add some port reservation logic

slurm/slurm_errno.h

@@ -168,6 +168,8 @@ enum {
 	ESLURM_RESERVATION_NOT_USABLE,
 	ESLURM_INVALID_WCKEY,
 	ESLURM_RESERVATION_OVERLAP,
+	ESLURM_PORTS_BUSY,
+	ESLURM_PORTS_INVALID,
 
 	/* switch specific error codes, specific values defined in plugin module */
 	ESLURM_SWITCH_MIN = 3000,

src/common/slurm_errno.c

 /*****************************************************************************\
  *  slurm_errno.c - error codes and functions for slurm
  ******************************************************************************
- *  Copyright (C) 2002-2006 The Regents of the University of California.
+ *  Copyright (C) 2002-2007 The Regents of the University of California.
+ *  Copyright (C) 2008-2009 Lawrence Livermore National Security.
  *  Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
  *  Written by Jim Garlick <garlick@llnl.gov>, et. al.
  *  CODE-OCEC-09-009. All rights reserved.
@@ -237,6 +238,10 @@ static slurm_errtab_t slurm_errtab[] = {
 	  "Requested reservation not usable now"		},
 	{ ESLURM_RESERVATION_OVERLAP, 
 	  "Requested reservation overlaps with another reservation"	},
+	{ ESLURM_PORTS_BUSY,
+	  "Requires ports are in use"				},
+	{ ESLURM_PORTS_INVALID,
+	  "Requires more ports than can be reserved"		},
 
 	/* slurmd error codes */
 

src/slurmctld/port_mgr.c

@@ -41,14 +41,77 @@
 #  include "config.h"
 #endif
 
+#include "src/common/bitstring.h"
+#include "src/common/hostlist.h"
+#include "src/common/xmalloc.h"
+#include "src/common/xstring.h"
+
 #include "src/slurmctld/slurmctld.h"
 
+bitstr_t **port_resv_table = (bitstr_t **) NULL;
+int        port_resv_cnt   = 0;
+int        port_resv_min   = 0;
+int        port_resv_max   = 0;
+
 /* Reserve ports for a job step
  * RET SLURM_SUCCESS or an error code */
 extern int reserve_ports(struct step_record *step_ptr)
 {
-info("reserving %u ports", step_ptr->resv_port_cnt);
-step_ptr->resv_ports = xstrdup("123-456");
+	int i, port_inx;
+	int *port_array = NULL;
+	char port_str[16];
+	hostlist_t hl;
+
+	if (step_ptr->resv_port_cnt > port_resv_cnt) {
+		info("step %u.%u needs %u reserved ports, but only %d exist",
+		     step_ptr->job_ptr->job_id, step_ptr->step_id,
+		     step_ptr->resv_port_cnt, port_resv_cnt);
+		return ESLURM_PORTS_INVALID;
+	}
+
+	/* Identify available ports */
+	port_array = xmalloc(sizeof(int) * step_ptr->resv_port_cnt);
+	port_inx = 0;
+	for (i=0; i<port_resv_cnt; i++) {
+		if (bit_overlap(step_ptr->step_node_bitmap,
+				port_resv_table[i]))
+			continue;
+		port_array[port_inx++] = i;
+		if (port_inx >= step_ptr->resv_port_cnt)
+			break;
+	}
+	if (port_inx < step_ptr->resv_port_cnt) {
+		info("insufficient ports for step %u.%u to reserve (%d of %u)",
+		     step_ptr->job_ptr->job_id, step_ptr->step_id,
+		     port_inx, step_ptr->resv_port_cnt);
+		xfree(port_array);
+		return ESLURM_PORTS_BUSY;
+	}
+
+	/* Reserve selected ports */
+	hl = hostlist_create(NULL);
+	if (hl == NULL)
+		fatal("malloc: hostlist_create");
+	for (i=0; i<port_inx; i++) {
+		bit_or(port_resv_table[port_array[i]], 
+		       step_ptr->step_node_bitmap);
+		snprintf(port_str, sizeof(port_str), 
+			 "%d", (port_array[i] + port_resv_min));
+		hostlist_push(hl, port_str);
+	}
+	hostlist_sort(hl);
+	for (i=1024; ; i*=2) {
+		step_ptr->resv_ports = xmalloc(i);
+		if (hostlist_ranged_string(hl, i, step_ptr->resv_ports) >= 0)
+			break;
+		xfree(step_ptr->resv_ports);
+	}
+	hostlist_destroy(hl);
+	xfree(port_array);
+	info("reserved ports %s for step %u.%u",
+	     step_ptr->resv_ports,
+	     step_ptr->job_ptr->job_id, step_ptr->step_id);
+
 	return SLURM_SUCCESS;
 }
 

src/srun/allocate.c

@@ -596,7 +596,8 @@ create_job_step(srun_job_t *job, bool use_all_cpus)
 		rc = slurm_get_errno();
 
 		if (opt.immediate ||
-		    ((rc != ESLURM_NODES_BUSY) && (rc != ESLURM_DISABLED))) {
+		    ((rc != ESLURM_NODES_BUSY) && (rc != ESLURM_PORTS_BUSY) &&
+		     (rc != ESLURM_DISABLED))) {
 			error ("Unable to create job step: %m");
 			return -1;
 		}