diff --git a/NEWS b/NEWS
index 7b1fbf19b0398eefe43eba40fe46800dcd40558c..0f2ad26d034e3d5909e8a90aa355e3d781ae78b3 100644
--- a/NEWS
+++ b/NEWS
@@ -46,6 +46,10 @@ documents those changes that are of interest to users and admins.
Moved existing digital signature logic into new plugin: crypto/openssl.
Added new support for crypto/munge (available with GPL license).
+* Changes in SLURM 1.2.16
+=========================
+ -- Add --overcommit option to the salloc command.
+
* Changes in SLURM 1.2.15
=========================
-- In sched/wiki2, fix bug processing hostlist expressions where hosts
diff --git a/doc/man/man1/salloc.1 b/doc/man/man1/salloc.1
index 0d004adac52f41b21be805e8e7671e08eca5fbd6..ba2869dcd81b55fc52d5bb92632a774302cb099d 100644
--- a/doc/man/man1/salloc.1
+++ b/doc/man/man1/salloc.1
@@ -1,4 +1,4 @@
-.TH "salloc" "1" "SLURM 1.2" "December 2006" "SLURM Commands"
+.TH "salloc" "1" "SLURM 1.2" "August 2007" "SLURM Commands"
.SH "NAME"
.LP
salloc \- Obtain a SLURM job allocation (a set of nodes), execute a command, and then release the allocation when the command is finished.
@@ -222,6 +222,13 @@ is presently ignored if SchedulerType=sched/maui.
\fB\-\-no\-bell\fR
Silence salloc's use of the terminal bell. Also see the option \fB\-\-bell\fR.
+.TP
+\fB\-O\fR, \fB\-\-overcommit\fR
+Overcommit resources. Normally, \fBsbatch\fR will allocate one cpu per
+task to be executed. By specifying \fB\-\-overcommit\fR you are explicitly
+allowing more than one process per cpu. However no more than
+\fBMAX_TASKS_PER_NODE\fR tasks are permitted to execute per node.
+
.TP
\fB\-p\fR, \fB\-\-partition\fR[=]<\fIpartition name\fR>
Request a specific partition for the resource allocation. If not specified,
@@ -376,6 +383,9 @@ Same as \fB\-\-no\-bell\fR.
\fBSALLOC_NO_ROTATE\fR
Same as \fB\-R\fR or \fB\-\-no\-rotate\fR.
.TP
+\fBSLURM_OVERCOMMIT\fR
+Same as \fB\-O, \-\-overcommit\fR
+.TP
\fBSALLOC_PARTITION\fR
Same as \fB\-p\fR or \fB\-\-partition\fR.
.TP
diff --git a/doc/man/man1/sbatch.1 b/doc/man/man1/sbatch.1
index 5ee2c5617a9fec033e40c58a629352057fe460ba..76c2c266acf6b4f7a5fa31629f1ac4180a0745ba 100644
--- a/doc/man/man1/sbatch.1
+++ b/doc/man/man1/sbatch.1
@@ -271,6 +271,7 @@ When a job is requeued, the batch script is initiated from its beginning.
Specify the number of tasks to be launched per node.
Equivalent to \fB\-\-tasks\-per\-node\fR.
+.TP
\fB\-O\fR, \fB\-\-overcommit\fR
Overcommit resources. Normally, \fBsbatch\fR will allocate one cpu per
task to be executed. By specifying \fB\-\-overcommit\fR you are explicitly
diff --git a/src/plugins/sched/wiki2/start_job.c b/src/plugins/sched/wiki2/start_job.c
index 6c578bdf17e69504bd0951f4ddb794d4f47f7406..224f05da8f097bafd1ad17e75b1bfee53f2ab59b 100644
--- a/src/plugins/sched/wiki2/start_job.c
+++ b/src/plugins/sched/wiki2/start_job.c
@@ -141,7 +141,6 @@ static int _start_job(uint32_t jobid, int task_cnt, char *hostlist,
size_t node_name_len;
static uint32_t cr_test = 0, cr_enabled = 0;
-info("task_cnt=%d, hostlist=%s, tasklist=%s", task_cnt, hostlist, tasklist);
if (cr_test == 0) {
select_g_get_info_from_plugin(SELECT_CR_PLUGIN,
&cr_enabled);
diff --git a/src/salloc/opt.c b/src/salloc/opt.c
index 245fc959d343865eed6e166609c94d5084394e3d..82e37115133f02ba259e75b78ea05c9dff33a0f8 100644
--- a/src/salloc/opt.c
+++ b/src/salloc/opt.c
@@ -85,6 +85,7 @@
#define OPT_NO_BELL 0x10
#define OPT_JOBID 0x11
#define OPT_EXCLUSIVE 0x12
+#define OPT_OVERCOMMIT 0x13
/* generic getopt_long flags, integers and *not* valid characters */
#define LONG_OPT_JOBID 0x105
@@ -441,6 +442,7 @@ static void _opt_default()
opt.kill_command_signal_set = false;
opt.immediate = false;
+ opt.overcommit = false;
opt.max_wait = 0;
opt.quiet = 0;
@@ -504,6 +506,7 @@ env_vars_t env_vars[] = {
{"SALLOC_BELL", OPT_BELL, NULL, NULL },
{"SALLOC_NO_BELL", OPT_NO_BELL, NULL, NULL },
{"SALLOC_EXCLUSIVE", OPT_EXCLUSIVE, NULL, NULL },
+ {"SALLOC_OVERCOMMIT", OPT_OVERCOMMIT, NULL, NULL },
{NULL, 0, NULL, NULL}
};
@@ -612,6 +615,9 @@ _process_env_var(env_vars_t *e, const char *val)
case OPT_EXCLUSIVE:
opt.shared = 0;
break;
+ case OPT_OVERCOMMIT:
+ opt.overcommit = true;
+ break;
default:
/* do nothing */
break;
@@ -660,6 +666,7 @@ void set_options(const int argc, char **argv)
{"kill-command", optional_argument, 0, 'K'},
{"tasks", required_argument, 0, 'n'},
{"nodes", required_argument, 0, 'N'},
+ {"overcommit", no_argument, 0, 'O'},
{"partition", required_argument, 0, 'p'},
{"quiet", no_argument, 0, 'q'},
{"no-rotate", no_argument, 0, 'R'},
@@ -694,7 +701,7 @@ void set_options(const int argc, char **argv)
{"reboot", no_argument, 0, LONG_OPT_REBOOT},
{NULL, 0, 0, 0}
};
- char *opt_string = "+a:c:C:d:F:g:hHIJ:kK::n:N:p:qR:st:uU:vVw:W:x:";
+ char *opt_string = "+a:c:C:d:F:g:hHIJ:kK::n:N:Op:qR:st:uU:vVw:W:x:";
opt.progname = xbasename(argv[0]);
optind = 0;
@@ -772,6 +779,9 @@ void set_options(const int argc, char **argv)
exit(1);
}
break;
+ case 'O':
+ opt.overcommit = true;
+ break;
case 'p':
xfree(opt.partition);
opt.partition = xstrdup(optarg);
@@ -1247,6 +1257,7 @@ static void _opt_list()
info("jobid : %u", opt.jobid);
info("verbose : %d", opt.verbose);
info("immediate : %s", tf_(opt.immediate));
+ info("overcommit : %s", tf_(opt.overcommit));
if (opt.time_limit == INFINITE)
info("time_limit : INFINITE");
else if (opt.time_limit != NO_VAL)
@@ -1288,7 +1299,7 @@ static void _usage(void)
printf(
"Usage: salloc [-N numnodes|[min nodes]-[max nodes]] [-n num-processors]\n"
" [[-c cpus-per-node] [-r n] [-p partition] [--hold] [-t minutes]\n"
-" [--immediate] [--no-kill]\n"
+" [--immediate] [--no-kill] [--overcommit]\n"
" [--share] [-J jobname] [--jobid=id]\n"
" [--verbose] [--gid=group] [--uid=user]\n"
" [-W sec] [--minsockets=n] [--mincores=n] [--minthreads=n]\n"
@@ -1318,6 +1329,7 @@ static void _help(void)
" -I, --immediate exit if resources are not immediately available\n"
" -k, --no-kill do not kill job on node failure\n"
" -K, --kill-command[=signal] signal to send terminating job\n"
+" -O, --overcommit overcommit resources\n"
" -s, --share share nodes with other jobs\n"
" -J, --job-name=jobname name of job\n"
" --jobid=id specify jobid to use\n"
diff --git a/src/salloc/opt.h b/src/salloc/opt.h
index 53a68871fd6f84061fbec2100f912110a177e4e4..c101b0bcb4d774d17d65d5b21d737b09c973276f 100644
--- a/src/salloc/opt.h
+++ b/src/salloc/opt.h
@@ -78,6 +78,7 @@ typedef struct salloc_options {
bool hold; /* --hold, -H */
bool no_kill; /* --no-kill, -k */
+ bool overcommit; /* --overcommit -O */
int kill_command_signal;/* --kill-command, -K */
bool kill_command_signal_set;
uint16_t shared; /* --share, -s */
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 1a359484f89eab7cda7c14b82e01bf9ac98b8ff9..06da0579f8665bf283306d13f783587621bfbb3d 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -164,6 +164,10 @@ int main(int argc, char *argv[])
/* Add default task count for srun, if not already set */
if (opt.nprocs_set)
env_array_append_fmt(&env, "SLURM_NPROCS", "%d", opt.nprocs);
+ if (opt.overcommit) {
+ env_array_append_fmt(&env, "SLURM_OVERCOMMIT", "%d",
+ opt.overcommit);
+ }
env_array_set_environment(env);
env_array_free(env);
pthread_mutex_lock(&allocation_state_lock);
@@ -284,7 +288,11 @@ static int fill_job_desc_from_opts(job_desc_msg_t *desc)
desc->job_min_memory = opt.realmem;
if (opt.tmpdisk > -1)
desc->job_min_tmp_disk = opt.tmpdisk;
- desc->num_procs = opt.nprocs * opt.cpus_per_task;
+ if (opt.overcommit) {
+ desc->num_procs = opt.min_nodes;
+ desc->overcommit = opt.overcommit;
+ } else
+ desc->num_procs = opt.nprocs * opt.cpus_per_task;
if (opt.nprocs_set)
desc->num_tasks = opt.nprocs;
if (opt.cpus_set)
diff --git a/testsuite/expect/Makefile.am b/testsuite/expect/Makefile.am
index 98309e73bb66e3b245ba88b0faa9885f2e62f0e5..bd9d6e5b33b1ae05794e387e6800eb29851b8f87 100644
--- a/testsuite/expect/Makefile.am
+++ b/testsuite/expect/Makefile.am
@@ -228,6 +228,7 @@ EXTRA_DIST = \
test15.21 \
test15.22 \
test15.23 \
+ test15.24 \
test16.1 \
test16.2 \
test16.3 \
diff --git a/testsuite/expect/Makefile.in b/testsuite/expect/Makefile.in
index de8bbbc2da3fffa53642f5ca8f10d5ffdced3e71..866bb0d7831a77ee8a51c958f4e30df3941249ad 100644
--- a/testsuite/expect/Makefile.in
+++ b/testsuite/expect/Makefile.in
@@ -454,6 +454,7 @@ EXTRA_DIST = \
test15.21 \
test15.22 \
test15.23 \
+ test15.24 \
test16.1 \
test16.2 \
test16.3 \
diff --git a/testsuite/expect/README b/testsuite/expect/README
index fd5de3c8e4d2e40ba62b378e92c441b03a0f3a65..809cc878ca2240998101a5a882fcb05922368a68 100644
--- a/testsuite/expect/README
+++ b/testsuite/expect/README
@@ -381,6 +381,7 @@ test15.22 Test of partition specification on job submission (--partition
option).
test15.23 Test of environment variables that control salloc actions:
SALLOC_ACCOUNT, SALLOC_DEBUG and SALLOC_TIMELIMIT
+test15.24 Test of --overcommit option.
test16.# Testing of sattach options.
diff --git a/testsuite/expect/test17.32 b/testsuite/expect/test17.32
index 032c1dba2a0f571b7a072af1ba7fc0586e4bdaeb..b26af694904dab4fbdf46cd997d6cde1b9ddd5ae 100755
--- a/testsuite/expect/test17.32
+++ b/testsuite/expect/test17.32
@@ -41,7 +41,7 @@ set task_cnt 10
print_header $test_id
#
-# Submit a slurm job that will execute 'id' on 4 tasks (or try anyway)
+# Submit a slurm job that will print slurm env vars and execute 'id'
#
file delete $file_in $file_out
make_bash_script $file_in "
@@ -110,11 +110,6 @@ if {$task_cnt != $tasks} {
set exit_code 1
}
-if {$task_cnt != $tasks} {
- send_user "\nFAILURE: Did not get proper number of tasks: "
- send_user "$task_cnt, $tasks\n"
- set exit_code 1
-}
if {$exit_code == 0} {
file delete $file_in $file_out
send_user "\nSUCCESS\n"