diff --git a/src/api/allocate.c b/src/api/allocate.c
index ef61fcf527a4cd1838b09fcf76bc36ba831aafd8..8fb5cfaf04368c0f222f3343f66bb28cef8631a1 100644
--- a/src/api/allocate.c
+++ b/src/api/allocate.c
@@ -32,7 +32,7 @@ main (int argc, char *argv[])
char *node_list, *job_id;
error_code = slurm_allocate
- ("User=1500 Script=/bin/hostname JobName=job01 TotalNodes=400 TotalProcs=1000 ReqNodes=lx[3000-3003] Partition=batch MinRealMemory=1024 MinTmpDisk=2034 Groups=students,employee MinProcs=4 Contiguous=YES Key=1234",
+ ("User=1500 JobName=job01 TotalNodes=400 TotalProcs=1000 ReqNodes=lx[3000-3003] Partition=batch MinRealMemory=1024 MinTmpDisk=2034 Groups=students,employee MinProcs=4 Contiguous=YES Key=1234",
&node_list, &job_id);
if (error_code)
printf ("allocate error %d\n", error_code);
@@ -44,7 +44,7 @@ main (int argc, char *argv[])
while (1) {
error_code = slurm_allocate
- ("User=1500 Script=/bin/hostname JobName=more TotalProcs=4000 Partition=batch Key=1234 ",
+ ("User=1500 JobName=more TotalProcs=4000 Partition=batch Key=1234 ",
&node_list, &job_id);
if (error_code) {
printf ("allocate error %d\n", error_code);
@@ -59,7 +59,7 @@ main (int argc, char *argv[])
while (1) {
error_code = slurm_allocate
- ("User=1500 Script=/bin/hostname JobName=more TotalProcs=40 Partition=batch Key=1234 ",
+ ("User=1500 JobName=more TotalProcs=40 Partition=batch Key=1234 ",
&node_list, &job_id);
if (error_code) {
printf ("allocate error %d\n", error_code);
@@ -84,7 +84,7 @@ main (int argc, char *argv[])
* output: job_id - node_list - list of allocated nodes
* returns 0 if no error, EINVAL if the request is invalid,
* EAGAIN if the request can not be satisfied at present
- * NOTE: required specifications include: User=<uid> Script=<pathname>
+ * NOTE: required specifications include: User=<uid>
* optional specifications include: Contiguous=<YES|NO>
* Distribution=<BLOCK|CYCLE> Features=<features> Groups=<groups>
* JobId=<id> JobName=<name> Key=<credential> MinProcs=<count>
diff --git a/src/api/node_info.c b/src/api/node_info.c
index 6718800e2f1207236c4538cce1605a14daa193b2..e825248f8346621708266dff6c01a79ead162546 100644
--- a/src/api/node_info.c
+++ b/src/api/node_info.c
@@ -174,7 +174,7 @@ slurm_load_node (time_t update_time, struct node_buffer **node_buffer_ptr)
unpackstr_ptr (&node[i].name, &uint16_tmp, &buf_ptr, &buffer_size);
if (node[i].name == NULL)
node[i].name = "";
- unpack32 (&node[i].node_state, &buf_ptr, &buffer_size);
+ unpack16 (&node[i].node_state, &buf_ptr, &buffer_size);
unpack32 (&node[i].cpus, &buf_ptr, &buffer_size);
unpack32 (&node[i].real_memory, &buf_ptr, &buffer_size);
unpack32 (&node[i].tmp_disk, &buf_ptr, &buffer_size);
diff --git a/src/api/submit.c b/src/api/submit.c
index 5c2097a947ed55dced75a5d65717c9a59ed20274..3c39771653f9291d25f8a59bc6614aa3310f67d1 100644
--- a/src/api/submit.c
+++ b/src/api/submit.c
@@ -68,6 +68,7 @@ main (int argc, char *argv[])
* 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: the caller must free the storage at job_id[0]
* NOTE: required specification include: Script=<script_path_name>
* User=<uid>
* NOTE: optional specifications include: Contiguous=<YES|NO>
@@ -76,7 +77,7 @@ main (int argc, char *argv[])
* MinRealMemory=<MB> MinTmpDisk=<MB> Partition=<part_name>
* Priority=<integer> ProcsPerTask=<count> ReqNodes=<node_list>
* Shared=<YES|NO> TimeLimit=<minutes> TotalNodes=<count>
- * TotalProcs=<count>
+ * TotalProcs=<count> Immediate=<YES|NO>
*/
int
slurm_submit (char *spec, char **job_id)
diff --git a/src/common/slurm.h b/src/common/slurm.h
index 7fc22650b4766cbdc399670e78164214ecdc3cf9..5f0cddf98377014c558fd128e78db7924d08afdf 100644
--- a/src/common/slurm.h
+++ b/src/common/slurm.h
@@ -74,7 +74,7 @@ extern time_t last_node_update; /* time of last update to node records */
struct node_record {
uint32_t magic; /* magic cookie to test data integrity */
char name[MAX_NAME_LEN]; /* name of the node. a null name indicates defunct node */
- int node_state; /* enum node_states, negative if down */
+ uint16_t node_state; /* enum node_states, ORed with STATE_NO_RESPOND if down */
time_t last_response; /* last response from the node */
uint32_t cpus; /* actual count of cpus running on the node */
uint32_t real_memory; /* actual megabytes of real memory on the node */
@@ -398,12 +398,13 @@ extern int job_cancel (char * job_id);
* new_job_id - location for storing new job's id
* output: new_job_id - the job's ID
* returns 0 on success, EINVAL if specification is invalid
+ * allocate - if set, job allocation only (no script required)
* globals: job_list - pointer to global job list
* list_part - global list of partition info
* default_part_loc - pointer to default partition
* NOTE: the calling program must xfree the memory pointed to by new_job_id
*/
-extern int job_create (char *job_specs, char **new_job_id);
+extern int job_create (char *job_specs, char **new_job_id, int allocate);
/* job_lock - lock the job information */
extern void job_lock ();
diff --git a/src/common/slurmlib.h b/src/common/slurmlib.h
index a06f290ae4def5f6ca15658a0eae1b3894d8f0b5..861b40575490336f6e486deafff5bc2ed08cbbfe 100644
--- a/src/common/slurmlib.h
+++ b/src/common/slurmlib.h
@@ -15,6 +15,7 @@
#define PART_STRUCT_VERSION 1
#define SLURMCTLD_HOST "134.9.55.42"
#define SLURMCTLD_PORT 1543
+#define STATE_NO_RESPOND 0x8000
/* INFINITE is used to identify unlimited configurations, */
/* eg. the maximum count of nodes any job may use in some partition */
@@ -38,8 +39,7 @@ enum task_dist {
};
/* last entry must be STATE_END, keep in sync with node_state_string */
-/* any value less than or equal to zero is down. if a node was in state */
-/* STATE_BUSY and stops responding, its state becomes -(STATE_BUSY), etc. */
+/* if a node ceases to respond, its last state is ORed with STATE_NO_RESPOND */
enum node_states {
STATE_DOWN, /* node is not responding */
STATE_UNKNOWN, /* node's initial state, unknown */
@@ -115,7 +115,7 @@ struct job_buffer {
struct node_table {
char *name; /* name of the node. a null name indicates defunct node */
- uint32_t node_state; /* state of the node, see node_states */
+ uint16_t node_state; /* state of the node, see node_states */
uint32_t cpus; /* count of processors running on the node */
uint32_t real_memory; /* megabytes of real memory on the node */
uint32_t tmp_disk; /* megabytes of total disk in TMP_FS */
@@ -137,9 +137,9 @@ struct part_table {
uint32_t max_nodes; /* per job or INFINITE */
uint32_t total_nodes; /* total number of nodes in the partition */
uint32_t total_cpus; /* total number of cpus in the partition */
- uint16_t default_part; /* 1 if if this is default partition */
+ uint16_t default_part; /* 1 if this is default partition */
uint16_t key; /* 1 if slurm distributed key is required for use */
- uint16_t shared; /* 1 if >1 job can share a node, 2 if required */
+ uint16_t shared; /* 1 if job can share nodes, 2 if job must share nodes */
uint16_t state_up; /* 1 if state is up, 0 if down */
char *nodes; /* comma delimited list names of nodes in partition */
int *node_inx; /* list index pairs into node_table:
@@ -161,7 +161,7 @@ struct part_buffer {
* output: job_id - node_list - list of allocated nodes
* returns 0 if no error, EINVAL if the request is invalid,
* EAGAIN if the request can not be satisfied at present
- * NOTE: required specifications include: User=<uid> Script=<pathname>
+ * NOTE: required specifications include: User=<uid>
* optional specifications include: Contiguous=<YES|NO>
* Distribution=<BLOCK|CYCLE> Features=<features> Groups=<groups>
* JobId=<id> JobName=<name> Key=<credential> MinProcs=<count>
@@ -170,6 +170,7 @@ struct part_buffer {
* Shared=<YES|NO> TimeLimit=<minutes> TotalNodes=<count>
* TotalProcs=<count>
* NOTE: the calling function must free the allocated storage at node_list[0]
+ * NOTE: the calling function must free the allocated storage at job_id[0]
*/
extern int slurm_allocate (char *spec, char **node_list, char **job_id);
@@ -269,6 +270,7 @@ extern int slurm_load_part (time_t update_time, struct part_buffer **part_buffer
* 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: the caller must free the storage at job_id[0]
* NOTE: required specification include: Script=<script_path_name>
* User=<uid>
* NOTE: optional specifications include: Contiguous=<YES|NO>
@@ -277,10 +279,29 @@ extern int slurm_load_part (time_t update_time, struct part_buffer **part_buffer
* MinRealMemory=<MB> MinTmpDisk=<MB> Partition=<part_name>
* Priority=<integer> ProcsPerTask=<count> ReqNodes=<node_list>
* Shared=<YES|NO> TimeLimit=<minutes> TotalNodes=<count>
- * TotalProcs=<count>
+ * TotalProcs=<count> Immediate=<YES|NO>
*/
extern int slurm_submit (char *spec, char **job_id);
+/*
+ * slurm_will_run - determine if a job would execute immediately
+ * if submitted.
+ * input: spec - specification of the job's constraints
+ * job_id - place to store id of submitted job
+ * output: returns 0 if job would run now, EINVAL if the request
+ * would never run, EAGAIN if job would run later
+ * NOTE: required specification include: Script=<script_path_name>
+ * User=<uid>
+ * NOTE: optional specifications include: Contiguous=<YES|NO>
+ * Features=<features> Groups=<groups>
+ * Key=<key> MinProcs=<count>
+ * MinRealMemory=<MB> MinTmpDisk=<MB> Partition=<part_name>
+ * Priority=<integer> ReqNodes=<node_list>
+ * Shared=<YES|NO> TimeLimit=<minutes> TotalNodes=<count>
+ * TotalProcs=<count>
+ */
+extern int slurm_will_run (char *spec, char **job_id);
+
/*
* parse_node_name - parse the node name for regular expressions and return a sprintf format
* generate multiple node names as needed.
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 69c914ce9848f9947e001e2c958d1042af97c1f2..15509745a4f28dbda8ce2f65332bced7f476d0b8 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -366,7 +366,7 @@ slurmctld_req (int sockfd) {
/* JobSubmit - submit a job to the slurm queue */
else if (strncmp ("JobSubmit", in_line, 9) == 0) {
time_stamp = NULL;
- error_code = job_create(&in_line[9], &job_id_ptr); /* skip "JobSubmit" */
+ error_code = job_create(&in_line[9], &job_id_ptr, 0); /* skip "JobSubmit" */
if (error_code)
info ("slurmctld_req: job_submit error %d, time=%ld",
error_code, (long) (clock () - start_time));
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 81092fbde86311c140fedbe7c97c60036ea89d59..451b35685c7adf6cbd463da81d51c15b901531f1 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -291,7 +291,7 @@ job_allocate (char *job_specs, char **new_job_id, char **node_list)
new_job_id[0] = node_list[0] = NULL;
- error_code = job_create (job_specs, new_job_id);
+ error_code = job_create (job_specs, new_job_id, 1);
if (error_code)
return error_code;
job_ptr = find_job_record (new_job_id[0]);
@@ -361,13 +361,14 @@ job_cancel (char * job_id)
* new_job_id - location for storing new job's id
* output: new_job_id - the job's ID
* returns 0 on success, EINVAL if specification is invalid
+ * allocate - if set, job allocation only (no script required)
* globals: job_list - pointer to global job list
* list_part - global list of partition info
* default_part_loc - pointer to default partition
* NOTE: the calling program must xfree the memory pointed to by new_job_id
*/
int
-job_create (char *job_specs, char **new_job_id)
+job_create (char *job_specs, char **new_job_id, int allocate)
{
char *req_features, *req_node_list, *job_name, *req_group;
char *req_partition, *script, *job_id;
@@ -408,7 +409,7 @@ job_create (char *job_specs, char **new_job_id)
error_code = EINVAL;
goto cleanup;
}
- if (script == NULL) {
+ if (allocate == 0 && script == NULL) {
info ("job_create: job failed to specify Script");
error_code = EINVAL;
goto cleanup;
@@ -1084,7 +1085,6 @@ reset_job_bitmaps ()
list_iterator_destroy (job_record_iterator);
last_job_update = time (NULL);
- return 0;
}
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 68f5bf42ced1dce37e474e117ea50a4ab7cf7554..12b594ebaf85c03a4633688ca1c2a763111fea9c 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1002,15 +1002,10 @@ pack_all_node (char **buffer_ptr, int *buffer_size, time_t * update_time)
int
pack_node (struct node_record *dump_node_ptr, void **buf_ptr, int *buf_len)
{
- int state;
char *partition = NULL;
- state = dump_node_ptr->node_state;
- if (state < 0)
- state = STATE_DOWN;
-
packstr (dump_node_ptr->name, buf_ptr, buf_len);
- pack32 (state, buf_ptr, buf_len);
+ pack16 (dump_node_ptr->node_state, buf_ptr, buf_len);
pack32 (dump_node_ptr->cpus, buf_ptr, buf_len);
pack32 (dump_node_ptr->real_memory, buf_ptr, buf_len);
pack32 (dump_node_ptr->tmp_disk, buf_ptr, buf_len);
@@ -1171,22 +1166,27 @@ update_node (char *node_names, char *spec)
}
if (state_val != NO_VAL) {
- if ((state_val == STATE_DOWN) &&
- (node_record_point->node_state != STATE_UNKNOWN))
- node_record_point->node_state =
- -(node_record_point->node_state);
- else
- node_record_point->node_state = state_val;
- if (state_val != STATE_IDLE)
+ if (state_val == STATE_DOWN) {
+ bit_clear (up_node_bitmap,
+ (int) (node_record_point -
+ node_record_table_ptr));
bit_clear (idle_node_bitmap,
(int) (node_record_point -
node_record_table_ptr));
- if (state_val == STATE_DOWN)
- bit_clear (up_node_bitmap,
+ }
+ else if (state_val != STATE_IDLE)
+ bit_clear (idle_node_bitmap,
(int) (node_record_point -
node_record_table_ptr));
+ else /* (state_val == STATE_IDLE) */
+ bit_set (idle_node_bitmap,
+ (int) (node_record_point -
+ node_record_table_ptr));
+
+ node_record_point->node_state = state_val;
info ("update_node: node %s state set to %s",
- &node_list[i*MAX_NAME_LEN], node_state_string[state_val]);
+ &node_list[i*MAX_NAME_LEN],
+ node_state_string[state_val]);
}
}
@@ -1250,11 +1250,10 @@ validate_node_specs (char *node_name, uint32_t cpus,
}
else {
info ("validate_node_specs: node %s has registered", node_name);
+ node_ptr->node_state &= (uint16_t) (~STATE_NO_RESPOND);
if ((node_ptr->node_state == STATE_DOWN) ||
(node_ptr->node_state == STATE_UNKNOWN))
node_ptr->node_state = STATE_IDLE;
- if (node_ptr->node_state < 0)
- node_ptr->node_state = -(node_ptr->node_state);
bit_set (up_node_bitmap, (node_ptr - node_record_table_ptr));
}
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 8eeba6917d9649b24feae0bb70128a8020a4011b..d3e1d9d5cd147a455d7e57f4f51f64d8397cdd0c 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -207,7 +207,8 @@ build_bitmaps () {
continue; /* defunct */
if (node_record_table_ptr[i].node_state == STATE_IDLE)
bit_set (idle_node_bitmap, i);
- if (node_record_table_ptr[i].node_state > STATE_DOWN)
+ if ((node_record_table_ptr[i].node_state != STATE_DOWN) &&
+ ((node_record_table_ptr[i].node_state & STATE_NO_RESPOND) == 0))
bit_set (up_node_bitmap, i);
if (node_record_table_ptr[i].config_ptr)
bit_set (node_record_table_ptr[i].config_ptr->node_bitmap, i);