svn merge -r10852:10883 https://eris.llnl.gov/svn/slurm/branches/slurm-1.1

NEWS

@@ -217,6 +217,11 @@ documents those changes that are of interest to users and admins.
     the code)
  -- Added support for OSX build.
 
+* Changes in SLURM 1.1.27
+=========================
+ - Fix possible race condition for two simultaneous "scontrol show config" 
+   calls resulting in slurm_xfree() Error: from read_config.c:642 
+
 * Changes in SLURM 1.1.26
 =========================
  - In sched/wiki2, fixes for support of job features.
@@ -669,6 +674,9 @@ documents those changes that are of interest to users and admins.
 * Changes in SLURM 1.0.17
 =========================
  -- Set correct user groups for task epilogs.
+ -- Set SLURM_DIST_CYCLIC = 1 (needed for HP MPI, slurm.hp.env.patch).
+ -- Add more debugging for tracking slow slurmd job initiations
+    (slurm.hp.replaydebug.patch).
 
 * Changes in SLURM 1.0.16
 =========================

slurm/slurm.h.in

@@ -309,7 +309,7 @@ enum jobacct_data_type {
 
 /* Possible task distributions across the nodes */
 typedef enum task_dist_states {
-	SLURM_DIST_CYCLIC,	/* distribute tasks 1 per node, round robin */
+	SLURM_DIST_CYCLIC = 1,	/* distribute tasks 1 per node, round robin */
 	SLURM_DIST_BLOCK,	/* distribute tasks filling node by node */
 	SLURM_DIST_ARBITRARY,	/* arbitrary task distribution  */
 	SLURM_DIST_PLANE,	/* distribute tasks by filling up

src/common/read_config.c

@@ -1023,9 +1023,11 @@ gethostname_short (char *name, size_t len)
 /* 
  * free_slurm_conf - free all storage associated with a slurm_ctl_conf_t.   
  * IN/OUT ctl_conf_ptr - pointer to data structure to be freed
+ * IN purge_node_hash - purge system-wide node hash table if set,
+ *			set to zero if clearing private copy of config data
  */
-void
-free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
+extern void
+free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr, bool purge_node_hash)
 {
 	xfree (ctl_conf_ptr->authtype);
 	xfree (ctl_conf_ptr->checkpoint_type);
@@ -1067,8 +1069,9 @@ free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
 	xfree (ctl_conf_ptr->srun_prolog);
 	xfree (ctl_conf_ptr->srun_epilog);
 	xfree (ctl_conf_ptr->node_prefix);
-	
-	_free_name_hashtbl();
+
+	if (purge_node_hash)
+		_free_name_hashtbl();
 }
 
 /* 
@@ -1189,7 +1192,7 @@ _destroy_slurm_conf()
 		s_p_hashtbl_destroy(default_partition_tbl);
 		default_partition_tbl = NULL;
 	}
-	free_slurm_conf(conf_ptr);
+	free_slurm_conf(conf_ptr, true);
 	conf_initialized = false;
 	
 	/* xfree(conf_ptr); */

src/common/read_config.h

@@ -281,8 +281,11 @@ extern void init_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr);
 /* 
  * free_slurm_conf - free all storage associated with a slurm_ctl_conf_t.   
  * IN/OUT ctl_conf_ptr - pointer to data structure to be freed
+ * IN purge_node_hash - purge system-wide node hash table if set,
+ *			set to zero if clearing private copy of config data
  */
-extern void free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr);
+extern void free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr,
+			     bool purge_node_hash);
 
 /*
  * gethostname_short - equivalent to gethostname(), but return only the first 

src/slurmctld/proc_req.c

@@ -548,7 +548,7 @@ static void _slurm_rpc_dump_conf(slurm_msg_t * msg)
 		
 		/* send message */
 		slurm_send_node_msg(msg->conn_fd, &response_msg);
-		free_slurm_conf(&config_tbl);
+		free_slurm_conf(&config_tbl, false);
 	}
 }
 

src/slurmd/slurmd/req.c

@@ -389,12 +389,15 @@ _send_slurmstepd_init(int fd, slurmd_step_type_t type, void *req,
 	free_buf(buffer);
 	
 	/* send cached group ids array for the relevant uid */
+	debug3("_send_slurmstepd_init: call to getpwuid");
 	if (!(pw = getpwuid(uid))) {
 		error("_send_slurmstepd_init getpwuid: %m");
 		len = 0;
 		safe_write(fd, &len, sizeof(int));
 		return -1;
 	}
+	debug3("_send_slurmstepd_init: return from getpwuid");
+
 	if ((gids = _gids_cache_lookup(pw->pw_name, pw->pw_gid))) {
 		int i;
 		uint32_t tmp32;
@@ -709,8 +712,10 @@ _rpc_launch_tasks(slurm_msg_t *msg)
 	adlen = sizeof(self);
 	_slurm_getsockname(msg->conn_fd, (struct sockaddr *)&self, &adlen);
 
+	debug3("_rpc_launch_tasks: call to _forkexec_slurmstepd");
 	errnum = _forkexec_slurmstepd(LAUNCH_TASKS, (void *)req, cli, &self,
 				      step_hset);
+	debug3("_rpc_launch_tasks: return from _forkexec_slurmstepd");
 
     done:
 	if (step_hset)
@@ -846,8 +851,10 @@ _rpc_batch_job(slurm_msg_t *msg)
 		info("Launching batch job %u.%u for UID %d",
 			req->job_id, req->step_id, req->uid);
 
+	debug3("_rpc_batch_job: call to _forkexec_slurmstepd");
 	rc = _forkexec_slurmstepd(LAUNCH_BATCH_JOB, (void *)req, cli, NULL,
 				  (hostset_t)NULL);
+	debug3("_rpc_batch_job: return from _forkexec_slurmstepd");
 
 	slurm_mutex_unlock(&launch_mutex);
 

src/srun/allocate.c

@@ -667,6 +667,7 @@ _step_req_create(srun_job_t *j)
 		break;
 
 	}
+	opt.distribution = r->task_dist;
 	
 	if (slurmctld_comm_addr.port) {
 		r->host = xstrdup(slurmctld_comm_addr.hostname);