diff --git a/src/common/slurm_errno.c b/src/common/slurm_errno.c
index 56533f1f4179d7aee6bc939ceaffee64a56ce7d1..4720b21895c488ad3cc9fd90f1d5a96065a08eee 100644
--- a/src/common/slurm_errno.c
+++ b/src/common/slurm_errno.c
@@ -122,7 +122,7 @@ static slurm_errtab_t slurm_errtab[] = {
{ ESLURM_TOO_MANY_REQUESTED_CPUS,
"More processors requested than permitted" },
{ ESLURM_TOO_MANY_REQUESTED_NODES,
- "More nodes requested than permitted" },
+ "Node count specification invalid" },
{ ESLURM_ERROR_ON_DESC_TO_RECORD_COPY,
"Unable to create job record, try again" },
{ ESLURM_JOB_MISSING_SIZE_SPECIFICATION,
diff --git a/src/plugins/select/bluegene/plugin/select_bluegene.c b/src/plugins/select/bluegene/plugin/select_bluegene.c
index 51d17dd599d0e31341170c07db4c855c6290d109..9b6cc45508d9f4c7f6c87f1882de7b8cf388296d 100644
--- a/src/plugins/select/bluegene/plugin/select_bluegene.c
+++ b/src/plugins/select/bluegene/plugin/select_bluegene.c
@@ -224,6 +224,8 @@ extern int fini ( void )
struct part_record *part_ptr = NULL;
ListIterator itr = list_iterator_create(part_list);
while((part_ptr = list_next(itr))) {
+ part_ptr->max_nodes = part_ptr->max_nodes_orig;
+ part_ptr->min_nodes = part_ptr->min_nodes_orig;
select_p_alter_node_cnt(SELECT_SET_BP_CNT,
&part_ptr->max_nodes);
select_p_alter_node_cnt(SELECT_SET_BP_CNT,
@@ -1164,8 +1166,8 @@ extern int select_p_update_node_state (int index, uint16_t state)
extern int select_p_alter_node_cnt(enum select_node_cnt type, void *data)
{
job_desc_msg_t *job_desc = (job_desc_msg_t *)data;
- uint32_t *nodes = (uint32_t *)data;
- int tmp, i;
+ uint32_t *nodes = (uint32_t *)data, tmp;
+ int i;
uint16_t req_geometry[BA_SYSTEM_DIMENSIONS];
if(!bluegene_bp_node_cnt) {
@@ -1179,7 +1181,9 @@ extern int select_p_alter_node_cnt(enum select_node_cnt type, void *data)
(*nodes) = bluegene_bp_node_cnt;
break;
case SELECT_SET_BP_CNT:
- if((*nodes) > bluegene_bp_node_cnt) {
+ if(((*nodes) == INFINITE) || ((*nodes) == NO_VAL))
+ tmp = (*nodes);
+ else if((*nodes) > bluegene_bp_node_cnt) {
tmp = (*nodes);
tmp /= bluegene_bp_node_cnt;
if(tmp < 1)
diff --git a/src/sbatch/opt.c b/src/sbatch/opt.c
index b5b8d933c8ea666e29b1e95465cf98b168151d37..8429c056c62b4a210575b407ee28ea3a907a64c0 100644
--- a/src/sbatch/opt.c
+++ b/src/sbatch/opt.c
@@ -226,7 +226,7 @@ static void _opt_default()
opt.nprocs_set = false;
opt.cpus_per_task = 1;
opt.cpus_set = false;
- opt.min_nodes = 1;
+ opt.min_nodes = 0;
opt.max_nodes = 0;
opt.nodes_set = false;
opt.min_sockets_per_node = NO_VAL; /* requested min/maxsockets */
@@ -1742,7 +1742,7 @@ static bool _opt_verify(void)
verified = false;
}
- if ((opt.min_nodes <= 0) || (opt.max_nodes < 0) ||
+ if ((opt.min_nodes < 0) || (opt.max_nodes < 0) ||
(opt.max_nodes && (opt.min_nodes > opt.max_nodes))) {
error("%s: invalid number of nodes (-N %d-%d)\n",
opt.progname, opt.min_nodes, opt.max_nodes);
@@ -1771,7 +1771,8 @@ static bool _opt_verify(void)
* environment are more extensive and are documented in the
* SLURM reference guide. */
if (opt.distribution == SLURM_DIST_PLANE && opt.plane_size) {
- if ((opt.nprocs/opt.plane_size) < opt.min_nodes) {
+ if ((opt.min_nodes <= 0) ||
+ ((opt.nprocs/opt.plane_size) < opt.min_nodes)) {
if (((opt.min_nodes-1)*opt.plane_size) >= opt.nprocs) {
#if(0)
info("Too few processes ((n/plane_size) %d < N %d) "
@@ -1815,7 +1816,7 @@ static bool _opt_verify(void)
/* massage the numbers */
if ((opt.nodes_set || opt.extra_set) && !opt.nprocs_set) {
/* 1 proc / node default */
- opt.nprocs = opt.min_nodes;
+ opt.nprocs = MAX(opt.min_nodes, 1);
/* 1 proc / min_[socket * core * thread] default */
if (opt.min_sockets_per_node > 0) {
diff --git a/src/sbatch/sbatch.c b/src/sbatch/sbatch.c
index 6c8d7141e37ffb0390cfa6290928361c3ce49782..8f1556abdf74d3684587384842a0dc22558c7330 100644
--- a/src/sbatch/sbatch.c
+++ b/src/sbatch/sbatch.c
@@ -163,7 +163,8 @@ static int fill_job_desc_from_opts(job_desc_msg_t *desc)
desc->req_nodes = opt.nodelist;
desc->exc_nodes = opt.exc_nodes;
desc->partition = opt.partition;
- desc->min_nodes = opt.min_nodes;
+ if (opt.min_nodes)
+ desc->min_nodes = opt.min_nodes;
if (opt.licenses)
desc->licenses = xstrdup(opt.licenses);
if (opt.max_nodes)
@@ -227,7 +228,7 @@ static int fill_job_desc_from_opts(job_desc_msg_t *desc)
if (opt.tmpdisk > -1)
desc->job_min_tmp_disk = opt.tmpdisk;
if (opt.overcommit) {
- desc->num_procs = opt.min_nodes;
+ desc->num_procs = MAX(opt.min_nodes, 1);
desc->overcommit = opt.overcommit;
} else
desc->num_procs = opt.nprocs * opt.cpus_per_task;
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index d828566a6efebf4f28678ca716bd5b5f6a366f53..bff6565b667798fe1d670dbb7d5e962432643614 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1808,7 +1808,7 @@ static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run,
bitstr_t *req_bitmap = NULL, *exc_bitmap = NULL;
bool super_user = false;
struct job_record *job_ptr;
- uint32_t total_nodes, max_procs;
+ uint32_t total_nodes, max_procs, max_nodes_orig;
acct_association_rec_t assoc_rec, *assoc_ptr;
List license_list = NULL;
bool valid;
@@ -1820,14 +1820,15 @@ static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run,
uint16_t conn_type;
#endif
- debug2("before alteration asking for nodes %u-%u procs %u",
+ max_nodes_orig = job_desc->max_nodes;
+ info("before alteration asking for nodes %u-%u procs %u",
job_desc->min_nodes, job_desc->max_nodes,
job_desc->num_procs);
select_g_alter_node_cnt(SELECT_SET_NODE_CNT, job_desc);
select_g_get_jobinfo(job_desc->select_jobinfo,
SELECT_DATA_MAX_PROCS, &max_procs);
- debug2("after alteration asking for nodes %u-%u procs %u-%u",
+ info("after alteration asking for nodes %u-%u procs %u-%u",
job_desc->min_nodes, job_desc->max_nodes,
job_desc->num_procs, max_procs);
@@ -1853,6 +1854,17 @@ static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run,
}
part_ptr = default_part_loc;
}
+ if (job_desc->min_nodes == NO_VAL)
+ job_desc->min_nodes = part_ptr->min_nodes_orig;
+ if (job_desc->max_nodes == NO_VAL)
+ job_desc->max_nodes = part_ptr->max_nodes_orig;
+ else if (max_nodes_orig < part_ptr->min_nodes_orig) {
+ info("_job_create: job's max nodes less than partition's "
+ "min nodes (%u < %u)",
+ max_nodes_orig, part_ptr->min_nodes_orig);
+ error_code = ESLURM_TOO_MANY_REQUESTED_NODES;
+ return error_code;
+ }
if ((job_desc->user_id == 0) && part_ptr->disable_root_jobs) {
error("Security violation, SUBMIT_JOB for user root disabled");
diff --git a/src/slurmctld/partition_mgr.c b/src/slurmctld/partition_mgr.c
index 958e01338b9a9b9ea4bf2340a783a4ec0468de11..b7d38e3f0d7b15f811dfc275f81d8e22ef5b48c7 100644
--- a/src/slurmctld/partition_mgr.c
+++ b/src/slurmctld/partition_mgr.c
@@ -217,17 +217,19 @@ struct part_record *create_part_record(void)
(struct part_record *) xmalloc(sizeof(struct part_record));
xassert (part_ptr->magic = PART_MAGIC); /* set value */
- part_ptr->name = xstrdup("DEFAULT");
+ part_ptr->name = xstrdup("DEFAULT");
part_ptr->disable_root_jobs = default_part.disable_root_jobs;
- part_ptr->hidden = default_part.hidden;
- part_ptr->max_time = default_part.max_time;
- part_ptr->max_nodes = default_part.max_nodes;
- part_ptr->min_nodes = default_part.min_nodes;
- part_ptr->root_only = default_part.root_only;
- part_ptr->state_up = default_part.state_up;
- part_ptr->max_share = default_part.max_share;
- part_ptr->priority = default_part.priority;
- part_ptr->node_bitmap = NULL;
+ part_ptr->hidden = default_part.hidden;
+ part_ptr->max_time = default_part.max_time;
+ part_ptr->max_nodes = default_part.max_nodes;
+ part_ptr->max_nodes_orig = default_part.max_nodes;
+ part_ptr->min_nodes = default_part.min_nodes;
+ part_ptr->min_nodes_orig = default_part.min_nodes;
+ part_ptr->root_only = default_part.root_only;
+ part_ptr->state_up = default_part.state_up;
+ part_ptr->max_share = default_part.max_share;
+ part_ptr->priority = default_part.priority;
+ part_ptr->node_bitmap = NULL;
if (default_part.allow_groups)
part_ptr->allow_groups = xstrdup(default_part.allow_groups);
@@ -367,20 +369,20 @@ static void _dump_part_state(struct part_record *part_ptr, Buf buffer)
else
default_part_flag = 0;
- packstr(part_ptr->name, buffer);
- pack32(part_ptr->max_time, buffer);
- pack32(part_ptr->max_nodes, buffer);
- pack32(part_ptr->min_nodes, buffer);
+ packstr(part_ptr->name, buffer);
+ pack32(part_ptr->max_time, buffer);
+ pack32(part_ptr->max_nodes_orig, buffer);
+ pack32(part_ptr->min_nodes_orig, buffer);
- pack16(default_part_flag, buffer);
- pack16(part_ptr->hidden, buffer);
- pack16(part_ptr->root_only, buffer);
- pack16(part_ptr->max_share, buffer);
- pack16(part_ptr->priority, buffer);
+ pack16(default_part_flag, buffer);
+ pack16(part_ptr->hidden, buffer);
+ pack16(part_ptr->root_only, buffer);
+ pack16(part_ptr->max_share, buffer);
+ pack16(part_ptr->priority, buffer);
- pack16(part_ptr->state_up, buffer);
- packstr(part_ptr->allow_groups, buffer);
- packstr(part_ptr->nodes, buffer);
+ pack16(part_ptr->state_up, buffer);
+ packstr(part_ptr->allow_groups, buffer);
+ packstr(part_ptr->nodes, buffer);
}
/*
@@ -489,21 +491,23 @@ int load_all_part_state(void)
if (part_ptr) {
part_cnt++;
- part_ptr->hidden = hidden;
- part_ptr->max_time = max_time;
- part_ptr->max_nodes = max_nodes;
- part_ptr->min_nodes = min_nodes;
+ part_ptr->hidden = hidden;
+ part_ptr->max_time = max_time;
+ part_ptr->max_nodes = max_nodes;
+ part_ptr->max_nodes_orig = max_nodes;
+ part_ptr->min_nodes = min_nodes;
+ part_ptr->min_nodes_orig = min_nodes;
if (def_part_flag) {
xfree(default_part_name);
default_part_name = xstrdup(part_name);
default_part_loc = part_ptr;
}
- part_ptr->root_only = root_only;
- part_ptr->max_share = max_share;
- part_ptr->priority = priority;
- part_ptr->state_up = state_up;
+ part_ptr->root_only = root_only;
+ part_ptr->max_share = max_share;
+ part_ptr->priority = priority;
+ part_ptr->state_up = state_up;
xfree(part_ptr->allow_groups);
- part_ptr->allow_groups = allow_groups;
+ part_ptr->allow_groups = allow_groups;
xfree(part_ptr->nodes);
part_ptr->nodes = nodes;
} else {
@@ -550,18 +554,20 @@ int init_part_conf(void)
last_part_update = time(NULL);
xfree(default_part.name); /* needed for reconfig */
- default_part.name = xstrdup("DEFAULT");
+ default_part.name = xstrdup("DEFAULT");
default_part.disable_root_jobs = slurmctld_conf.disable_root_jobs;
- default_part.hidden = 0;
- default_part.max_time = INFINITE;
- default_part.max_nodes = INFINITE;
- default_part.min_nodes = 1;
- default_part.root_only = 0;
- default_part.state_up = 1;
- default_part.max_share = 1;
- default_part.priority = 1;
- default_part.total_nodes = 0;
- default_part.total_cpus = 0;
+ default_part.hidden = 0;
+ default_part.max_time = INFINITE;
+ default_part.max_nodes = INFINITE;
+ default_part.max_nodes_orig = INFINITE;
+ default_part.min_nodes = 1;
+ default_part.min_nodes_orig = 1;
+ default_part.root_only = 0;
+ default_part.state_up = 1;
+ default_part.max_share = 1;
+ default_part.priority = 1;
+ default_part.total_nodes = 0;
+ default_part.total_cpus = 0;
xfree(default_part.nodes);
xfree(default_part.allow_groups);
xfree(default_part.allow_uids);
@@ -745,14 +751,8 @@ void pack_part(struct part_record *part_ptr, Buf buffer)
packstr(part_ptr->name, buffer);
pack32(part_ptr->max_time, buffer);
- altered = part_ptr->max_nodes;
- select_g_alter_node_cnt(SELECT_APPLY_NODE_MAX_OFFSET,
- &altered);
- pack32(altered, buffer);
- altered = part_ptr->min_nodes;
- select_g_alter_node_cnt(SELECT_APPLY_NODE_MIN_OFFSET,
- &altered);
- pack32(altered, buffer);
+ pack32(part_ptr->max_nodes_orig, buffer);
+ pack32(part_ptr->min_nodes_orig, buffer);
altered = part_ptr->total_nodes;
select_g_alter_node_cnt(SELECT_APPLY_NODE_MAX_OFFSET,
&altered);
@@ -826,7 +826,8 @@ int update_part(update_part_msg_t * part_desc)
if (part_desc->max_nodes != NO_VAL) {
info("update_part: setting max_nodes to %u for partition %s",
part_desc->max_nodes, part_desc->name);
- part_ptr->max_nodes = part_desc->max_nodes;
+ part_ptr->max_nodes = part_desc->max_nodes;
+ part_ptr->max_nodes_orig = part_desc->max_nodes;
select_g_alter_node_cnt(SELECT_SET_BP_CNT,
&part_ptr->max_nodes);
}
@@ -834,7 +835,8 @@ int update_part(update_part_msg_t * part_desc)
if (part_desc->min_nodes != NO_VAL) {
info("update_part: setting min_nodes to %u for partition %s",
part_desc->min_nodes, part_desc->name);
- part_ptr->min_nodes = part_desc->min_nodes;
+ part_ptr->min_nodes = part_desc->min_nodes;
+ part_ptr->min_nodes_orig = part_desc->min_nodes;
select_g_alter_node_cnt(SELECT_SET_BP_CNT,
&part_ptr->min_nodes);
}
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index d35b144c1a8d41fb0afb1493ddc618d15a2908ab..c1bd2188f4da39cc3b429f875f94e624ed8ab936 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -628,14 +628,16 @@ static int _build_single_partitionline_info(slurm_conf_partition_t *part)
if(part_ptr->disable_root_jobs)
debug2("partition %s does not allow root jobs", part_ptr->name);
- part_ptr->hidden = part->hidden_flag ? 1 : 0;
- part_ptr->max_time = part->max_time;
- part_ptr->max_share = part->max_share;
- part_ptr->max_nodes = part->max_nodes;
- part_ptr->min_nodes = part->min_nodes;
- part_ptr->priority = part->priority;
- part_ptr->root_only = part->root_only_flag ? 1 : 0;
- part_ptr->state_up = part->state_up_flag ? 1 : 0;
+ part_ptr->hidden = part->hidden_flag ? 1 : 0;
+ part_ptr->max_time = part->max_time;
+ part_ptr->max_share = part->max_share;
+ part_ptr->max_nodes = part->max_nodes;
+ part_ptr->max_nodes_orig = part->max_nodes;
+ part_ptr->min_nodes = part->min_nodes;
+ part_ptr->min_nodes_orig = part->min_nodes;
+ part_ptr->priority = part->priority;
+ part_ptr->root_only = part->root_only_flag ? 1 : 0;
+ part_ptr->state_up = part->state_up_flag ? 1 : 0;
if (part->allow_groups) {
xfree(part_ptr->allow_groups);
part_ptr->allow_groups = xstrdup(part->allow_groups);
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 6d0d8d9f85e62b732672561db48053357348645d..6ff3e1ee44dbebf9d50fec6e9a4a9076242cfa06 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -247,7 +247,9 @@ struct part_record {
uint16_t hidden; /* 1 if hidden by default */
uint32_t max_time; /* minutes or INFINITE */
uint32_t max_nodes; /* per job or INFINITE */
+ uint32_t max_nodes_orig;/* unscaled value (c-nodes on BlueGene) */
uint32_t min_nodes; /* per job */
+ uint32_t min_nodes_orig;/* unscaled value (c-nodes on BlueGene) */
uint32_t total_nodes; /* total number of nodes in the partition */
uint32_t total_cpus; /* total number of cpus in the partition */
uint32_t min_offset; /* select plugin min offset */