diff --git a/META b/META
index f8f6e5a5b79829fa49de864468f810214d21cb7e..fbe6e9222c1887af3f868beb75610ec6e52a8759 100644
--- a/META
+++ b/META
@@ -10,4 +10,4 @@
Micro: 0
Version: 0.3.0
Release: 1
- API: 0x010000
+ API: 0x010100
diff --git a/NEWS b/NEWS
index d5c9f17b71675fac5ea039b36fb2fcb87954b758..a6f8de32c51d33ae9472a0041281a6890645ad41 100644
--- a/NEWS
+++ b/NEWS
@@ -1,8 +1,9 @@
This file describes changes in recent versions of SLURM. It primarily
documents those changes that are of interest to users and admins.
-* Changes in SLURM 0.3.1
+* Changes in SLURM 0.3.1 (not tagged yet)
========================
+ -- Set SLURM_TASKS_PER_NODE env var for batch jobs (and LAM/MPI).
-- Fix for slurmd spinning when stdin buffers full (gnats:434)
* Changes in SLURM 0.3.0
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 64d0b76a3fc60e232dfdf58cadc4df4bdac93233..1579bf6535c79a25ff4d7a92391d7f0dfee608c2 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -145,6 +145,8 @@ void slurm_free_job_launch_msg(batch_job_launch_msg_t * msg)
if (msg) {
xfree(msg->nodes);
+ xfree(msg->cpus_per_node);
+ xfree(msg->cpu_count_reps);
xfree(msg->script);
xfree(msg->err);
xfree(msg->in);
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 2febadd293442dc64b72ec01a005f81ca8c0f344..47c7f3add66c29d10363d0f7e1c35422c7807d8a 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -325,6 +325,9 @@ typedef struct batch_job_launch_msg {
uint32_t job_id;
uint32_t uid;
uint32_t nprocs; /* number of tasks in this job */
+ uint16_t num_cpu_groups;/* elements in below cpu arrays */
+ uint32_t *cpus_per_node;/* cpus per node */
+ uint32_t *cpu_count_reps;/* how many nodes have same cpu count */
char *nodes; /* list of nodes allocated to job_step */
char *script; /* the actual job script, default NONE */
char *err; /* pathname of stderr */
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 8d29d678c52d9073f4664d0d3f282f8d3bcfa4de..94c9c49b3c53814ab073bb5fa945f3ef5892b123 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -3,7 +3,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Kevin Tew <tew1@llnl.gov>, Moe Jette <jette1@llnl.gov>, et. al.
+ * Written by Kevin Tew <tew1@llnl.gov>, et. al.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -2486,6 +2486,10 @@ _pack_batch_job_launch_msg(batch_job_launch_msg_t * msg, Buf buffer)
pack32(msg->job_id, buffer);
pack32(msg->uid, buffer);
+ pack16(msg->num_cpu_groups, buffer);
+ pack32_array(msg->cpus_per_node, msg->num_cpu_groups, buffer);
+ pack32_array(msg->cpu_count_reps, msg->num_cpu_groups, buffer);
+
packstr(msg->nodes, buffer);
packstr(msg->script, buffer);
packstr(msg->work_dir, buffer);
@@ -2505,6 +2509,7 @@ static int
_unpack_batch_job_launch_msg(batch_job_launch_msg_t ** msg, Buf buffer)
{
uint16_t uint16_tmp;
+ uint32_t uint32_tmp;
batch_job_launch_msg_t *launch_msg_ptr;
assert(msg != NULL);
@@ -2514,6 +2519,23 @@ _unpack_batch_job_launch_msg(batch_job_launch_msg_t ** msg, Buf buffer)
safe_unpack32(&launch_msg_ptr->job_id, buffer);
safe_unpack32(&launch_msg_ptr->uid, buffer);
+ safe_unpack16(&launch_msg_ptr->num_cpu_groups, buffer);
+ if (launch_msg_ptr->num_cpu_groups > 0) {
+ safe_unpack32_array((uint32_t **) &
+ (launch_msg_ptr->cpus_per_node), &uint32_tmp,
+ buffer);
+ if (launch_msg_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ safe_unpack32_array((uint32_t **) &
+ (launch_msg_ptr->cpu_count_reps), &uint32_tmp,
+ buffer);
+ if (launch_msg_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ } else {
+ launch_msg_ptr->cpus_per_node = NULL;
+ launch_msg_ptr->cpu_count_reps = NULL;
+ }
+
safe_unpackstr_xmalloc(&launch_msg_ptr->nodes, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&launch_msg_ptr->script, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&launch_msg_ptr->work_dir, &uint16_tmp,
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 3aa75a0b373a3630e24a0632768fcb7d48ed9171..9a47443f49723ee7bca3f536c9c9aedf73fa42e7 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -3,7 +3,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Jette <jette@llnl.gov>, Kevin Tew <tew1@llnl.gov>, et. al.
+ * Written by Morris Jette <jette@llnl.gov>, Kevin Tew <tew1@llnl.gov>, et. al.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -338,7 +338,7 @@ static void _init_config(void)
* Set the (per thread) stack size to a
* more "reasonable" value to avoid running
* out of virtual memory and dying */
- rlim.rlim_cur = 1024 * 1024;
+ rlim.rlim_cur = rlim.rlim_max;
(void) setrlimit(RLIMIT_STACK, &rlim);
}
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 599115011f6e51eeb906f3801ea28bd2d7f09bbc..9793abfd3fd20b9385b942d50ad577fa4a143e83 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -4,7 +4,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Jette <jette1@llnl.gov>
+ * Written by Morris Jette <jette1@llnl.gov>
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -269,6 +269,16 @@ static void _launch_job(struct job_record *job_ptr)
launch_msg_ptr->environment =
get_job_env(job_ptr, &launch_msg_ptr->envc);
+ launch_msg_ptr->num_cpu_groups = job_ptr->num_cpu_groups;
+ launch_msg_ptr->cpus_per_node = xmalloc(sizeof(uint32_t) *
+ job_ptr->num_cpu_groups);
+ memcpy(launch_msg_ptr->cpus_per_node, job_ptr->cpus_per_node,
+ (sizeof(uint32_t) * job_ptr->num_cpu_groups));
+ launch_msg_ptr->cpu_count_reps = xmalloc(sizeof(uint32_t) *
+ job_ptr->num_cpu_groups);
+ memcpy(launch_msg_ptr->cpu_count_reps, job_ptr->cpu_count_reps,
+ (sizeof(uint32_t) * job_ptr->num_cpu_groups));
+
agent_arg_ptr = (agent_arg_t *) xmalloc(sizeof(agent_arg_t));
agent_arg_ptr->node_count = 1;
agent_arg_ptr->retry = 0;
diff --git a/src/slurmd/mgr.c b/src/slurmd/mgr.c
index dc6ffa919b4d6b3046501116d90e6a5d17bf82ae..329a06b96ad295a6ef4feead374f65a197910937 100644
--- a/src/slurmd/mgr.c
+++ b/src/slurmd/mgr.c
@@ -132,13 +132,13 @@ static void _setargs(slurmd_job_t *job);
static void _set_launch_ip_in_env(slurmd_job_t *, slurm_addr *cli);
static void _random_sleep(slurmd_job_t *job);
-
+static char *_sprint_task_cnt(batch_job_launch_msg_t *msg);
/*
* Batch job mangement prototypes:
*/
static char * _make_batch_dir(slurmd_job_t *job);
static char * _make_batch_script(batch_job_launch_msg_t *msg, char *path);
-static int _setup_batch_env(slurmd_job_t *job, char *nodes);
+static int _setup_batch_env(slurmd_job_t *job, batch_job_launch_msg_t *msg);
static int _complete_job(uint32_t jobid, int err, int status);
@@ -205,7 +205,7 @@ mgr_launch_batch_job(batch_job_launch_msg_t *msg, slurm_addr *cli)
if ((job->argv[0] = _make_batch_script(msg, batchdir)) == NULL)
goto cleanup2;
- if ((rc = _setup_batch_env(job, msg->nodes)) < 0)
+ if ((rc = _setup_batch_env(job, msg)) < 0)
goto cleanup2;
status = _job_mgr(job);
@@ -910,10 +910,10 @@ _make_batch_script(batch_job_launch_msg_t *msg, char *path)
}
static int
-_setup_batch_env(slurmd_job_t *job, char *nodes)
+_setup_batch_env(slurmd_job_t *job, batch_job_launch_msg_t *msg)
{
- char buf[1024];
- hostlist_t hl = hostlist_create(nodes);
+ char buf[1024], *task_buf;
+ hostlist_t hl = hostlist_create(msg->nodes);
if (!hl)
return SLURM_ERROR;
@@ -924,9 +924,34 @@ _setup_batch_env(slurmd_job_t *job, char *nodes)
setenvpf(&job->env, "SLURM_NODELIST", "%s", buf);
hostlist_destroy(hl);
+ task_buf = _sprint_task_cnt(msg);
+ setenvpf(&job->env, "SLURM_TASKS_PER_NODE", "%s", task_buf);
+ xfree(task_buf);
+
return 0;
}
+static char *
+_sprint_task_cnt(batch_job_launch_msg_t *msg)
+{
+ int i;
+ char *task_str = xstrdup("");
+ char tmp[16], *comma = "";
+
+ for (i=0; i<msg->num_cpu_groups; i++) {
+ if (i == 1)
+ comma = ",";
+ if (msg->cpu_count_reps[i] > 1)
+ sprintf(tmp, "%s%d(x%d)", comma, msg->cpus_per_node[i],
+ msg->cpu_count_reps[i]);
+ else
+ sprintf(tmp, "%s%d", comma, msg->cpus_per_node[i]);
+ xstrcat(task_str, tmp);
+ }
+
+ return task_str;
+}
+
static void
_send_launch_failure (launch_tasks_request_msg_t *msg, slurm_addr *cli, int rc)
{