diff --git a/src/common/gres.c b/src/common/gres.c
index d5933fdff327f099872723d5d04eb55ea96e1980..6366159ffdda0aa9a7789aedcb91348c90af05ca 100644
--- a/src/common/gres.c
+++ b/src/common/gres.c
@@ -82,6 +82,9 @@ typedef struct slurm_gres_ops {
int (*load_node_config) ( void );
int (*pack_node_config) ( Buf buffer );
int (*unpack_node_config) ( Buf buffer );
+ int (*node_config_init) ( char *node_name,
+ char *orig_config,
+ void **gres_data );
int (*node_config_validate) ( char *node_name,
char *orig_config,
char **new_config,
@@ -173,6 +176,7 @@ static int _load_gres_plugin(char *plugin_name,
"load_node_config",
"pack_node_config",
"unpack_node_config",
+ "node_config_init",
"node_config_validate",
"node_reconfig",
"node_config_delete",
@@ -600,6 +604,50 @@ static void _gres_node_list_delete(void *list_element)
slurm_mutex_unlock(&gres_context_lock);
}
+/*
+ * Build a node's gres record based only upon the slurm.conf contents
+ * IN node_name - name of the node for which the gres information applies
+ * IN orig_config - Gres information supplied from slurm.conf
+ * IN/OUT gres_list - List of Gres records for this node to track usage
+ */
+extern int gres_plugin_init_node_config(char *node_name, char *orig_config,
+ List *gres_list)
+{
+ int i, rc;
+ ListIterator gres_iter;
+ gres_state_t *gres_ptr;
+
+ rc = gres_plugin_init();
+
+ slurm_mutex_lock(&gres_context_lock);
+ if ((gres_context_cnt > 0) && (*gres_list == NULL)) {
+ *gres_list = list_create(_gres_node_list_delete);
+ if (*gres_list == NULL)
+ fatal("list_create malloc failure");
+ }
+ for (i=0; ((i < gres_context_cnt) && (rc == SLURM_SUCCESS)); i++) {
+ /* Find or create gres_state entry on the list */
+ gres_iter = list_iterator_create(*gres_list);
+ while ((gres_ptr = (gres_state_t *) list_next(gres_iter))) {
+ if (gres_ptr->plugin_id ==
+ *(gres_context[i].ops.plugin_id))
+ break;
+ }
+ list_iterator_destroy(gres_iter);
+ if (gres_ptr == NULL) {
+ gres_ptr = xmalloc(sizeof(gres_state_t));
+ gres_ptr->plugin_id = *(gres_context[i].ops.plugin_id);
+ list_append(*gres_list, gres_ptr);
+ }
+
+ rc = (*(gres_context[i].ops.node_config_init))
+ (node_name, orig_config, &gres_ptr->gres_data);
+ }
+ slurm_mutex_unlock(&gres_context_lock);
+
+ return rc;
+}
+
/*
* Validate a node's configuration and put a gres record onto a list
* Called immediately after gres_plugin_unpack_node_config().
@@ -612,7 +660,7 @@ static void _gres_node_list_delete(void *list_element)
* 2: Don't validate hardware, use slurm.conf configuration
* OUT reason_down - set to an explanation of failure, if any, don't set if NULL
*/
-extern int gres_plugin_node_config_validate(char *node_name,
+extern int gres_plugin_node_config_validate(char *node_name,
char *orig_config,
char **new_config,
List *gres_list,
@@ -1320,7 +1368,7 @@ extern uint32_t gres_plugin_job_test(List job_gres_list, List node_gres_list,
if (job_gres_list == NULL)
return NO_VAL;
if (node_gres_list == NULL)
- return (uint32_t) 0;
+ return NO_VAL;
cpu_cnt = NO_VAL;
(void) gres_plugin_init();
diff --git a/src/common/gres.h b/src/common/gres.h
index 25128c6aa3b53c193c132d4f5ce0330f95ee8b12..acc85e8e0c181256b1be24dc59a87e8a826474fc 100644
--- a/src/common/gres.h
+++ b/src/common/gres.h
@@ -95,6 +95,15 @@ extern int gres_plugin_pack_node_config(Buf buffer);
* PLUGIN CALLS FOR SLURMCTLD DAEMON *
**************************************************************************
*/
+/*
+ * Build a node's gres record based only upon the slurm.conf contents
+ * IN node_name - name of the node for which the gres information applies
+ * IN orig_config - Gres information supplied from slurm.conf
+ * IN/OUT gres_list - List of Gres records for this node to track usage
+ */
+extern int gres_plugin_init_node_config(char *node_name, char *orig_config,
+ List *gres_list);
+
/*
* Unpack this node's configuration from a buffer
* IN buffer - message buffer to unpack
diff --git a/src/plugins/gres/gpu/gres_gpu.c b/src/plugins/gres/gpu/gres_gpu.c
index de61ea95208b1cf882e707979ed62bf5991f3931..35dd1f48efdcff309eb6895a918aa02072d0e253 100644
--- a/src/plugins/gres/gpu/gres_gpu.c
+++ b/src/plugins/gres/gpu/gres_gpu.c
@@ -248,6 +248,65 @@ extern void node_config_delete(void *gres_data)
xfree(gres_ptr);
}
+extern int node_config_init(char *node_name, char *orig_config,
+ void **gres_data)
+{
+ int rc = SLURM_SUCCESS;
+ gpu_node_state_t *gres_ptr;
+ char *node_gres_config, *tok, *last = NULL;
+ int32_t gres_config_cnt = 0;
+ bool updated_config = false;
+
+ xassert(gres_data);
+ gres_ptr = (gpu_node_state_t *) *gres_data;
+ if (gres_ptr == NULL) {
+ gres_ptr = xmalloc(sizeof(gpu_node_state_t));
+ *gres_data = gres_ptr;
+ gres_ptr->gpu_cnt_config = NO_VAL;
+ gres_ptr->gpu_cnt_found = NO_VAL;
+ updated_config = true;
+ }
+
+ /* If the resource isn't configured for use with this node,
+ * just return leaving gpu_cnt_config=0, gpu_cnt_avail=0, etc. */
+ if ((orig_config == NULL) || (orig_config[0] == '\0') ||
+ (updated_config == false))
+ return rc;
+
+ node_gres_config = xstrdup(orig_config);
+ tok = strtok_r(node_gres_config, ",", &last);
+ while (tok) {
+ if (!strcmp(tok, "gpu")) {
+ gres_config_cnt = 1;
+ break;
+ }
+ if (!strncmp(tok, "gpu:", 4)) {
+ gres_config_cnt = strtol(tok+4, &last, 10);
+ if (last[0] == '\0')
+ ;
+ else if ((last[0] == 'k') || (last[0] == 'K'))
+ gres_config_cnt *= 1024;
+ break;
+ }
+ tok = strtok_r(NULL, ",", &last);
+ }
+ xfree(node_gres_config);
+
+ gres_ptr->gpu_cnt_config = gres_config_cnt;
+ gres_ptr->gpu_cnt_avail = gres_config_cnt;
+ if (gres_ptr->gpu_bit_alloc == NULL) {
+ gres_ptr->gpu_bit_alloc = bit_alloc(gres_ptr->gpu_cnt_avail);
+ } else if (gres_ptr->gpu_cnt_avail >
+ bit_size(gres_ptr->gpu_bit_alloc)) {
+ gres_ptr->gpu_bit_alloc = bit_realloc(gres_ptr->gpu_bit_alloc,
+ gres_ptr->gpu_cnt_avail);
+ }
+ if (gres_ptr->gpu_bit_alloc == NULL)
+ fatal("bit_alloc: malloc failure");
+
+ return rc;
+}
+
/*
* Validate a node's configuration and put a gres record onto a list
* Called immediately after unpack_node_config().
@@ -268,7 +327,7 @@ extern int node_config_validate(char *node_name,
int rc = SLURM_SUCCESS;
gpu_node_state_t *gres_ptr;
char *node_gres_config, *tok, *last = NULL;
- int32_t gres_config_cnt = -1;
+ uint32_t gres_config_cnt = 0;
bool updated_config = false;
xassert(gres_data);
@@ -276,7 +335,8 @@ extern int node_config_validate(char *node_name,
if (gres_ptr == NULL) {
gres_ptr = xmalloc(sizeof(gpu_node_state_t));
*gres_data = gres_ptr;
- gres_ptr->gpu_cnt_found = gres_config.gpu_cnt;
+ gres_ptr->gpu_cnt_config = NO_VAL;
+ gres_ptr->gpu_cnt_found = gres_config.gpu_cnt;
updated_config = true;
} else if (gres_ptr->gpu_cnt_found != gres_config.gpu_cnt) {
if (gres_ptr->gpu_cnt_found != NO_VAL) {
@@ -287,39 +347,37 @@ extern int node_config_validate(char *node_name,
gres_ptr->gpu_cnt_found = gres_config.gpu_cnt;
updated_config = true;
}
-
- /* If the resource isn't configured for use with this node,
- * just return leaving gpu_cnt_config=0, gpu_cnt_avail=0, etc. */
- if ((orig_config == NULL) || (orig_config[0] == '\0') ||
- (updated_config == false))
+ if (updated_config == false)
return SLURM_SUCCESS;
- node_gres_config = xstrdup(orig_config);
- tok = strtok_r(node_gres_config, ",", &last);
- while (tok) {
- if (!strcmp(tok, "gpu")) {
- gres_config_cnt = 1;
- break;
- }
- if (!strncmp(tok, "gpu:", 4)) {
- gres_config_cnt = strtol(tok+4, &last, 10);
- if (last[0] == '\0')
- ;
- else if ((last[0] == 'k') || (last[0] == 'K'))
- gres_config_cnt *= 1024;
- break;
+ if ((orig_config == NULL) || (orig_config[0] == '\0'))
+ gres_ptr->gpu_cnt_config = 0;
+ else if (gres_ptr->gpu_cnt_config == NO_VAL) {
+ node_gres_config = xstrdup(orig_config);
+ tok = strtok_r(node_gres_config, ",", &last);
+ while (tok) {
+ if (!strcmp(tok, "gpu")) {
+ gres_config_cnt = 1;
+ break;
+ }
+ if (!strncmp(tok, "gpu:", 4)) {
+ gres_config_cnt = strtol(tok+4, &last, 10);
+ if (last[0] == '\0')
+ ;
+ else if ((last[0] == 'k') || (last[0] == 'K'))
+ gres_config_cnt *= 1024;
+ break;
+ }
+ tok = strtok_r(NULL, ",", &last);
}
- tok = strtok_r(NULL, ",", &last);
+ xfree(node_gres_config);
+ gres_ptr->gpu_cnt_config = gres_config_cnt;
}
- gres_ptr->gpu_cnt_config = gres_config_cnt;
- xfree(node_gres_config);
- if (gres_config_cnt < 0)
- ; /* not configured for use */
- else if (fast_schedule == 0)
- gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_found;
- else
+ if ((gres_ptr->gpu_cnt_config == 0) || (fast_schedule > 0))
gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_config;
+ else
+ gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_found;
if (gres_ptr->gpu_bit_alloc == NULL) {
gres_ptr->gpu_bit_alloc = bit_alloc(gres_ptr->gpu_cnt_avail);
@@ -327,9 +385,9 @@ extern int node_config_validate(char *node_name,
bit_size(gres_ptr->gpu_bit_alloc)) {
gres_ptr->gpu_bit_alloc = bit_realloc(gres_ptr->gpu_bit_alloc,
gres_ptr->gpu_cnt_avail);
+ if (gres_ptr->gpu_bit_alloc == NULL)
+ fatal("bit_alloc: malloc failure");
}
- if (gres_ptr->gpu_bit_alloc == NULL)
- fatal("bit_alloc: malloc failure");
if ((fast_schedule < 2) &&
(gres_ptr->gpu_cnt_found < gres_ptr->gpu_cnt_config)) {
@@ -385,9 +443,10 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
xassert(gres_data);
gres_ptr = (gpu_node_state_t *) *gres_data;
if (gres_ptr == NULL) {
- /* Assume that node has not yet registerd */
- info("%s record is NULL for node %s", plugin_name, node_name);
- return rc;
+ gres_ptr = xmalloc(sizeof(gpu_node_state_t));
+ *gres_data = gres_ptr;
+ gres_ptr->gpu_cnt_config = NO_VAL;
+ gres_ptr->gpu_cnt_found = NO_VAL;
}
node_gres_config = xstrdup(orig_config);
@@ -410,10 +469,11 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
gres_ptr->gpu_cnt_config = gres_config_cnt;
xfree(node_gres_config);
- if (fast_schedule == 0)
- gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_found;
- else if (gres_ptr->gpu_cnt_config != NO_VAL)
+ if ((gres_ptr->gpu_cnt_config == 0) || (fast_schedule > 0) ||
+ (gres_ptr->gpu_cnt_found == NO_VAL))
gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_config;
+ else
+ gres_ptr->gpu_cnt_avail = gres_ptr->gpu_cnt_found;
if (gres_ptr->gpu_bit_alloc == NULL) {
gres_ptr->gpu_bit_alloc = bit_alloc(gres_ptr->gpu_cnt_avail);
@@ -425,13 +485,15 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
if (gres_ptr->gpu_bit_alloc == NULL)
fatal("bit_alloc: malloc failure");
- if ((fast_schedule < 2) &&
- (gres_ptr->gpu_cnt_found < gres_ptr->gpu_cnt_config)) {
+ if ((fast_schedule < 2) &&
+ (gres_ptr->gpu_cnt_found != NO_VAL) &&
+ (gres_ptr->gpu_cnt_found < gres_ptr->gpu_cnt_config)) {
/* Do not set node DOWN, but give the node
* a chance to register with more resources */
gres_ptr->gpu_cnt_found = NO_VAL;
- } else if ((fast_schedule == 0) &&
- (gres_ptr->gpu_cnt_found > gres_ptr->gpu_cnt_config)) {
+ } else if ((fast_schedule == 0) &&
+ (gres_ptr->gpu_cnt_found != NO_VAL) &&
+ (gres_ptr->gpu_cnt_found > gres_ptr->gpu_cnt_config)) {
/* need to rebuild new_config */
char *new_configured_res = NULL;
if (*new_config)
@@ -567,7 +629,19 @@ extern int node_state_realloc(void *job_gres_data, int node_offset,
job_bit_size = bit_size(job_gres_ptr->gpu_bit_alloc[node_offset]);
node_bit_size = bit_size(node_gres_ptr->gpu_bit_alloc);
- if (job_bit_size != node_bit_size) {
+ if (job_bit_size > node_bit_size) {
+ error("%s job/node bit size mismatch (%d != %d)",
+ plugin_name, job_bit_size, node_bit_size);
+ /* Node needs to register with more resources, expand
+ * node's bitmap now so we can merge the data */
+ node_gres_ptr->gpu_bit_alloc =
+ bit_realloc(node_gres_ptr->gpu_bit_alloc,
+ job_bit_size);
+ if (node_gres_ptr->gpu_bit_alloc == NULL)
+ fatal("bit_realloc: malloc failure");
+ node_bit_size = job_bit_size;
+ }
+ if (job_bit_size < node_bit_size) {
error("%s job/node bit size mismatch (%d != %d)",
plugin_name, job_bit_size, node_bit_size);
/* Update what we can */
diff --git a/src/plugins/gres/nic/gres_nic.c b/src/plugins/gres/nic/gres_nic.c
index e15fc7c47f849f2a7a1829f23531a620977bac39..2daf2b1ae55cc7cad7668b4d9e24fb240ed93c31 100644
--- a/src/plugins/gres/nic/gres_nic.c
+++ b/src/plugins/gres/nic/gres_nic.c
@@ -248,6 +248,65 @@ extern void node_config_delete(void *gres_data)
xfree(gres_ptr);
}
+extern int node_config_init(char *node_name, char *orig_config,
+ void **gres_data)
+{
+ int rc = SLURM_SUCCESS;
+ nic_node_state_t *gres_ptr;
+ char *node_gres_config, *tok, *last = NULL;
+ int32_t gres_config_cnt = 0;
+ bool updated_config = false;
+
+ xassert(gres_data);
+ gres_ptr = (nic_node_state_t *) *gres_data;
+ if (gres_ptr == NULL) {
+ gres_ptr = xmalloc(sizeof(nic_node_state_t));
+ *gres_data = gres_ptr;
+ gres_ptr->nic_cnt_config = NO_VAL;
+ gres_ptr->nic_cnt_found = NO_VAL;
+ updated_config = true;
+ }
+
+ /* If the resource isn't configured for use with this node,
+ * just return leaving nic_cnt_config=0, nic_cnt_avail=0, etc. */
+ if ((orig_config == NULL) || (orig_config[0] == '\0') ||
+ (updated_config == false))
+ return rc;
+
+ node_gres_config = xstrdup(orig_config);
+ tok = strtok_r(node_gres_config, ",", &last);
+ while (tok) {
+ if (!strcmp(tok, "nic")) {
+ gres_config_cnt = 1;
+ break;
+ }
+ if (!strncmp(tok, "nic:", 4)) {
+ gres_config_cnt = strtol(tok+4, &last, 10);
+ if (last[0] == '\0')
+ ;
+ else if ((last[0] == 'k') || (last[0] == 'K'))
+ gres_config_cnt *= 1024;
+ break;
+ }
+ tok = strtok_r(NULL, ",", &last);
+ }
+ xfree(node_gres_config);
+
+ gres_ptr->nic_cnt_config = gres_config_cnt;
+ gres_ptr->nic_cnt_avail = gres_config_cnt;
+ if (gres_ptr->nic_bit_alloc == NULL) {
+ gres_ptr->nic_bit_alloc = bit_alloc(gres_ptr->nic_cnt_avail);
+ } else if (gres_ptr->nic_cnt_avail >
+ bit_size(gres_ptr->nic_bit_alloc)) {
+ gres_ptr->nic_bit_alloc = bit_realloc(gres_ptr->nic_bit_alloc,
+ gres_ptr->nic_cnt_avail);
+ }
+ if (gres_ptr->nic_bit_alloc == NULL)
+ fatal("bit_alloc: malloc failure");
+
+ return rc;
+}
+
/*
* Validate a node's configuration and put a gres record onto a list
* Called immediately after unpack_node_config().
@@ -268,7 +327,7 @@ extern int node_config_validate(char *node_name,
int rc = SLURM_SUCCESS;
nic_node_state_t *gres_ptr;
char *node_gres_config, *tok, *last = NULL;
- int32_t gres_config_cnt = -1;
+ uint32_t gres_config_cnt = 0;
bool updated_config = false;
xassert(gres_data);
@@ -276,7 +335,8 @@ extern int node_config_validate(char *node_name,
if (gres_ptr == NULL) {
gres_ptr = xmalloc(sizeof(nic_node_state_t));
*gres_data = gres_ptr;
- gres_ptr->nic_cnt_found = gres_config.nic_cnt;
+ gres_ptr->nic_cnt_config = NO_VAL;
+ gres_ptr->nic_cnt_found = gres_config.nic_cnt;
updated_config = true;
} else if (gres_ptr->nic_cnt_found != gres_config.nic_cnt) {
if (gres_ptr->nic_cnt_found != NO_VAL) {
@@ -287,39 +347,37 @@ extern int node_config_validate(char *node_name,
gres_ptr->nic_cnt_found = gres_config.nic_cnt;
updated_config = true;
}
-
- /* If the resource isn't configured for use with this node,
- * just return leaving nic_cnt_config=0, nic_cnt_avail=0, etc. */
- if ((orig_config == NULL) || (orig_config[0] == '\0') ||
- (updated_config == false))
+ if (updated_config == false)
return SLURM_SUCCESS;
- node_gres_config = xstrdup(orig_config);
- tok = strtok_r(node_gres_config, ",", &last);
- while (tok) {
- if (!strcmp(tok, "nic")) {
- gres_config_cnt = 1;
- break;
- }
- if (!strncmp(tok, "nic:", 4)) {
- gres_config_cnt = strtol(tok+4, &last, 10);
- if (last[0] == '\0')
- ;
- else if ((last[0] == 'k') || (last[0] == 'K'))
- gres_config_cnt *= 1024;
- break;
+ if ((orig_config == NULL) || (orig_config[0] == '\0'))
+ gres_ptr->nic_cnt_config = 0;
+ else if (gres_ptr->nic_cnt_config == NO_VAL) {
+ node_gres_config = xstrdup(orig_config);
+ tok = strtok_r(node_gres_config, ",", &last);
+ while (tok) {
+ if (!strcmp(tok, "nic")) {
+ gres_config_cnt = 1;
+ break;
+ }
+ if (!strncmp(tok, "nic:", 4)) {
+ gres_config_cnt = strtol(tok+4, &last, 10);
+ if (last[0] == '\0')
+ ;
+ else if ((last[0] == 'k') || (last[0] == 'K'))
+ gres_config_cnt *= 1024;
+ break;
+ }
+ tok = strtok_r(NULL, ",", &last);
}
- tok = strtok_r(NULL, ",", &last);
+ xfree(node_gres_config);
+ gres_ptr->nic_cnt_config = gres_config_cnt;
}
- gres_ptr->nic_cnt_config = gres_config_cnt;
- xfree(node_gres_config);
- if (gres_config_cnt < 0)
- ; /* not configured for use */
- else if (fast_schedule == 0)
- gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_found;
- else
+ if ((gres_ptr->nic_cnt_config == 0) || (fast_schedule > 0))
gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_config;
+ else
+ gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_found;
if (gres_ptr->nic_bit_alloc == NULL) {
gres_ptr->nic_bit_alloc = bit_alloc(gres_ptr->nic_cnt_avail);
@@ -327,9 +385,9 @@ extern int node_config_validate(char *node_name,
bit_size(gres_ptr->nic_bit_alloc)) {
gres_ptr->nic_bit_alloc = bit_realloc(gres_ptr->nic_bit_alloc,
gres_ptr->nic_cnt_avail);
+ if (gres_ptr->nic_bit_alloc == NULL)
+ fatal("bit_alloc: malloc failure");
}
- if (gres_ptr->nic_bit_alloc == NULL)
- fatal("bit_alloc: malloc failure");
if ((fast_schedule < 2) &&
(gres_ptr->nic_cnt_found < gres_ptr->nic_cnt_config)) {
@@ -385,9 +443,10 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
xassert(gres_data);
gres_ptr = (nic_node_state_t *) *gres_data;
if (gres_ptr == NULL) {
- /* Assume that node has not yet registerd */
- info("%s record is NULL for node %s", plugin_name, node_name);
- return rc;
+ gres_ptr = xmalloc(sizeof(nic_node_state_t));
+ *gres_data = gres_ptr;
+ gres_ptr->nic_cnt_config = NO_VAL;
+ gres_ptr->nic_cnt_found = NO_VAL;
}
node_gres_config = xstrdup(orig_config);
@@ -410,10 +469,11 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
gres_ptr->nic_cnt_config = gres_config_cnt;
xfree(node_gres_config);
- if (fast_schedule == 0)
- gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_found;
- else if (gres_ptr->nic_cnt_config != NO_VAL)
+ if ((gres_ptr->nic_cnt_config == 0) || (fast_schedule > 0) ||
+ (gres_ptr->nic_cnt_found == NO_VAL))
gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_config;
+ else
+ gres_ptr->nic_cnt_avail = gres_ptr->nic_cnt_found;
if (gres_ptr->nic_bit_alloc == NULL) {
gres_ptr->nic_bit_alloc = bit_alloc(gres_ptr->nic_cnt_avail);
@@ -425,13 +485,15 @@ extern int node_reconfig(char *node_name, char *orig_config, char **new_config,
if (gres_ptr->nic_bit_alloc == NULL)
fatal("bit_alloc: malloc failure");
- if ((fast_schedule < 2) &&
- (gres_ptr->nic_cnt_found < gres_ptr->nic_cnt_config)) {
+ if ((fast_schedule < 2) &&
+ (gres_ptr->nic_cnt_found != NO_VAL) &&
+ (gres_ptr->nic_cnt_found < gres_ptr->nic_cnt_config)) {
/* Do not set node DOWN, but give the node
* a chance to register with more resources */
gres_ptr->nic_cnt_found = NO_VAL;
- } else if ((fast_schedule == 0) &&
- (gres_ptr->nic_cnt_found > gres_ptr->nic_cnt_config)) {
+ } else if ((fast_schedule == 0) &&
+ (gres_ptr->nic_cnt_found != NO_VAL) &&
+ (gres_ptr->nic_cnt_found > gres_ptr->nic_cnt_config)) {
/* need to rebuild new_config */
char *new_configured_res = NULL;
if (*new_config)
@@ -567,7 +629,19 @@ extern int node_state_realloc(void *job_gres_data, int node_offset,
job_bit_size = bit_size(job_gres_ptr->nic_bit_alloc[node_offset]);
node_bit_size = bit_size(node_gres_ptr->nic_bit_alloc);
- if (job_bit_size != node_bit_size) {
+ if (job_bit_size > node_bit_size) {
+ error("%s job/node bit size mismatch (%d != %d)",
+ plugin_name, job_bit_size, node_bit_size);
+ /* Node needs to register with more resources, expand
+ * node's bitmap now so we can merge the data */
+ node_gres_ptr->nic_bit_alloc =
+ bit_realloc(node_gres_ptr->nic_bit_alloc,
+ job_bit_size);
+ if (node_gres_ptr->nic_bit_alloc == NULL)
+ fatal("bit_realloc: malloc failure");
+ node_bit_size = job_bit_size;
+ }
+ if (job_bit_size < node_bit_size) {
error("%s job/node bit size mismatch (%d != %d)",
plugin_name, job_bit_size, node_bit_size);
/* Update what we can */
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 16a01dbc88fdefd0ff44fff93c6cfce25e213949..5d9a333a8cc9a1b6c99be4bbbf8cfb668eb5eeb5 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -824,10 +824,21 @@ static void _gres_reconig(bool reconfig)
ListIterator job_iterator;
int i, i_first, i_last, node_offset;
bool gres_active, gres_changed = false;
- char *plugin_names;
+ char *gres_name, *plugin_names;
- if (reconfig)
+ if (reconfig) {
gres_plugin_reconfig(&gres_changed);
+ } else {
+ for (i = 0, node_ptr = node_record_table_ptr;
+ i < node_record_count; i++, node_ptr++) {
+ if (node_ptr->gres)
+ gres_name = node_ptr->gres;
+ else
+ gres_name = node_ptr->config_ptr->gres;
+ gres_plugin_init_node_config(node_ptr->name, gres_name,
+ &node_ptr->gres_list);
+ }
+ }
plugin_names = slurm_get_gres_plugins();
if (plugin_names && plugin_names[0])