerely,Edwin Hall
> 
>     On Tuesday, February 1, 2022, 02:00:49 AM EST, Emmanuel Thomé 
> <emmanuel.thome@inria.fr> wrote:  
>  
>  Hi,
> 
> On Tue, Feb 01, 2022 at 03:20:43AM +0000, Ed Hall wrote:
> > Hello Team,
> >
> > It there a manner to have CADO-NFS issue a 410, closing the clients
> > when tasks.sieve.run=false?  If not, could a method be added?  When
> > performing polyselect only, for larger jobs (as well as other
> > taskings), the clients keep pinging the server machine after the server
> > has exited.  My work around for an automated setup with many external
> > clients, has been to run a smaller complete job after the polyselect
> > has finished.  This is not ideal. Thanks.
> 
> Yes, this is a good suggestion.
> 
> Implementation is trickier than it seems.  We're touching on some of the
> brittle aspects of the cado-nfs server, here. The server would have to
> wait for the clients to ask for work so that it gets a chance to answer
> 410. Presumably, waiting for (what the server think is) the clients sleep
> time could be appropriate.
> 
> I may have a try at this, but that won't be immediate.
> 
> Best regards,
> 
> E.
> 
> > 
> > Sincerely,Ed Hall
>