[GE users] A "do not disturb" flag for low prio jobs?

Reuti reuti at staff.uni-marburg.de
Wed Dec 21 15:24:34 GMT 2005


Hi,

Am 21.12.2005 um 15:54 schrieb Stefan.O.Nordlander at astrazeneca.com:

> Hey,
>
> I was just asked if there is a way to make SGE aware that a job is  
> currently
> in a "sensitive" state and shouldn't be suspended. Lets say you  
> have 56
> normal priority queues and they're all full. Then a job with higher  
> priority
> comes along and ends up on node 050. However, the original job on  
> node 050
> just sent away a web request to a database/whatever and if that  
> process gets
> interrupted the job will fail once resumed.

In the job, you can set and reset a context with:

qalter -ac dontstop $JOB_ID

(similar -sc and -dc).

This can be checked in a custom defined suspend method for this queue  
to wait a little bit. Then of course you need some logic in the  
suspend script, to wait for this flag to be reset (and not to be set  
again too fast) - this handshaking isn't atomic and might lead to  
race conditions anyway.

-- Reuti

> Is there a way to solve this? The chance of it happening is remote,  
> but
> still, it's there and we'd like to prevent it if possible. A "Do Not
> Disturb" flag that could be set prior to sending the database  
> request and
> then released when it's done would be an alternative I guess. But  
> is it
> possible?
>
>
> Thanks, and a merry x-mas!
> /Stefan
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list