diff --git a/NEWS b/NEWS
index 6b364ba51b285287e6fd3810dc6e36cd25c0682f..cf475475db943421f5b8bfb6572eea81827ca19c 100644
--- a/NEWS
+++ b/NEWS
@@ -72,8 +72,10 @@ documents those changes that are of interest to users and admins.
-- Fix for checking for a non-existant job when querying steps
-- For job steps with the --exclusive option, base initial time in
exponential back-off be partly based upon the process ID for better
- performance with many job steps started at the same time.
+ performance with many job steps started at the same time.
-- Fix for correct step ordering in sview.
+ -- Support optional argument to srun and salloc --immediate option. Specify
+ timeout value in seconds for job or step to be allocated resources.
* Changes in SLURM 2.0.2
========================
diff --git a/doc/man/man1/salloc.1 b/doc/man/man1/salloc.1
index 04388acc37804372cca7dace566005fa1f4dab8b..51088bf9dd3c0b0efa74442719819c364164cd9e 100644
--- a/doc/man/man1/salloc.1
+++ b/doc/man/man1/salloc.1
@@ -326,10 +326,13 @@ show this help message
.RE
.TP
-\fB\-I\fR, \fB\-\-immediate\fR
-Grab the requested resources immediately, or abort if the resources
-are not currently available. The \fIcommand\fR parameter will not be
-run if the resources are not available.
+\fB\-I\fR, \fB\-\-immediate\fR[=<\fIseconds\fR>]
+exit if resources are not available within the
+time period specified.
+If no argument is given, resources must be available immediately
+for the request to succeed.
+By default, \fB\-\-immediate\fR is off, and the command
+will block until resources become available.
.TP
\fB\-J\fR, \fB\-\-job\-name\fR=<\fIjobname\fR>
diff --git a/doc/man/man1/srun.1 b/doc/man/man1/srun.1
index 940d10144ee61c60f73dd955bd5022b205c46745..3b00b8e71b2d9a2e9ef89d2441cb6de113c01ef9 100644
--- a/doc/man/man1/srun.1
+++ b/doc/man/man1/srun.1
@@ -372,10 +372,12 @@ show this help message
.RE
.TP
-\fB\-I\fR, \fB\-\-immediate\fR
-exit if resources are not immediately available. By default,
-\fB\-\-immediate\fR is off, and
-.B srun
+\fB\-I\fR, \fB\-\-immediate\fR[=<\fIseconds\fR>]
+exit if resources are not available within the
+time period specified.
+If no argument is given, resources must be available immediately
+for the request to succeed.
+By default, \fB\-\-immediate\fR is off, and the command
will block until resources become available.
.TP
diff --git a/src/api/allocate.c b/src/api/allocate.c
index 1c351538551e75d6f5a8677cdf6472916f19d99f..4292e694a3b6346d245af5b53c088bee47fc455d 100644
--- a/src/api/allocate.c
+++ b/src/api/allocate.c
@@ -252,7 +252,7 @@ slurm_allocate_resources_blocking (const job_desc_msg_t *user_req,
/* yes, allocation has been granted */
errno = SLURM_PROTOCOL_SUCCESS;
} else if (!req->immediate) {
- if(resp->error_code != SLURM_SUCCESS)
+ if (resp->error_code != SLURM_SUCCESS)
info("%s", slurm_strerror(resp->error_code));
/* no, we need to wait for a response */
job_id = resp->job_id;
@@ -263,7 +263,7 @@ slurm_allocate_resources_blocking (const job_desc_msg_t *user_req,
timeout);
/* If NULL, we didn't get the allocation in
the time desired, so just free the job id */
- if (resp == NULL && errno != ESLURM_ALREADY_DONE) {
+ if ((resp == NULL) && (errno != ESLURM_ALREADY_DONE)) {
errnum = errno;
slurm_complete_job(job_id, -1);
}
diff --git a/src/salloc/opt.c b/src/salloc/opt.c
index b36b821a2a4d70c67223e18d63ef6f9826e76320..62ca9a98e7ceb2eb04f1668f3c1d195c9b4c303f 100644
--- a/src/salloc/opt.c
+++ b/src/salloc/opt.c
@@ -98,6 +98,7 @@
#define OPT_ACCTG_FREQ 0x0f
#define OPT_CPU_BIND 0x10
#define OPT_MEM_BIND 0x11
+#define OPT_IMMEDIATE 0x12
#define OPT_WCKEY 0x15
/* generic getopt_long flags, integers and *not* valid characters */
@@ -272,7 +273,7 @@ static void _opt_default()
opt.kill_command_signal = SIGTERM;
opt.kill_command_signal_set = false;
- opt.immediate = false;
+ opt.immediate = 0;
opt.overcommit = false;
opt.max_wait = 0;
@@ -331,25 +332,25 @@ struct env_vars {
};
env_vars_t env_vars[] = {
- {"SALLOC_ACCOUNT", OPT_STRING, &opt.account, NULL },
- {"SALLOC_ACCTG_FREQ", OPT_INT, &opt.acctg_freq, NULL },
- {"SALLOC_BELL", OPT_BELL, NULL, NULL },
- {"SALLOC_CONN_TYPE", OPT_CONN_TYPE, NULL, NULL },
- {"SALLOC_CPU_BIND", OPT_CPU_BIND, NULL, NULL },
- {"SALLOC_DEBUG", OPT_DEBUG, NULL, NULL },
- {"SALLOC_EXCLUSIVE", OPT_EXCLUSIVE, NULL, NULL },
- {"SALLOC_GEOMETRY", OPT_GEOMETRY, NULL, NULL },
- {"SALLOC_IMMEDIATE", OPT_BOOL, &opt.immediate, NULL },
- {"SALLOC_JOBID", OPT_JOBID, NULL, NULL },
- {"SALLOC_MEM_BIND", OPT_MEM_BIND, NULL, NULL },
- {"SALLOC_NETWORK", OPT_STRING , &opt.network, NULL },
- {"SALLOC_NO_BELL", OPT_NO_BELL, NULL, NULL },
- {"SALLOC_NO_ROTATE", OPT_NO_ROTATE, NULL, NULL },
- {"SALLOC_OVERCOMMIT", OPT_OVERCOMMIT, NULL, NULL },
- {"SALLOC_PARTITION", OPT_STRING, &opt.partition, NULL },
- {"SALLOC_TIMELIMIT", OPT_STRING, &opt.time_limit_str,NULL },
- {"SALLOC_WAIT", OPT_INT, &opt.max_wait, NULL },
- {"SALLOC_WCKEY", OPT_STRING, &opt.wckey, NULL },
+ {"SALLOC_ACCOUNT", OPT_STRING, &opt.account, NULL },
+ {"SALLOC_ACCTG_FREQ", OPT_INT, &opt.acctg_freq, NULL },
+ {"SALLOC_BELL", OPT_BELL, NULL, NULL },
+ {"SALLOC_CONN_TYPE", OPT_CONN_TYPE, NULL, NULL },
+ {"SALLOC_CPU_BIND", OPT_CPU_BIND, NULL, NULL },
+ {"SALLOC_DEBUG", OPT_DEBUG, NULL, NULL },
+ {"SALLOC_EXCLUSIVE", OPT_EXCLUSIVE, NULL, NULL },
+ {"SALLOC_GEOMETRY", OPT_GEOMETRY, NULL, NULL },
+ {"SALLOC_IMMEDIATE", OPT_IMMEDIATE, NULL, NULL },
+ {"SALLOC_JOBID", OPT_JOBID, NULL, NULL },
+ {"SALLOC_MEM_BIND", OPT_MEM_BIND, NULL, NULL },
+ {"SALLOC_NETWORK", OPT_STRING , &opt.network, NULL },
+ {"SALLOC_NO_BELL", OPT_NO_BELL, NULL, NULL },
+ {"SALLOC_NO_ROTATE", OPT_NO_ROTATE, NULL, NULL },
+ {"SALLOC_OVERCOMMIT", OPT_OVERCOMMIT, NULL, NULL },
+ {"SALLOC_PARTITION", OPT_STRING, &opt.partition, NULL },
+ {"SALLOC_TIMELIMIT", OPT_STRING, &opt.time_limit_str,NULL },
+ {"SALLOC_WAIT", OPT_INT, &opt.max_wait, NULL },
+ {"SALLOC_WCKEY", OPT_STRING, &opt.wckey, NULL },
{NULL, 0, NULL, NULL}
};
@@ -390,8 +391,10 @@ _process_env_var(env_vars_t *e, const char *val)
case OPT_INT:
if (val != NULL) {
*((int *) e->arg) = (int) strtol(val, &end, 10);
- if (!(end && *end == '\0'))
- error("%s=%s invalid. ignoring...", e->var, val);
+ if (!(end && *end == '\0')) {
+ error("%s=%s invalid. ignoring...",
+ e->var, val);
+ }
}
break;
@@ -444,6 +447,14 @@ _process_env_var(env_vars_t *e, const char *val)
e->var, val);
}
break;
+
+ case OPT_IMMEDIATE:
+ if (val)
+ opt.immediate = strtol(val, NULL, 10);
+ else
+ opt.immediate = DEFAULT_IMMEDIATE;
+ break;
+
case OPT_BELL:
opt.bell = BELL_ALWAYS;
break;
@@ -518,7 +529,7 @@ void set_options(const int argc, char **argv)
{"geometry", required_argument, 0, 'g'},
{"help", no_argument, 0, 'h'},
{"hold", no_argument, 0, 'H'},
- {"immediate", no_argument, 0, 'I'},
+ {"immediate", optional_argument, 0, 'I'},
{"job-name", required_argument, 0, 'J'},
{"no-kill", no_argument, 0, 'k'},
{"kill-command", optional_argument, 0, 'K'},
@@ -658,7 +669,10 @@ void set_options(const int argc, char **argv)
opt.hold = true;
break;
case 'I':
- opt.immediate = true;
+ if (optarg)
+ opt.immediate = strtol(optarg, NULL, 10);
+ else
+ opt.immediate = DEFAULT_IMMEDIATE;
break;
case 'J':
xfree(opt.job_name);
@@ -1265,16 +1279,6 @@ static bool _opt_verify(void)
opt.time_limit = INFINITE;
}
- if (opt.immediate) {
- char *sched_name = slurm_get_sched_type();
- if (strcmp(sched_name, "sched/wiki") == 0) {
- info("WARNING: Ignoring the -I/--immediate option "
- "(not supported by Maui)");
- opt.immediate = false;
- }
- xfree(sched_name);
- }
-
#ifdef HAVE_AIX
if (opt.network == NULL)
opt.network = "us,sn_all,bulk_xfer";
@@ -1309,6 +1313,11 @@ static bool _opt_verify(void)
opt.ntasks_per_node);
}
+ if (opt.max_wait) {
+ /* FIXME: Eliminate max_wait in slurm v2.1 */
+ opt.immediate = MAX(opt.immediate, opt.max_wait);
+ }
+
return verified;
}
@@ -1536,7 +1545,10 @@ static void _opt_list()
if(opt.distribution == SLURM_DIST_PLANE)
info("plane size : %u", opt.plane_size);
info("verbose : %d", opt.verbose);
- info("immediate : %s", tf_(opt.immediate));
+ if (opt.immediate <= 1)
+ info("immediate : %s", tf_(opt.immediate));
+ else
+ info("immediate : %d secs", (opt.immediate - 1));
info("overcommit : %s", tf_(opt.overcommit));
if (opt.time_limit == INFINITE)
info("time_limit : INFINITE");
@@ -1610,7 +1622,7 @@ 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] [--overcommit] [-D path]\n"
+" [--immediate[=secs]] [--no-kill] [--overcommit] [-D path]\n"
" [--share] [-J jobname] [--jobid=id]\n"
" [--verbose] [--gid=group] [--uid=user] [--licenses=names]\n"
" [-W sec] [--minsockets=n] [--mincores=n] [--minthreads=n]\n"
@@ -1650,7 +1662,7 @@ static void _help(void)
" --get-user-env used by Moab. See srun man page.\n"
" --gid=group_id group ID to run job as (user root only)\n"
" -H, --hold submit job in held state\n"
-" -I, --immediate exit if resources are not immediately available\n"
+" -I, --immediate[=secs] exit if resources not available in \"secs\"\n"
" --jobid=id specify jobid to use\n"
" -J, --job-name=jobname name of job\n"
" -k, --no-kill do not kill job on node failure\n"
diff --git a/src/salloc/opt.h b/src/salloc/opt.h
index ffddb92c3f095e873acc6ab370e2c7f7a6f39227..3175f0d48423de9f489ba2619532cdb4341fcacc 100644
--- a/src/salloc/opt.h
+++ b/src/salloc/opt.h
@@ -34,15 +34,16 @@
# include "config.h"
#endif
-#include <time.h>
#include <sys/types.h>
+#include <time.h>
#include <unistd.h>
-#include "src/common/macros.h" /* true and false */
#include "src/common/env.h"
+#include "src/common/macros.h" /* true and false */
-#define MAX_USERNAME 9
-#define DEFAULT_BELL_DELAY 10
+#define DEFAULT_IMMEDIATE 1
+#define MAX_USERNAME 9
+#define DEFAULT_BELL_DELAY 10
typedef enum {BELL_NEVER, BELL_AFTER_DELAY, BELL_ALWAYS} bell_flag_t;
@@ -92,7 +93,7 @@ typedef struct salloc_options {
char *account; /* --account, -U acct_name */
char *comment; /* --comment */
- int immediate; /* -i, --immediate */
+ int immediate; /* -I, --immediate */
bool hold; /* --hold, -H */
bool no_kill; /* --no-kill, -k */
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 6ecce218e3d8199d6f0ec6d528a2abfb398f6db7..68fee32342973c78fdb59219ea67643e52396491 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -193,7 +193,8 @@ int main(int argc, char *argv[])
callbacks.user_msg = _user_msg_handler;
callbacks.node_fail = _node_fail_handler;
/* create message thread to handle pings and such from slurmctld */
- msg_thr = slurm_allocation_msg_thr_create(&desc.other_port, &callbacks);
+ msg_thr = slurm_allocation_msg_thr_create(&desc.other_port,
+ &callbacks);
xsignal(SIGHUP, _signal_while_allocating);
xsignal(SIGINT, _signal_while_allocating);
@@ -204,7 +205,7 @@ int main(int argc, char *argv[])
xsignal(SIGUSR2, _signal_while_allocating);
before = time(NULL);
- while ((alloc = slurm_allocate_resources_blocking(&desc, opt.max_wait,
+ while ((alloc = slurm_allocate_resources_blocking(&desc, opt.immediate,
_pending_callback)) == NULL) {
if ((errno != ESLURM_ERROR_ON_DESC_TO_RECORD_COPY) ||
(retries >= MAX_RETRIES))
@@ -222,12 +223,15 @@ int main(int argc, char *argv[])
} else if (errno == EINTR) {
error("Interrupted by signal."
" Allocation request rescinded.");
+ } else if ((errno == ETIMEDOUT) && opt.immediate) {
+ error("Unable to allocate resources: %s",
+ slurm_strerror(ESLURM_NODES_BUSY));
} else {
error("Failed to allocate resources: %m");
}
slurm_allocation_msg_thr_destroy(msg_thr);
exit(1);
- } else if(!allocation_interrupted) {
+ } else if (!allocation_interrupted) {
/*
* Allocation granted!
*/
@@ -382,7 +386,8 @@ static int _fill_job_desc_from_opts(job_desc_msg_t *desc)
{
desc->contiguous = opt.contiguous ? 1 : 0;
desc->features = opt.constraints;
- desc->immediate = opt.immediate ? 1 : 0;
+ if (opt.immediate == 1)
+ desc->immediate = 1;
desc->name = xstrdup(opt.job_name);
desc->reservation = xstrdup(opt.reservation);
desc->wckey = xstrdup(opt.wckey);
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index f596f9961c8583a4fdb770fa80a0971739add788..d41dd84fea1babe44b8142eb33942d1c0b7fe7b5 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -177,7 +177,7 @@ static void _node_fail_handler(srun_node_fail_msg_t *msg)
-static bool _retry()
+static bool _retry(void)
{
static int retries = 0;
static char *msg = "Slurm controller not responding, "
@@ -194,10 +194,16 @@ static bool _retry()
} else if (errno == EINTR) {
/* srun may be interrupted by the BLCR checkpoint signal */
/*
- * XXX: this will cause the old job cancelled and a new job allocated
+ * XXX: this will cause the old job cancelled and a new
+ * job allocated
*/
- debug("Syscall interrupted while allocating resources, retrying.");
+ debug("Syscall interrupted while allocating resources, "
+ "retrying.");
return true;
+ } else if ((errno == ETIMEDOUT) && opt.immediate) {
+ error("Unable to allocate resources: %s",
+ slurm_strerror(ESLURM_NODES_BUSY));
+ return false;
} else {
error("Unable to allocate resources: %m");
return false;
@@ -350,14 +356,14 @@ allocate_nodes(void)
job_desc_msg_t *j = job_desc_msg_create_from_opts();
slurm_allocation_callbacks_t callbacks;
- if(!j)
+ if (!j)
return NULL;
/* Do not re-use existing job id when submitting new job
* from within a running job */
if ((j->job_id != NO_VAL) && !opt.jobid_set) {
info("WARNING: Creating SLURM job allocation from within "
- "another allocation");
+ "another allocation");
info("WARNING: You are attempting to initiate a second job");
if (!opt.jobid_set) /* Let slurmctld set jobid */
j->job_id = NO_VAL;
@@ -380,7 +386,7 @@ allocate_nodes(void)
xsignal(SIGUSR2, _signal_while_allocating);
while (!resp) {
- resp = slurm_allocate_resources_blocking(j, 0,
+ resp = slurm_allocate_resources_blocking(j, opt.immediate,
_set_pending_job_id);
if (destroy_job) {
/* cancelled by signal */
@@ -390,7 +396,7 @@ allocate_nodes(void)
}
}
- if(resp && !destroy_job) {
+ if (resp && !destroy_job) {
/*
* Allocation granted!
*/
@@ -518,7 +524,8 @@ job_desc_msg_create_from_opts ()
j->contiguous = opt.contiguous;
j->features = opt.constraints;
- j->immediate = opt.immediate;
+ if (opt.immediate == 1)
+ j->immediate = opt.immediate;
if (opt.job_name)
j->name = xstrdup(opt.job_name);
else
@@ -529,7 +536,7 @@ job_desc_msg_create_from_opts ()
j->req_nodes = xstrdup(opt.nodelist);
/* simplify the job allocation nodelist,
- not laying out tasks until step */
+ * not laying out tasks until step */
if(j->req_nodes) {
hl = hostlist_create(j->req_nodes);
hostlist_ranged_string(hl, sizeof(buf), buf);
@@ -696,6 +703,7 @@ create_job_step(srun_job_t *job, bool use_all_cpus)
int i, rc;
SigFunc *oquitf = NULL, *ointf = NULL, *otermf = NULL;
unsigned long my_sleep = 0;
+ time_t begin_time;
slurm_step_ctx_params_t_init(&job->ctx_params);
job->ctx_params.job_id = job->jobid;
@@ -721,7 +729,8 @@ create_job_step(srun_job_t *job, bool use_all_cpus)
job->ctx_params.ckpt_interval = (uint16_t)opt.ckpt_interval;
job->ctx_params.ckpt_dir = opt.ckpt_dir;
job->ctx_params.exclusive = (uint16_t)opt.exclusive;
- job->ctx_params.immediate = (uint16_t)opt.immediate;
+ if (opt.immediate == 1)
+ job->ctx_params.immediate = (uint16_t)opt.immediate;
if (opt.time_limit != NO_VAL)
job->ctx_params.time_limit = (uint32_t)opt.time_limit;
job->ctx_params.verbose_level = (uint16_t)_verbose;
@@ -767,9 +776,10 @@ create_job_step(srun_job_t *job, bool use_all_cpus)
debug("cpus %u, tasks %u, name %s, relative %u",
job->ctx_params.cpu_count, job->ctx_params.task_count,
job->ctx_params.name, job->ctx_params.relative);
+ begin_time = time(NULL);
for (i=0; (!destroy_job); i++) {
- if(opt.no_alloc) {
+ if (opt.no_alloc) {
job->step_ctx = slurm_step_ctx_create_no_alloc(
&job->ctx_params, job->stepid);
} else
@@ -783,7 +793,8 @@ create_job_step(srun_job_t *job, bool use_all_cpus)
}
rc = slurm_get_errno();
- if (opt.immediate ||
+ if (((opt.immediate != 0) &&
+ (difftime(time(NULL), begin_time) > opt.immediate)) ||
((rc != ESLURM_NODES_BUSY) && (rc != ESLURM_PORTS_BUSY) &&
(rc != ESLURM_PROLOG_RUNNING) &&
(rc != ESLURM_DISABLED))) {
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 72941235468751cbbfa90166a151569a99ea10f4..e0c92c3a094b4f2b6cf6cfeef73b9143d1ee8c39 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -98,6 +98,7 @@
#define OPT_NONE 0x00
#define OPT_INT 0x01
#define OPT_STRING 0x02
+#define OPT_IMMEDIATE 0x03
#define OPT_DISTRIB 0x04
#define OPT_NODES 0x05
#define OPT_OVERCOMMIT 0x06
@@ -380,7 +381,7 @@ static void _opt_default()
opt.no_kill = false;
opt.kill_bad_exit = false;
- opt.immediate = false;
+ opt.immediate = 0;
opt.join = false;
opt.max_wait = slurm_get_wait_time();
@@ -481,7 +482,7 @@ env_vars_t env_vars[] = {
{"SLURM_DEPENDENCY", OPT_STRING, &opt.dependency, NULL },
{"SLURM_DISTRIBUTION", OPT_DISTRIB, NULL, NULL },
{"SLURM_GEOMETRY", OPT_GEOMETRY, NULL, NULL },
-{"SLURM_IMMEDIATE", OPT_INT, &opt.immediate, NULL },
+{"SLURM_IMMEDIATE", OPT_IMMEDIATE, NULL, NULL },
{"SLURM_JOB_NAME", OPT_STRING, &opt.job_name,
&opt.job_name_set_env},
{"SLURM_JOB_ID", OPT_INT, &opt.jobid, NULL },
@@ -652,6 +653,13 @@ _process_env_var(env_vars_t *e, const char *val)
}
break;
+ case OPT_IMMEDIATE:
+ if (val)
+ opt.immediate = strtol(val, NULL, 10);
+ else
+ opt.immediate = DEFAULT_IMMEDIATE;
+ break;
+
case OPT_MPI:
if (mpi_hook_client_init((char *)val) == SLURM_ERROR) {
fatal("\"%s=%s\" -- invalid MPI type, "
@@ -710,7 +718,7 @@ static void set_options(const int argc, char **argv)
{"geometry", required_argument, 0, 'g'},
{"hold", no_argument, 0, 'H'},
{"input", required_argument, 0, 'i'},
- {"immediate", no_argument, 0, 'I'},
+ {"immediate", optional_argument, 0, 'I'},
{"join", no_argument, 0, 'j'},
{"job-name", required_argument, 0, 'J'},
{"no-kill", no_argument, 0, 'k'},
@@ -876,8 +884,10 @@ static void set_options(const int argc, char **argv)
opt.cwd = xstrdup(optarg);
break;
case (int)'e':
- if (opt.pty)
- fatal("--error incompatable with --pty option");
+ if (opt.pty) {
+ fatal("--error incompatable with --pty "
+ "option");
+ }
xfree(opt.efname);
if (strncasecmp(optarg, "none", (size_t) 4) == 0)
opt.efname = xstrdup("/dev/null");
@@ -895,8 +905,10 @@ static void set_options(const int argc, char **argv)
opt.hold = true;
break;
case (int)'i':
- if (opt.pty)
- fatal("--input incompatable with --pty option");
+ if (opt.pty) {
+ fatal("--input incompatable with "
+ "--pty option");
+ }
xfree(opt.ifname);
if (strncasecmp(optarg, "none", (size_t) 4) == 0)
opt.ifname = xstrdup("/dev/null");
@@ -904,7 +916,10 @@ static void set_options(const int argc, char **argv)
opt.ifname = xstrdup(optarg);
break;
case (int)'I':
- opt.immediate = true;
+ if (optarg)
+ opt.immediate = strtol(optarg, NULL, 10);
+ else
+ opt.immediate = DEFAULT_IMMEDIATE;
break;
case (int)'j':
opt.join = true;
@@ -1814,7 +1829,8 @@ static bool _opt_verify(void)
if (opt.ckpt_interval_str) {
opt.ckpt_interval = time_str2mins(opt.ckpt_interval_str);
- if ((opt.ckpt_interval < 0) && (opt.ckpt_interval != INFINITE)) {
+ if ((opt.ckpt_interval < 0) &&
+ (opt.ckpt_interval != INFINITE)) {
error("Invalid checkpoint interval specification");
exit(1);
}
@@ -1829,16 +1845,6 @@ static bool _opt_verify(void)
if ((opt.egid != (gid_t) -1) && (opt.egid != opt.gid))
opt.gid = opt.egid;
- if (opt.immediate) {
- char *sched_name = slurm_get_sched_type();
- if (strcmp(sched_name, "sched/wiki") == 0) {
- info("WARNING: Ignoring the -I/--immediate option "
- "(not supported by Maui)");
- opt.immediate = false;
- }
- xfree(sched_name);
- }
-
if (slurm_verify_cpu_bind(NULL, &opt.cpu_bind,
&opt.cpu_bind_type))
exit(1);
@@ -2022,7 +2028,10 @@ static void _opt_list()
info("core format : %s", core_format_name (opt.core_type));
info("verbose : %d", _verbose);
info("slurmd_debug : %d", opt.slurmd_debug);
- info("immediate : %s", tf_(opt.immediate));
+ if (opt.immediate <= 1)
+ info("immediate : %s", tf_(opt.immediate));
+ else
+ info("immediate : %d secs", (opt.immediate - 1));
info("label output : %s", tf_(opt.labelio));
info("unbuffered IO : %s", tf_(opt.unbuffered));
info("overcommit : %s", tf_(opt.overcommit));
@@ -2122,7 +2131,7 @@ static void _usage(void)
printf(
"Usage: srun [-N nnodes] [-n ntasks] [-i in] [-o out] [-e err]\n"
" [-c ncpus] [-r n] [-p partition] [--hold] [-t minutes]\n"
-" [-D path] [--immediate] [--overcommit] [--no-kill]\n"
+" [-D path] [--immediate[=secs]] [--overcommit] [--no-kill]\n"
" [--share] [--label] [--unbuffered] [-m dist] [-J jobname]\n"
" [--jobid=id] [--verbose] [--slurmd_debug=#]\n"
" [--core=type] [-T threads] [-W sec] [--checkpoint=time]\n"
@@ -2177,7 +2186,7 @@ static void _help(void)
" --get-user-env used by Moab. See srun man page.\n"
" -H, --hold submit job in held state\n"
" -i, --input=in location of stdin redirection\n"
-" -I, --immediate exit if resources are not immediately available\n"
+" -I, --immediate[=secs] exit if resources not available in \"secs\"\n"
" --jobid=id run under already allocated job\n"
" -J, --job-name=jobname name of job\n"
" -k, --no-kill do not kill job on node failure\n"
diff --git a/src/srun/opt.h b/src/srun/opt.h
index fc12cb42d3f37e62255d42cf169246817cf73917..adabce9233eb5db382cd5d5f38f08d2be7f78db3 100644
--- a/src/srun/opt.h
+++ b/src/srun/opt.h
@@ -54,8 +54,9 @@
#include "src/common/env.h"
#include "src/srun/fname.h"
-#define MAX_THREADS 60
-#define MAX_USERNAME 9
+#define DEFAULT_IMMEDIATE 1
+#define MAX_THREADS 60
+#define MAX_USERNAME 9
#define INT_UNASSIGNED ((int)-1)
@@ -64,14 +65,15 @@ extern int _verbose;
extern enum modes mode;
-#define format_task_dist_states(t) (t == SLURM_DIST_BLOCK) ? "block" : \
- (t == SLURM_DIST_CYCLIC) ? "cyclic" : \
- (t == SLURM_DIST_PLANE) ? "plane" : \
- (t == SLURM_DIST_CYCLIC_CYCLIC) ? "cyclic:cyclic" : \
- (t == SLURM_DIST_CYCLIC_BLOCK) ? "cyclic:block" : \
- (t == SLURM_DIST_BLOCK_CYCLIC) ? "block:cyclic" : \
- (t == SLURM_DIST_BLOCK_BLOCK) ? "block:block" : \
- (t == SLURM_DIST_ARBITRARY) ? "arbitrary" : \
+#define format_task_dist_states(t) \
+ (t == SLURM_DIST_BLOCK) ? "block" : \
+ (t == SLURM_DIST_CYCLIC) ? "cyclic" : \
+ (t == SLURM_DIST_PLANE) ? "plane" : \
+ (t == SLURM_DIST_CYCLIC_CYCLIC) ? "cyclic:cyclic" : \
+ (t == SLURM_DIST_CYCLIC_BLOCK) ? "cyclic:block" : \
+ (t == SLURM_DIST_BLOCK_CYCLIC) ? "block:cyclic" : \
+ (t == SLURM_DIST_BLOCK_BLOCK) ? "block:block" : \
+ (t == SLURM_DIST_ARBITRARY) ? "arbitrary" : \
"unknown"
typedef struct srun_options {
@@ -146,7 +148,7 @@ typedef struct srun_options {
/*int verbose;*/ /* -v, --verbose */
/*int debug;*/ /* -d, --debug */
- int immediate; /* -i, --immediate */
+ int immediate; /* -I, --immediate=secs */
bool hold; /* --hold, -H */
bool labelio; /* --label-output, -l */