Okay, thanks for looking into it.
In the meantime I've started using a TaskContinuousMapperResource and
making subgrid.size() jobs available initially and then feeding out an
additional job in each result() call.
It was actually surprisingly easy to change the Task implementation. I
love that the Ignite interfaces are so clean!
The users like that jobs are now being completed in-order.
So far it isn't crashing. Since making the change I've noticed that
several of the compute nodes have become idle for 2-3 hour periods
during the compute. Perhaps I've got an off-by-one error in my job
feeding logic. That part of the code is dead-simple so I don't see how
it could be off but I'm going to add more logging to track down when the
mapper makes each job available vs when each node starts working on the job.
I'd still prefer using the parallelJobsNumber setting.
Thanks again,
Ryan
On 4/13/2017 2:18 AM, vkulichenko wrote:
Hi Ryan,
I will take a look at your sample in the next few days.
-Val
--
View this message in context:
http://apache-ignite-users.70518.x6.nabble.com/Concurrent-job-execution-and-FifoQueueCollisionSpi-parallelJobsNumber-1-tp8697p11935.html
Sent from the Apache Ignite Users mailing list archive at Nabble.com.
