From 184eb1733631f227bdc1a6baf4dcab946aa4f9eb Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 27 Sep 2002 19:04:02 +0000
Subject: [PATCH] Fix some batch job submit bugs.
---
src/slurmctld/agent.c | 38 +++++++++++++++++++++++++++++------
src/slurmctld/job_mgr.c | 19 +++++++++++++-----
src/slurmctld/job_scheduler.c | 5 ++---
src/slurmctld/slurmctld.h | 1 +
4 files changed, 49 insertions(+), 14 deletions(-)
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 6837952e11a..466520a202f 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -104,6 +104,7 @@ typedef struct task_info {
static void alarm_handler(int dummy);
static void queue_agent_retry (agent_info_t *agent_info_ptr, int count);
+static void slurmctld_free_job_launch_msg(batch_job_launch_msg_t * msg);
static void spawn_retry_agent (agent_arg_t *agent_arg_ptr);
static void *thread_per_node_rpc (void *args);
static void *wdog (void *args);
@@ -141,7 +142,8 @@ agent (void *args)
fatal ("agent passed NULL node name list");
if ((agent_arg_ptr->msg_type != REQUEST_REVOKE_JOB_CREDENTIAL) &&
(agent_arg_ptr->msg_type != REQUEST_NODE_REGISTRATION_STATUS) &&
- (agent_arg_ptr->msg_type != REQUEST_PING))
+ (agent_arg_ptr->msg_type != REQUEST_PING) &&
+ (agent_arg_ptr->msg_type != REQUEST_BATCH_JOB_LAUNCH))
fatal ("agent passed invalid message type %d", agent_arg_ptr->msg_type);
/* initialize the data structures */
@@ -239,8 +241,12 @@ cleanup:
xfree (agent_arg_ptr->slurm_addr);
if (agent_arg_ptr->node_names)
xfree (agent_arg_ptr->node_names);
- if (agent_arg_ptr->msg_args)
- xfree (agent_arg_ptr->msg_args);
+ if (agent_arg_ptr->msg_args) {
+ if (agent_arg_ptr->msg_type == REQUEST_BATCH_JOB_LAUNCH)
+ slurmctld_free_job_launch_msg (agent_arg_ptr->msg_args);
+ else
+ xfree (agent_arg_ptr->msg_args);
+ }
xfree (agent_arg_ptr);
}
#endif
@@ -437,13 +443,13 @@ thread_per_node_rpc (void *args)
rc = slurm_rc_msg->return_code;
slurm_free_return_code_msg ( slurm_rc_msg );
if (rc)
- error ("thread_per_node_rpc/rc error %d", rc);
+ error ("thread_per_node_rpc/rc error %s",
+ slurm_strerror (rc));
else {
debug3 ("agent sucessfully processed RPC to node %s",
thread_ptr->node_name);
- thread_state = DSH_DONE;
}
-
+ thread_state = DSH_DONE;
break ;
default:
error ("thread_per_node_rpc bad msg_type %d",response_msg.msg_type);
@@ -608,3 +614,23 @@ spawn_retry_agent (agent_arg_t *agent_arg_ptr)
fatal ("pthread_create error %m");
}
}
+
+/* slurmctld_free_job_launch_msg is a variant of slurm_free_job_launch_msg
+ * because all environment variables currently loaded in one xmalloc
+ * buffer (see get_job_env()), which is different from how slurmd
+ * assembles the data from a message */
+
+void slurmctld_free_job_launch_msg(batch_job_launch_msg_t * msg)
+{
+ if (msg) {
+ if (msg->environment) {
+ if (msg->environment[0])
+ xfree(msg->environment[0]);
+
+ xfree(msg->environment);
+ msg->environment = NULL;
+ }
+ slurm_free_job_launch_msg (msg);
+ }
+}
+
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 5881a30144e..4f84265587d 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -361,9 +361,11 @@ dump_job_details_state (struct job_details *detail_ptr, void **buf_ptr, int *buf
pack32 ((uint32_t) detail_ptr->num_procs, buf_ptr, buf_len);
pack32 ((uint32_t) detail_ptr->num_nodes, buf_ptr, buf_len);
+
pack16 ((uint16_t) detail_ptr->shared, buf_ptr, buf_len);
pack16 ((uint16_t) detail_ptr->contiguous, buf_ptr, buf_len);
pack16 ((uint16_t) detail_ptr->kill_on_node_fail, buf_ptr, buf_len);
+ pack16 ((uint16_t) detail_ptr->batch_flag, buf_ptr, buf_len);
pack32 ((uint32_t) detail_ptr->min_procs, buf_ptr, buf_len);
pack32 ((uint32_t) detail_ptr->min_memory, buf_ptr, buf_len);
@@ -466,7 +468,7 @@ load_job_state ( void )
uint16_t job_state, next_step_id, details;
char *nodes = NULL, *partition = NULL, *name = NULL;
uint32_t num_procs, num_nodes, min_procs, min_memory, min_tmp_disk, submit_time;
- uint16_t shared, contiguous, kill_on_node_fail, name_len;
+ uint16_t shared, contiguous, kill_on_node_fail, name_len, batch_flag;
char *req_nodes = NULL, *features = NULL;
char *stderr = NULL, *stdin = NULL, *stdout = NULL, *work_dir = NULL;
slurm_job_credential_t *credential_ptr = NULL;
@@ -536,9 +538,11 @@ load_job_state ( void )
safe_unpack32 (&num_procs, &buf_ptr, &buffer_size);
safe_unpack32 (&num_nodes, &buf_ptr, &buffer_size);
+
safe_unpack16 (&shared, &buf_ptr, &buffer_size);
safe_unpack16 (&contiguous, &buf_ptr, &buffer_size);
safe_unpack16 (&kill_on_node_fail, &buf_ptr, &buffer_size);
+ safe_unpack16 (&batch_flag, &buf_ptr, &buffer_size);
safe_unpack32 (&min_procs, &buf_ptr, &buffer_size);
safe_unpack32 (&min_memory, &buf_ptr, &buffer_size);
@@ -617,7 +621,7 @@ load_job_state ( void )
job_ptr->details->shared = shared;
job_ptr->details->contiguous = contiguous;
job_ptr->details->kill_on_node_fail = kill_on_node_fail;
- job_ptr->details->kill_on_node_fail = 1;
+ job_ptr->details->batch_flag = batch_flag;
job_ptr->details->min_procs = min_procs;
job_ptr->details->min_memory = min_memory;
job_ptr->details->min_tmp_disk = min_tmp_disk;
@@ -1275,10 +1279,15 @@ job_create ( job_desc_msg_t *job_desc, uint32_t *new_job_id, int allocate,
goto cleanup ;
}
- if ( ( error_code = copy_job_desc_to_file ( job_desc , (*job_rec_ptr)->job_id ) ) ) {
- error_code = ESLURM_WRITING_TO_FILE ;
- goto cleanup ;
+ if (job_desc->script) {
+ if ( ( error_code = copy_job_desc_to_file ( job_desc , (*job_rec_ptr)->job_id ) ) ) {
+ error_code = ESLURM_WRITING_TO_FILE ;
+ goto cleanup ;
+ }
+ (*job_rec_ptr)->details->batch_flag = 1;
}
+ else
+ (*job_rec_ptr)->details->batch_flag = 0;
if (part_ptr->shared == SHARED_FORCE) /* shared=force */
(*job_rec_ptr)->details->shared = 1;
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 29f6661194d..864e7a1d092 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -208,7 +208,7 @@ launch_job (struct job_record *job_ptr)
pthread_attr_t attr_agent;
pthread_t thread_agent;
-/* if (job_ptr->details->batch_flag == 0) */
+ if (job_ptr->details->batch_flag == 0)
return;
node_ptr = find_first_node_record (job_ptr -> node_bitmap);
@@ -235,10 +235,9 @@ launch_job (struct job_record *job_ptr)
agent_arg_ptr -> slurm_addr = xmalloc (sizeof (struct sockaddr_in));
memcpy (agent_arg_ptr -> slurm_addr,
&(node_ptr -> slurm_addr), sizeof (struct sockaddr_in));
- agent_arg_ptr -> node_names = node_ptr -> name;
+ agent_arg_ptr -> node_names = xstrdup (node_ptr -> name);
agent_arg_ptr -> msg_type = REQUEST_BATCH_JOB_LAUNCH;
agent_arg_ptr -> msg_args = (void *)launch_msg_ptr;
-/* FIXME: Agent must perform full data structure cleanup for launch_msg_ptr */
/* Launch the RPC via agent */
debug3 ("Spawning job launch agent for job_id %u", job_ptr -> job_id);
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 0f321c41861..d7c73171897 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -171,6 +171,7 @@ extern int job_count; /* number of jobs in the system */
/* job_details - specification of a job's constraints */
struct job_details {
uint32_t magic; /* magic cookie to test data integrity */
+ uint16_t batch_flag; /* 1 if batch job (with script) */
uint32_t num_procs; /* minimum number of processors */
uint32_t num_nodes; /* minimum number of nodes */
char *req_nodes; /* required nodes */
--
GitLab