Commit bd90a558 authored by Brewer, Wes's avatar Brewer, Wes
Browse files

Fix issue with `python main.py -w benchmark` hanging

parent a8d186eb

raps/engine.py

+6 −3
Original line number Diff line number Diff line
@@ -194,12 +194,15 @@ class Engine: 


        for timestep in range(timesteps):

            while self.current_time >= last_submit_time and jobs:

                self.scheduler.schedule(self.queue, self.running, self.available_nodes, self.current_time)



                job = jobs.pop(0)

                job = Job(job_info, self.current_time)

                self.scheduler.schedule([job], self.running, self.available_nodes, self.current_time)



                if jobs:

                    last_submit_time = job.submit_time

                else:  # No more jobs, set submit_time to infinity to avoid triggering again

                    last_submit_time = float('inf')

                else:

                    last_submit_time = float('inf')  # Avoid infinite loop



            yield self.tick()

raps/schedulers/default.py

+8 −7
Original line number Diff line number Diff line
@@ -50,10 +50,12 @@ class Scheduler: 
        job.state = JobState.RUNNING  # Job is now running





    def schedule(self, queue, running, available_nodes, current_time):

        """Schedules jobs from the queue to available nodes."""

        while queue:

            job = queue.pop(0)

    def schedule(self, job_list, running, available_nodes, current_time):

        """Schedules jobs from the given job_list directly, modifying available_nodes."""

        

        while job_list:

            job = job_list.pop(0)



            if len(available_nodes) >= job.nodes_required:

                job.scheduled_nodes = available_nodes[:job.nodes_required]

                available_nodes[:] = available_nodes[job.nodes_required:]
@@ -61,10 +63,9 @@ class Scheduler: 
                job.end_time = current_time + job.wall_time

                job.state = JobState.RUNNING

                running.append(job)

                #print(f"t={current_time}: Scheduled job {job.id} on nodes {summarize_ranges(job.scheduled_nodes)}")

            else:

                queue.insert(0, job)  # Keep the job at the front if it can't be scheduled

                break

                job_list.insert(0, job)  # Put job back at the front if it can't be scheduled

                break  # Stop scheduling if no nodes are available





    def schedule2(self, queue, running, available_nodes, current_time, debug=False):