Job dependencies not propertly tracked for jobs in routing queue
gbeane opened this issue · comments
If we have jobs in a routing queue waiting for slots to open up in an execution queue (due to max_user_queuable limits), and those jobs have dependencies on jobs already in the execution queue, then the job may get rejected when it does finally route if one or more of the jobs it depends on have already executed and completed. upon routing, job is immediately rejected: "Job rejected by all possible destinations (check syntax, queue resources, ...)"
We need job dependencies to be tracked properly for jobs in routing queues.
that's great!
When will we see a Torque release with these fixes? We are currently on Torque 6.0.x on our clusters.