From 9da29a38e47767cfdb20d21ca51f4301ae5bb2d5 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 2 Jul 2002 23:54:45 +0000
Subject: [PATCH] The submit API via new comm layer was not functional, it now
is working.
---
src/api/slurm.h | 20 +++------------
src/api/submit.c | 8 +++---
src/common/slurm_protocol_defs.h | 5 ++++
src/common/slurm_protocol_pack.c | 27 ++++++++++++++++++--
src/common/slurm_protocol_pack.h | 3 +++
src/slurmctld/controller.c | 6 ++---
src/slurmctld/job_scheduler.c | 14 +++++-----
src/slurmctld/node_mgr.c | 4 +--
src/slurmctld/node_scheduler.c | 7 ++---
src/slurmctld/slurmctld.h | 3 +--
testsuite/slurm_unit/api/manual/submit-tst.c | 26 +++++++++++++++----
11 files changed, 74 insertions(+), 49 deletions(-)
diff --git a/src/api/slurm.h b/src/api/slurm.h
index 589cf1fb233..0c9c078c85d 100644
--- a/src/api/slurm.h
+++ b/src/api/slurm.h
@@ -129,23 +129,9 @@ extern int slurm_load_node (time_t update_time, node_info_msg_t **node_info_msg_
*/
extern int slurm_load_partitions (time_t update_time, partition_info_msg_t **part_buffer_ptr);
-/*
- * slurm_submit - submit/queue a job with supplied contraints.
- * input: spec - specification of the job's constraints
- * job_id - place to store id of submitted job
- * output: job_id - the job's id
- * returns 0 if no error, EINVAL if the request is invalid
- * NOTE: required specification include: Script=<script_path_name>
- * User=<uid>
- * NOTE: optional specifications include: Contiguous=<YES|NO>
- * Distribution=<BLOCK|CYCLE> Features=<features> Groups=<groups>
- * JobId=<id> JobName=<name> Key=<key> MinProcs=<count>
- * MinRealMemory=<MB> MinTmpDisk=<MB> Partition=<part_name>
- * Priority=<integer> ProcsPerTask=<count> ReqNodes=<node_list>
- * Shared=<YES|NO> TimeLimit=<minutes> TotalNodes=<count>
- * TotalProcs=<count> Immediate=<YES|NO>
- */
-extern int slurm_submit_batch_job (job_desc_msg_t * job_desc_msg );
+/* slurm_submit_job - load the supplied node information buffer if changed */
+extern int slurm_submit_batch_job (job_desc_msg_t * job_desc_msg,
+ submit_response_msg_t ** slurm_alloc_msg );
/*
* slurm_will_run - determine if a job would execute immediately
diff --git a/src/api/submit.c b/src/api/submit.c
index 484a39db272..d5d06977cfd 100644
--- a/src/api/submit.c
+++ b/src/api/submit.c
@@ -20,7 +20,7 @@
/* slurm_submit_job - load the supplied node information buffer if changed */
int
-slurm_submit_batch_job (job_desc_msg_t * job_desc_msg )
+slurm_submit_batch_job (job_desc_msg_t * job_desc_msg, submit_response_msg_t ** slurm_alloc_msg )
{
int msg_size ;
int rc ;
@@ -28,7 +28,6 @@ slurm_submit_batch_job (job_desc_msg_t * job_desc_msg )
slurm_msg_t request_msg ;
slurm_msg_t response_msg ;
return_code_msg_t * slurm_rc_msg ;
- resource_allocation_response_msg_t * slurm_aloc_resp_msg;
/* init message connection for message communication with controller */
if ( ( sockfd = slurm_open_controller_conn ( ) ) == SLURM_SOCKET_ERROR )
@@ -55,9 +54,8 @@ slurm_submit_batch_job (job_desc_msg_t * job_desc_msg )
return (int) slurm_rc_msg->return_code ;
break ;
case RESPONSE_SUBMIT_BATCH_JOB:
- slurm_aloc_resp_msg = ( resource_allocation_response_msg_t * ) response_msg . data ;
- job_desc_msg->job_id = slurm_aloc_resp_msg->job_id;
- return 0;
+ *slurm_alloc_msg = ( submit_response_msg_t * ) response_msg . data ;
+ return SLURM_SUCCESS;
break;
default:
return SLURM_UNEXPECTED_MSG_ERROR ;
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index b959cd8ef92..4cfa822b712 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -215,6 +215,11 @@ typedef struct resource_allocation_response_msg
int32_t* cpu_count_reps;
} resource_allocation_response_msg_t ;
+typedef struct submit_response_msg
+{
+ uint32_t job_id;
+} submit_response_msg_t ;
+
typedef struct job_desc_msg { /* Job descriptor for submit, allocate, and update requests */
uint16_t contiguous; /* 1 if job requires contiguous nodes, 0 otherwise,
* default=0 */
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index efe4e03695f..4f942e54821 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -82,6 +82,9 @@ int pack_msg ( slurm_msg_t const * msg , char ** buffer , uint32_t * buf_len )
case REQUEST_RECONFIGURE :
/* Message contains no body/information */
break ;
+ case RESPONSE_SUBMIT_BATCH_JOB:
+ pack_submit_response_msg ( ( submit_response_msg_t * ) msg -> data , ( void ** ) buffer , buf_len ) ;
+ break ;
case RESPONSE_RESOURCE_ALLOCATION :
case RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION :
case RESPONSE_JOB_WILL_RUN :
@@ -145,7 +148,7 @@ int pack_msg ( slurm_msg_t const * msg , char ** buffer , uint32_t * buf_len )
pack_return_code ( ( return_code_msg_t * ) msg -> data , ( void ** ) buffer , buf_len ) ;
break;
default :
- debug ( "No pack method for msg type %i", msg -> msg_type ) ;
+ error ( "No pack method for msg type %i", msg -> msg_type ) ;
return EINVAL ;
break;
@@ -196,12 +199,14 @@ int unpack_msg ( slurm_msg_t * msg , char ** buffer , uint32_t * buf_len )
case REQUEST_RECONFIGURE :
/* Message contains no body/information */
break ;
+ case RESPONSE_SUBMIT_BATCH_JOB :
+ unpack_submit_response_msg ( ( submit_response_msg_t ** ) & ( msg -> data ) , ( void ** ) buffer , buf_len ) ;
+ break ;
case RESPONSE_RESOURCE_ALLOCATION :
case RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION :
case RESPONSE_JOB_WILL_RUN :
unpack_resource_allocation_response_msg ( ( resource_allocation_response_msg_t ** ) & ( msg -> data ) , ( void ** ) buffer , buf_len ) ;
break ;
-
case REQUEST_UPDATE_NODE :
unpack_update_node_msg ( ( update_node_msg_t ** ) & ( msg-> data ) , ( void ** ) buffer , buf_len ) ;
@@ -351,6 +356,24 @@ int unpack_resource_allocation_response_msg ( resource_allocation_response_msg_t
return 0 ;
}
+void pack_submit_response_msg ( submit_response_msg_t * msg, void ** buffer , int * length )
+{
+ pack32 ( msg->job_id , ( void ** ) buffer , length ) ;
+}
+
+int unpack_submit_response_msg ( submit_response_msg_t ** msg , void ** buffer , int * length )
+{
+ submit_response_msg_t * tmp_ptr ;
+ /* alloc memory for structure */
+ tmp_ptr = xmalloc ( sizeof ( submit_response_msg_t ) ) ;
+ if (tmp_ptr == NULL)
+ return ENOMEM;
+
+ /* load the data values */
+ unpack32 ( & tmp_ptr -> job_id , ( void ** ) buffer , length ) ;
+ *msg = tmp_ptr ;
+ return 0 ;
+}
void pack_node_info_msg ( slurm_msg_t * msg, void ** buf_ptr , int * buffer_size )
{
assert ( msg != NULL );
diff --git a/src/common/slurm_protocol_pack.h b/src/common/slurm_protocol_pack.h
index 8a392ea33a7..a48bb6c5291 100644
--- a/src/common/slurm_protocol_pack.h
+++ b/src/common/slurm_protocol_pack.h
@@ -61,6 +61,9 @@ int unpack_node_table ( node_table_msg_t * node , void ** buf_ptr , int * buffer
void pack_resource_allocation_response_msg ( resource_allocation_response_msg_t * msg, void ** buffer , int * length );
int unpack_resource_allocation_response_msg ( resource_allocation_response_msg_t ** msg , void ** buffer , int * length );
+void pack_submit_response_msg ( submit_response_msg_t * msg, void ** buffer , int * length );
+int unpack_submit_response_msg ( submit_response_msg_t ** msg , void ** buffer , int * length );
+
void pack_update_node_msg ( update_node_msg_t * msg, void ** buffer , uint32_t * length );
int unpack_update_node_msg ( update_node_msg_t ** msg , void ** buffer , uint32_t * length );
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index ba990bfee8a..843d15cb2cf 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -448,7 +448,7 @@ slurm_rpc_submit_batch_job ( slurm_msg_t * msg )
struct job_record *job_rec_ptr;
uint32_t job_id ;
slurm_msg_t response_msg ;
- job_id_msg_t job_id_msg ;
+ submit_response_msg_t submit_msg ;
job_desc_msg_t * job_desc_msg = ( job_desc_msg_t * ) msg-> data ;
start_time = clock ();
@@ -469,9 +469,9 @@ slurm_rpc_submit_batch_job ( slurm_msg_t * msg )
info ("slurmctld_req: job_submit success for id=%u, time=%ld",
job_id, (long) (clock () - start_time));
/* send job_ID */
- job_id_msg . job_id = job_id ;
+ submit_msg . job_id = job_id ;
response_msg . msg_type = RESPONSE_SUBMIT_BATCH_JOB ;
- response_msg . data = & job_id_msg ;
+ response_msg . data = & submit_msg ;
slurm_send_node_msg ( msg->conn_fd , & response_msg ) ;
}
schedule();
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 1cc34faec2d..e10231427bf 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -107,22 +107,22 @@ schedule()
}
if (j < failed_part_cnt) continue;
error_code = select_nodes(job_ptr, 0);
- if (error_code == EAGAIN) {
+ if (error_code == ESLURM_NODES_BUSY) {
xrealloc(failed_parts,
(failed_part_cnt+1)*sizeof(struct part_record *));
failed_parts[failed_part_cnt++] = job_ptr->part_ptr;
}
- else if (error_code == EINVAL) {
+ else if (error_code == SLURM_SUCCESS) { /* job initiated */
+ last_job_update = time (NULL);
+ info ("schedule: job_id %u on nodes %s",
+ job_ptr->job_id, job_ptr->nodes);
+ }
+ else {
last_job_update = time (NULL);
job_ptr->job_state = JOB_FAILED;
job_ptr->start_time = job_ptr->end_time = time(NULL);
delete_job_details(job_ptr);
}
- else { /* job initiated */
- last_job_update = time (NULL);
- info ("schedule: job_id %u on nodes %s",
- job_ptr->job_id, job_ptr->nodes);
- }
}
if (failed_parts)
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 66d950c1bac..c3fd4acb93b 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -250,11 +250,10 @@ main (int argc, char *argv[])
* input: bitmap - bitmap pointer
* node_list - place to put node list
* output: node_list - set to node list or NULL on error
- * returns 0 if no error, errno otherwise
* globals: node_record_table_ptr - pointer to node table
* NOTE: the caller must xfree the memory at node_list when no longer required
*/
-int
+void
bitmap2node_name (bitstr_t *bitmap, char **node_list)
{
int node_list_size, i;
@@ -344,7 +343,6 @@ bitmap2node_name (bitstr_t *bitmap, char **node_list)
strcat (node_list[0], last_suffix);
}
xrealloc (node_list[0], strlen (node_list[0]) + 1);
- return 0;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 99e15f9326d..702e44e55cf 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -698,6 +698,7 @@ select_nodes (struct job_record *job_ptr, int test_only)
struct part_record *part_ptr;
int tmp_feature, check_node_config;
+ error_code = SLURM_SUCCESS;
req_bitmap = scratch_bitmap = NULL;
config_record_iterator = (ListIterator) NULL;
node_set_ptr = NULL;
@@ -849,11 +850,7 @@ select_nodes (struct job_record *job_ptr, int test_only)
}
/* assign the nodes and stage_in the job */
- error_code = bitmap2node_name (req_bitmap, &(job_ptr->nodes));
- if (error_code) {
- error ("bitmap2node_name error %d", error_code);
- goto cleanup;
- }
+ bitmap2node_name (req_bitmap, &(job_ptr->nodes));
build_node_list (req_bitmap,
&job_ptr->details->node_list,
&job_ptr->details->total_procs);
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index b26c5a5ab76..b430c70f18a 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -203,11 +203,10 @@ extern void allocate_nodes (unsigned *bitmap);
* input: bitmap - bitmap pointer
* node_list - place to put node list
* output: node_list - set to node list or null on error
- * returns 0 if no error, otherwise einval or enomem
* NOTE: consider returning the node list as a regular expression if helpful
* NOTE: the caller must free memory at node_list when no longer required
*/
-extern int bitmap2node_name (bitstr_t *bitmap, char **node_list);
+extern void bitmap2node_name (bitstr_t *bitmap, char **node_list);
/*
* block_or_cycle - map string into integer
diff --git a/testsuite/slurm_unit/api/manual/submit-tst.c b/testsuite/slurm_unit/api/manual/submit-tst.c
index bbe93fb134a..2d4201f183c 100644
--- a/testsuite/slurm_unit/api/manual/submit-tst.c
+++ b/testsuite/slurm_unit/api/manual/submit-tst.c
@@ -12,6 +12,7 @@ main (int argc, char *argv[])
{
int error_code, i, count;
job_desc_msg_t job_mesg;
+ submit_response_msg_t *resp_msg;
slurm_init_job_desc_msg( &job_mesg );
job_mesg. contiguous = 1;
@@ -31,13 +32,13 @@ main (int argc, char *argv[])
job_mesg. num_nodes = 400;
job_mesg. user_id = 1500;
- error_code = slurm_submit_batch_job( &job_mesg );
+ error_code = slurm_submit_batch_job( &job_mesg, &resp_msg );
if (error_code) {
printf ("submit error %d\n", error_code);
return (error_code);
}
else
- printf ("job %u submitted\n", job_mesg.job_id);
+ printf ("job %u submitted\n", resp_msg->job_id);
if (argc > 1)
count = atoi (argv[1]);
@@ -45,14 +46,29 @@ main (int argc, char *argv[])
count = 5;
for (i=0; i<count; i++) {
- job_mesg.job_id = job_mesg.job_id + i;
- error_code = slurm_submit_batch_job( &job_mesg);
+ slurm_init_job_desc_msg( &job_mesg );
+ job_mesg. contiguous = 1;
+ job_mesg. groups = ("students,employee\0");
+ job_mesg. name = ("job01\0");
+ job_mesg. partition_key = NULL;
+ job_mesg. min_procs = 4;
+ job_mesg. min_memory = 1024 + i;
+ job_mesg. min_tmp_disk = 2034 + i;
+ job_mesg. partition = "batch\0";
+ job_mesg. priority = 100 + i;
+ job_mesg. job_script = "/bin/hostname\0";
+ job_mesg. shared = 0;
+ job_mesg. time_limit = 100 + i;
+ job_mesg. num_procs = 1000 + i;
+ job_mesg. num_nodes = 400 + i;
+ job_mesg. user_id = 1500;
+ error_code = slurm_submit_batch_job( &job_mesg, &resp_msg );
if (error_code) {
printf ("submit error %d\n", error_code);
break;
}
else {
- printf ("job %u submitted\n", job_mesg.job_id);
+ printf ("job %u submitted\n", resp_msg->job_id);
}
}
--
GitLab