diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index d8c79133d8ad0ea8c68951455626480206f42941..f27c137cfd80e6d5055a71e42f749aec4dee6740 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -98,27 +98,17 @@ allocate_nodes(void)
return resp;
}
-/* Return jobid from environment
- *
+/*
* Returns jobid if SLURM_JOBID was set in the user's environment
- * else returns 0
+ * or if --jobid option was given, else returns 0
*/
uint32_t
jobid_from_env(void)
{
- char *p, *q;
- uint32_t jobid;
-
- if (!(p = getenv("SLURM_JOBID")))
- return 0;
-
- jobid = (uint32_t) strtoul(p, &q, 10);
- if (*q != '\0') {
- error ("Invalid value for SLURM_JOBID: `%s'", p);
- return 0;
- }
-
- return jobid;
+ if (opt.jobid != NO_VAL)
+ return ((uint32_t) opt.jobid);
+ else
+ return (0);
}
resource_allocation_response_msg_t *
diff --git a/src/srun/job.c b/src/srun/job.c
index 57d8ae95cfbd276ac6a5483bd262a02d03fe3bf5..c4082d7f0ca6324fd542b3e8188e5bc18950be4c 100644
--- a/src/srun/job.c
+++ b/src/srun/job.c
@@ -648,8 +648,8 @@ _relative_hosts(hostlist_t hl)
*/
n = hostlist_count(rlist);
if (n < opt.min_nodes) {
- info("Warning: Only %d node(s) beyond relative location "
- "specified, resetting nnodes to %d", n, n);
+ info("Warning: Only %d node%s available in relative set, "
+ "resetting nnodes to %d", n, (n>1 ? "s":""), n);
opt.min_nodes = n;
}
diff --git a/src/srun/launch.c b/src/srun/launch.c
index ac337f33d77ce8472768aabd1a540470e6afa232..d70d555fcdd9a060312fc93cf5bc66bfd6c726d9 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -370,7 +370,7 @@ _send_msg_rc(slurm_msg_t *msg)
int rc = 0;
int errnum = 0;
- if ((rc = slurm_send_recv_rc_msg(msg, &errnum, 0)) < 0)
+ if ((rc = slurm_send_recv_rc_msg(msg, &errnum, opt.msg_timeout)) < 0)
return SLURM_ERROR;
if (errnum != 0)
diff --git a/src/srun/msg.c b/src/srun/msg.c
index 09beab07f69612246cfd931ae57f81363921e753..355b1e0b864ca4d815f0c012b4e8eb0cab403233 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -100,59 +100,29 @@ static void _print_pid_list(const char *host, int ntasks,
#ifdef HAVE_TOTALVIEW
-/* Convert node name to address string, eg. "123.45.67.8",
- * also return the index in the job table (-1 if not found) */
-static char *
-_node_name_to_addr(const char *name, job_t *job, int *inx)
-{
- int i;
- char *buf = xmalloc(28);
- char *colon;
-
- for (i=0; i<job->nhosts; i++) {
- if (strcmp(name, job->host[i]))
- continue;
- slurm_print_slurm_addr(&job->slurmd_addr[i], buf, 128);
- /* This returns address:port, we need to remove ":port" */
- colon = strchr(buf, (int)':');
- if (colon)
- colon[0] = '\0';
- *inx = i;
- return buf;
- }
-
- error("_node_name_to_addr error on %s", name);
- *inx = -1;
- return NULL;
-}
-
+/*
+ * Install entry in the MPI_proctable for host with node id `nodeid'
+ * and the number of tasks `ntasks' with pid array `pid'
+ */
static void
-_build_tv_list(job_t *job, char *host, int ntasks, uint32_t *pid)
+_build_tv_list(job_t *job, char *host, int nodeid, int ntasks, uint32_t *pid)
{
- MPIR_PROCDESC * tv_tasks;
- int i, node_inx, task_id;
- char *node_addr;
+ int i;
static int tasks_recorded = 0;
- node_addr = _node_name_to_addr(host, job, &node_inx);
- if ((node_addr == NULL) || (node_inx < 0))
- return;
-
if (MPIR_proctable_size == 0) {
MPIR_proctable_size = opt.nprocs;
MPIR_proctable = xmalloc(sizeof(MPIR_PROCDESC) * opt.nprocs);
}
for (i = 0; i < ntasks; i++) {
+ int taskid = job->tids[nodeid][i];
+ MPIR_PROCDESC *tv = &MPIR_proctable[taskid];
+ tv->host_name = job->host[nodeid];
+ tv->executable_name = remote_argv[0];
+ tv->pid = pid[i];
+
tasks_recorded++;
- task_id = job->tids[node_inx][i];
- tv_tasks = &MPIR_proctable[task_id];
- tv_tasks->host_name = node_addr;
- tv_tasks->executable_name = remote_argv[0];
- tv_tasks->pid = pid[i];
- debug("task=%d host=%s executable=%s pid=%d", task_id,
- tv_tasks->host_name, tv_tasks->executable_name,
- tv_tasks->pid);
}
if (tasks_recorded == opt.nprocs) {
@@ -185,26 +155,21 @@ static bool _job_msg_done(job_t *job)
static void
_process_launch_resp(job_t *job, launch_tasks_response_msg_t *msg)
{
- if ((msg->srun_node_id >= 0)
- && (msg->srun_node_id < job->nhosts)) {
+ if ((msg->srun_node_id < 0) || (msg->srun_node_id >= job->nhosts)) {
+ error ("Bad launch response from %s", msg->node_name);
+ return;
+ }
- pthread_mutex_lock(&job->task_mutex);
- job->host_state[msg->srun_node_id] = SRUN_HOST_REPLIED;
- pthread_mutex_unlock(&job->task_mutex);
+ pthread_mutex_lock(&job->task_mutex);
+ job->host_state[msg->srun_node_id] = SRUN_HOST_REPLIED;
+ pthread_mutex_unlock(&job->task_mutex);
#ifdef HAVE_TOTALVIEW
- _build_tv_list( job, msg->node_name, msg->count_of_pids,
- msg->local_pids );
+ _build_tv_list( job, msg->node_name, msg->srun_node_id,
+ msg->count_of_pids, msg->local_pids );
#endif
- _print_pid_list( msg->node_name, msg->count_of_pids,
- msg->local_pids, remote_argv[0] );
+ _print_pid_list( msg->node_name, msg->count_of_pids,
+ msg->local_pids, remote_argv[0] );
- } else {
- error("launch resp from %s has bad task id %d",
- msg->node_name, msg->srun_node_id);
-#ifdef HAVE_TOTALVIEW
- tv_launch_failure();
-#endif
- }
}
static void
@@ -349,7 +314,8 @@ _reattach_handler(job_t *job, slurm_msg_t *msg)
resp->executable_name = NULL; /* nothing left to free */
remote_argv[1] = NULL;
}
- _build_tv_list(job, resp->node_name, resp->ntasks, resp->local_pids);
+ _build_tv_list(job, resp->node_name, resp->srun_node_id,
+ resp->ntasks, resp->local_pids );
#endif
_print_pid_list(resp->node_name, resp->ntasks, resp->local_pids,
resp->executable_name);
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 2ab38ac2a398032d36e7733c19a377a29e227a63..a9f165bc3ddc76fbf132a8e2b34a4ae18db1e8f6 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -106,6 +106,7 @@
#define OPT_OVERCOMMIT 0x1a
#define OPT_HOLD 0x1b
#define OPT_RELATIVE 0x1c
+#define OPT_JOBID 0x1d
/* constraint type options */
#define OPT_MINCPUS 0x50
@@ -223,6 +224,8 @@ struct poptOption runTable[] = {
{"job-name", 'J', POPT_ARG_STRING, &opt.job_name, 0,
"name of job",
"jobname"},
+ {"jobid", '\0', POPT_ARG_INT, NULL, OPT_JOBID,
+ "run under already allocated job", "id" },
{"output", 'o', POPT_ARG_STRING, 0, OPT_OUTPUT,
"location of stdout redirection",
"out"},
@@ -245,10 +248,12 @@ struct poptOption runTable[] = {
{"wait", 'W', POPT_ARG_INT, &opt.max_wait, OPT_WAIT,
"seconds to wait after first task ends before killing job",
"sec"},
- {"max-launch-time", '\0', POPT_ARG_INT | POPT_ARGFLAG_DOC_HIDDEN,
- &opt.max_launch_time, 0, NULL, NULL },
+ {"max-launch-time", '\0', POPT_ARG_INT | POPT_ARGFLAG_DOC_HIDDEN,
+ &opt.max_launch_time, 0, NULL, NULL },
{"max-exit-timeout", '\0', POPT_ARG_INT | POPT_ARGFLAG_DOC_HIDDEN,
&opt.max_exit_timeout, 0, NULL, NULL },
+ {"msg-timeout", '\0', POPT_ARG_INT | POPT_ARGFLAG_DOC_HIDDEN,
+ &opt.msg_timeout, 0, NULL, NULL },
POPT_TABLEEND
};
@@ -565,6 +570,7 @@ static void _opt_default()
opt.max_threads = MAX_THREADS;
opt.job_name = NULL;
+ opt.jobid = NO_VAL;
opt.distribution = SRUN_DIST_UNKNOWN;
@@ -580,9 +586,6 @@ static void _opt_default()
opt.batch = false;
opt.share = false;
opt.no_kill = false;
-#ifdef HAVE_TOTALVIEW
- opt.totalview = _under_totalview();
-#endif
opt.immediate = false;
@@ -606,9 +609,22 @@ static void _opt_default()
opt.exc_nodes = NULL;
opt.max_launch_time = 60; /* 60 seconds to launch job */
opt.max_exit_timeout= 60; /* Warn user 60 seconds after task exit */
+ opt.msg_timeout = 2; /* Default launch msg timeout */
mode = MODE_NORMAL;
+#ifdef HAVE_TOTALVIEW
+ /*
+ * Reset some default values if running under TotalView:
+ */
+ if ((opt.totalview = _under_totalview())) {
+ opt.max_launch_time = 120;
+ opt.max_threads = 1;
+ opt.msg_timeout = 15;
+ }
+
+#endif
+
}
/*---[ env var processing ]-----------------------------------------------*/
@@ -630,6 +646,7 @@ struct env_vars {
};
env_vars_t env_vars[] = {
+ {"SLURM_JOBID", OPT_INT, &opt.jobid, NULL },
{"SLURM_NPROCS", OPT_INT, &opt.nprocs, &opt.nprocs_set},
{"SLURM_CPUS_PER_TASK", OPT_INT, &opt.cpus_per_task, &opt.cpus_set },
{"SLURM_PARTITION", OPT_STRING, &opt.partition, NULL },
@@ -735,7 +752,7 @@ _get_int(const char *arg, const char *what)
char *p;
long int result = strtol(arg, &p, 10);
- if (*p != '\0') {
+ if ((*p != '\0') || (result < 0L)) {
error ("Invalid numeric value \"%s\" for %s.", arg, what);
exit(1);
}
@@ -866,6 +883,10 @@ static void _opt_args(int ac, char **av)
}
break;
+ case OPT_JOBID:
+ opt.jobid = _get_int(arg, "jobid");
+ break;
+
case OPT_REALMEM:
opt.realmem = (int) _to_bytes(arg);
if (opt.realmem < 0) {
diff --git a/src/srun/opt.h b/src/srun/opt.h
index 84c54567f0ebcab29dc4dc296770ea9c8198740d..f3f7521a8672560bbf2030c0c604e6c8f3cef2fd 100644
--- a/src/srun/opt.h
+++ b/src/srun/opt.h
@@ -110,6 +110,7 @@ typedef struct srun_options {
enum distribution_t
distribution; /* --distribution=, -m dist */
char *job_name; /* --job-name=, -J name */
+ unsigned int jobid; /* --jobid=jobid */
char *ofname; /* --output -o filename */
char *ifname; /* --input -i filename */
@@ -151,6 +152,7 @@ typedef struct srun_options {
bool no_alloc; /* --no-allocate, -Z */
int max_launch_time; /* Undocumented */
int max_exit_timeout; /* Undocumented */
+ int msg_timeout; /* Undocumented */
} opt_t;
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 2b6fefe1884e81a67af4b8401ea230b8b31254c9..a72c94d23e91bb5b63fb3f0c0de6a54b01c3385c 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -190,9 +190,12 @@ int main(int ac, char **av)
/* job structure should now be filled in */
/*
- * Set nodelist environment variable
+ * Enhance environment for job
*/
setenvf("SLURM_NODELIST=%s", job->nodelist);
+ setenvf("SLURM_JOBID=%u", job->jobid);
+ setenvf("SLURM_NPROCS=%d", opt.nprocs);
+ setenvf("SLURM_NNODES=%d", job->nhosts);
if (msg_thr_create(job) < 0)
job_fatal(job, "Unable to create msg thread");