SLURM Scheduler

1.src/plugins/sched/
built-in: will initiate jobs strictly in their priority order, typically (first-in-first-out)
backfill: will initiate a lower-priority job if doing so does not delay the expected initiation time of higher priority jobs; essentially using smaller jobs to fill holes in the resource allocation plan. Effective backfill scheduling does require users to specify job time limits.
gang: gang time-slices jobs in the same partition/queue and can be used to preempt jobs from lower-priority queues in order to execute jobs in higher priority queues.
wiki: is an interface for use with The Maui Scheduler
wiki2: is an interface for use with the Moab Cluster Suite


2. slurm.conf 
# SCHEDULING 
FastSchedule=0
SchedulerType=sched/backfill
#SchedulerPort=7321 
SelectType=select/cons_res




3. slurmctld/job_scheduler.c


/*
  * build_job_queue - build (non-priority ordered) list of pending jobs
* IN clear_start - if set then clear the start_time for pending jobs
  * RET the job queue
  * NOTE: the caller must call list_destroy() on RET value to free memory
  */
extern List build_job_queue(bool clear_start);


/*
  * schedule - attempt to schedule all pending jobs
  * pending jobs for each partition will be scheduled in priority
  * order until a request fails
  * IN job_limit - maximum number of jobs to test now, avoid testing the full
  *  queue on every job submit (0 means to use the system default,
  *  SchedulerParameters for default_queue_depth)
  * RET count of jobs scheduled
  * Note: We re-build the queue every time. Jobs can not only be added
  * or removed from the queue, but have their priority or partition
  * changed with the update_job RPC. In general nodes will be in priority
  * order (by submit time), so the sorting should be pretty fast.
  */
extern int schedule(uint32_t job_limit);


/* Determine if a pending job will run using only the specified nodes
  * (in job_desc_msg->req_nodes), build response message and return
  * SLURM_SUCCESS on success. Otherwise return an error code. Caller
  * must free response message */
extern int job_start_data(job_desc_msg_t *job_desc_msg, will_run_response_msg_t **resp);


/*
  * launch_job - send an RPC to a slurmd to initiate a batch job
  * IN job_ptr - pointer to job that will be initiated
  */
extern void launch_job(struct job_record *job_ptr);