From 386397bd04b247432ac06a2e553fc03af656c8a0 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 3 Dec 2002 00:00:31 +0000
Subject: [PATCH] Make srun verbose messages more useful (task numbers by host
reported along with step id)
---
src/srun/io.c | 4 ++--
src/srun/launch.c | 41 ++++++++++++++++++++++++++++++++---------
src/srun/srun.c | 7 +++----
3 files changed, 37 insertions(+), 15 deletions(-)
diff --git a/src/srun/io.c b/src/srun/io.c
index 3544534fe2f..5f7aa0b90b3 100644
--- a/src/srun/io.c
+++ b/src/srun/io.c
@@ -404,7 +404,7 @@ _accept_io_stream(job_t *job, int i)
int len = size_io_stream_header();
int j;
int fd = job->iofd[i];
- verbose("Activity on IO server port %d fd %d", i, fd);
+ debug("Activity on IO server port %d fd %d", i, fd);
for (j = 0; j < 15; i++) {
int sd, size_read;
@@ -470,7 +470,7 @@ _accept_io_stream(job_t *job, int i)
else
job->err[hdr.task_id] = sd;
- verbose("accepted %s connection from %s task %ld, sd=%d",
+ debug("accepted %s connection from %s task %ld, sd=%d",
(hdr.type ? "stderr" : "stdout"),
buf, hdr.task_id, sd );
}
diff --git a/src/srun/launch.c b/src/srun/launch.c
index 986864430e3..2bd9df7f4d0 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -66,7 +66,8 @@ static void _dist_block(job_t *job, uint32_t **task_ids);
static void _dist_cyclic(job_t *job, uint32_t **task_ids);
static void _p_launch(slurm_msg_t *req_array_ptr, job_t *job);
static void * _p_launch_task(void *args);
-static void _print_launch_msg(launch_tasks_request_msg_t *msg);
+static void _print_launch_msg(launch_tasks_request_msg_t *msg,
+ char * hostname);
static int _envcount(char **env);
static void
@@ -261,8 +262,8 @@ static void * _p_launch_task(void *args)
int host_inx = msg_ptr->srun_node_id;
int failure = 0;
- debug3("launching on host %s", job_ptr->host[host_inx]);
- _print_launch_msg(msg_ptr);
+ if (_verbose || _debug)
+ _print_launch_msg(msg_ptr, job_ptr->host[host_inx]);
if (slurm_send_only_node_msg(req_ptr) < 0) { /* Has timeout */
error("task launch error on %s: %m", job_ptr->host[host_inx]);
pthread_mutex_lock(&job_ptr->task_mutex);
@@ -288,13 +289,35 @@ static void * _p_launch_task(void *args)
}
-static void _print_launch_msg(launch_tasks_request_msg_t *msg)
+static void
+_print_launch_msg(launch_tasks_request_msg_t *msg, char * hostname)
{
- debug3("%d.%d uid:%ld n:%ld cwd:%s %d [%d-%d]",
- msg->job_id, msg->job_step_id, (long) msg->uid,
- (long) msg->tasks_to_launch, msg->cwd,
- msg->srun_node_id, msg->global_task_ids[0],
- msg->global_task_ids[msg->tasks_to_launch-1]);
+ int i;
+ char tmp_str[10], task_list[4096];
+
+ if (opt.distribution == SRUN_DIST_BLOCK) {
+ sprintf(task_list, "%u-%u",
+ msg->global_task_ids[0],
+ msg->global_task_ids[(msg->tasks_to_launch-1)]);
+ } else {
+ for (i=0; i<msg->tasks_to_launch; i++) {
+ sprintf(tmp_str, ",%u", msg->global_task_ids[i]);
+ if (i == 0)
+ strcpy(task_list, &tmp_str[1]);
+ else if ((strlen(tmp_str) + strlen(task_list)) <
+ sizeof(task_list))
+ strcat(task_list, tmp_str);
+ else
+ break;
+ }
+ }
+
+ printf("launching %u.%u on host %s, %u tasks: %s\n",
+ msg->job_id, msg->job_step_id, hostname,
+ msg->tasks_to_launch, task_list);
+
+ debug3("uid:%ld cwd:%s %d",
+ (long) msg->uid, msg->cwd, msg->srun_node_id);
}
static int
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 226ad006ce4..c484220c8e2 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -175,18 +175,17 @@ main(int ac, char **av)
if ( !(resp = _allocate_nodes()) )
exit(1);
job = job_create(resp);
- if (_verbose || _debug)
+ if (_debug)
_print_job_information(resp);
_run_job_script(resp->job_id);
slurm_complete_job(resp->job_id, 0, 0);
- if (_verbose || _debug)
- info ("Spawned srun shell terminated");
+ debug ("Spawned srun shell terminated");
exit (0);
} else {
if ( !(resp = _allocate_nodes()) )
exit(1);
- if (_verbose || _debug)
+ if (_debug)
_print_job_information(resp);
job = job_create(resp);
--
GitLab