From be5880e4db5e66e15c41a4860c402377f4ec9828 Mon Sep 17 00:00:00 2001
From: Mark Grondona <mgrondona@llnl.gov>
Date: Thu, 20 May 2004 23:14:14 +0000
Subject: [PATCH] o job_desc_msg_create() ->
job_desc_msg_create_from_options() o Use job_desc_msg_create_from_options()
to build job_desc_msg for batch jobs (less places to update in future)
---
src/srun/allocate.c | 23 +++++++++-
src/srun/allocate.h | 6 +--
src/srun/env.c | 11 +++++
src/srun/env.h | 5 +++
src/srun/launch.c | 13 +-----
src/srun/srun.c | 102 +++++++++-----------------------------------
6 files changed, 62 insertions(+), 98 deletions(-)
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 57f67a66ecc..b63c6624590 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -67,7 +67,7 @@ allocate_nodes(void)
SigFunc *oquitf, *ointf, *otermf;
sigset_t oset;
resource_allocation_response_msg_t *resp = NULL;
- job_desc_msg_t *j = job_desc_msg_create();
+ job_desc_msg_t *j = job_desc_msg_create_from_opts (NULL);
oquitf = xsignal(SIGQUIT, _intr_handler);
ointf = xsignal(SIGINT, _intr_handler);
@@ -214,8 +214,9 @@ _intr_handler(int signo)
* (see opt.h)
*/
job_desc_msg_t *
-job_desc_msg_create(void)
+job_desc_msg_create_from_opts (const char *script)
{
+ extern char **environ;
job_desc_msg_t *j = xmalloc(sizeof(*j));
slurm_init_job_desc_msg(j);
@@ -265,6 +266,24 @@ job_desc_msg_create(void)
else
j->host = NULL;
+ if (script) {
+ /*
+ * If script is set then we are building a request for
+ * a batch job
+ */
+ xassert (opt.batch);
+
+ j->environment = environ;
+ j->env_size = envcount (environ);
+ j->script = script;
+ j->argv = remote_argv;
+ j->argc = remote_argc;
+ j->err = opt.efname;
+ j->in = opt.ifname;
+ j->out = opt.ofname;
+ j->work_dir = opt.cwd;
+ }
+
return (j);
}
diff --git a/src/srun/allocate.h b/src/srun/allocate.h
index b3fa5e19315..4b8da6d0fc4 100644
--- a/src/srun/allocate.h
+++ b/src/srun/allocate.h
@@ -44,16 +44,16 @@ resource_allocation_response_msg_t * allocate_nodes(void);
/*
* Create a job_desc_msg_t object, filled in from the current srun options
- * (see opt.h)
+ * (see opt.h), if script != NULL then this is a batch job.
* The resulting memory must be freed with job_desc_msg_destroy()
*/
-job_desc_msg_t * job_desc_msg_create(void);
+job_desc_msg_t * job_desc_msg_create_from_opts (const char *script);
/*
* Destroy (free memory from) a job_desc_msg_t object allocated with
* job_desc_msg_create()
*/
-void job_desc_msg_destroy(job_desc_msg_t *j);
+void job_desc_msg_destroy (job_desc_msg_t *j);
/*
* Check for SLURM_JOBID environment variable, and if it is a valid
diff --git a/src/srun/env.c b/src/srun/env.c
index 21b60b0c05d..96305a117c0 100644
--- a/src/srun/env.c
+++ b/src/srun/env.c
@@ -33,3 +33,14 @@ setenvf(const char *fmt, ...)
return putenv(bufcpy);
}
+/*
+ * Return the number of elements in the environment `env'
+ */
+int
+envcount (const char **env)
+{
+ int envc = 0;
+ while (env[envc] != NULL)
+ envc++;
+ return (envc);
+}
diff --git a/src/srun/env.h b/src/srun/env.h
index ede76189f53..4d74bc5a92a 100644
--- a/src/srun/env.h
+++ b/src/srun/env.h
@@ -14,4 +14,9 @@
*/
int setenvf(const char *fmt, ...);
+/*
+ * Return the number of elements in the environment array `env'
+ */
+int envcount (char **env);
+
#endif /* _HAVE_ENV_H */
diff --git a/src/srun/launch.c b/src/srun/launch.c
index 0d53e984d61..1f1600c7ace 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -43,6 +43,7 @@
#include "src/srun/job.h"
#include "src/srun/launch.h"
#include "src/srun/opt.h"
+#include "src/srun/env.h"
extern char **environ;
@@ -73,7 +74,6 @@ 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,
char * hostname);
-static int _envcount(char **env);
int
launch_thr_create(job_t *job)
@@ -109,7 +109,7 @@ launch(void *arg)
msg_array_ptr =
xmalloc(sizeof(launch_tasks_request_msg_t)*job->nhosts);
req_array_ptr = xmalloc(sizeof(slurm_msg_t) * job->nhosts);
- my_envc = _envcount(environ);
+ my_envc = envcount(environ);
for (i = 0; i < job->nhosts; i++) {
launch_tasks_request_msg_t *r = &msg_array_ptr[i];
slurm_msg_t *m = &req_array_ptr[i];
@@ -445,12 +445,3 @@ _print_launch_msg(launch_tasks_request_msg_t *msg, char * hostname)
debug3("uid:%ld cwd:%s %d",
(long) msg->uid, msg->cwd, msg->srun_node_id);
}
-
-static int
-_envcount(char **environ)
-{
- int envc = 0;
- while (environ[envc] != NULL)
- envc++;
- return envc;
-}
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 52128f78179..026baa520b0 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -345,10 +345,10 @@ _run_batch_job(void)
{
int file_type, retries;
int rc = SLURM_SUCCESS;
- job_desc_msg_t job;
+ job_desc_msg_t *req;
submit_response_msg_t *resp;
- extern char **environ;
- char *job_script;
+ char *script;
+ void (*log_msg) (const char *fmt, ...) = (void (*)) &error;
if ((remote_argc == 0) || (remote_argv[0] == NULL))
return SLURM_ERROR;
@@ -361,100 +361,38 @@ _run_batch_job(void)
* }
*/
- job_script = _build_script (remote_argv[0], file_type);
- if (job_script == NULL) {
+ if ((script = _build_script (remote_argv[0], file_type)) == NULL) {
error ("unable to build script from file %s", remote_argv[0]);
return SLURM_ERROR;
}
- slurm_init_job_desc_msg(&job);
-
- job.contiguous = opt.contiguous;
- job.features = opt.constraints;
-
- job.name = opt.job_name;
-
- job.partition = opt.partition;
-
- if (opt.hold)
- job.priority = 0;
- if (opt.mincpus > -1)
- job.min_procs = opt.mincpus;
- if (opt.realmem > -1)
- job.min_memory = opt.realmem;
- if (opt.tmpdisk > -1)
- job.min_tmp_disk = opt.tmpdisk;
-
- job.req_nodes = opt.nodelist;
- job.exc_nodes = opt.exc_nodes;
-
- if (opt.overcommit)
- job.num_procs = opt.min_nodes;
- else
- job.num_procs = opt.nprocs * opt.cpus_per_task;
-
- job.min_nodes = opt.min_nodes;
- if (opt.max_nodes)
- job.max_nodes = opt.max_nodes;
-
- job.num_tasks = opt.nprocs;
-
- job.user_id = opt.uid;
- job.group_id = getgid();
-
- if (opt.hold)
- job.priority = 0;
- if (opt.no_kill)
- job.kill_on_node_fail = 0;
- if (opt.time_limit > -1)
- job.time_limit = opt.time_limit;
- if (opt.share)
- job.shared = 1;
-
- /* _set_batch_script_env(job); */
- job.environment = environ;
-
- job.env_size = 0;
- while (environ[job.env_size] != NULL)
- job.env_size++;
-
- job.script = job_script;
- job.argv = remote_argv;
- job.argc = remote_argc;
- job.err = opt.efname;
- job.in = opt.ifname;
- job.out = opt.ofname;
- job.work_dir = opt.cwd;
+ if (!(req = job_desc_msg_create_from_opts (script)))
+ fatal ("Unable to create job request");
retries = 0;
- while ((rc = slurm_submit_batch_job(&job, &resp)) < 0) {
- if ( (errno == ESLURM_ERROR_ON_DESC_TO_RECORD_COPY)
- && (retries < MAX_RETRIES) ) {
- if (retries == 0)
- error ("Slurm controller not responding, "
- "sleeping and retrying");
- else
- debug ("Slurm controller not responding, "
- "sleeping and retrying");
+ while ( (retries < MAX_RETRIES)
+ && (rc = slurm_submit_batch_job(req, &resp)) < 0) {
- sleep (++retries);
- }
- else {
- error("Unable to submit batch job resources: %s",
- slurm_strerror(errno));
- return SLURM_ERROR;
- }
+ if (errno != ESLURM_ERROR_ON_DESC_TO_RECORD_COPY)
+ return (error("Unable to submit batch job: %m"));
+
+ (*log_msg) ("Controller not responding, retrying...");
+ log_msg = &debug;
+ sleep (++retries);
}
if (rc == SLURM_SUCCESS) {
- info("jobid %u submitted",resp->job_id);
+ info ("jobid %u submitted",resp->job_id);
if (resp->error_code)
info("Warning: %s", slurm_strerror(resp->error_code));
slurm_free_submit_response_response_msg (resp);
}
- xfree (job_script);
- return rc;
+
+ job_desc_msg_destroy (req);
+ xfree (script);
+
+ return (rc);
}
/* _get_shell - return a string containing the default shell for this user
--
GitLab