diff --git a/src/common/node_conf.c b/src/common/node_conf.c
index 4aff65daad0bd03d632b0eec5f5f629de6779060..e4c1c66947d577b1f22c604b3df6af8e7d132d8e 100644
--- a/src/common/node_conf.c
+++ b/src/common/node_conf.c
@@ -78,8 +78,9 @@
#define _DEBUG 0
/* Global variables */
-List config_list = NULL; /* list of config_record entries */
-List feature_list = NULL; /* list of features_record entries */
+List config_list = NULL; /* list of config_record entries */
+List feature_list = NULL; /* list of features_record entries */
+List front_end_list = NULL; /* list of slurm_conf_frontend_t entries */
time_t last_node_update = (time_t) 0; /* time of last update */
struct node_record *node_record_table_ptr = NULL; /* node records */
struct node_record **node_hash_table = NULL; /* node_record hash table */
@@ -100,7 +101,6 @@ static int _list_find_config (void *config_entry, void *key);
static int _list_find_feature (void *feature_entry, void *key);
-
static void _add_config_feature(char *feature, bitstr_t *node_bitmap)
{
struct features_record *feature_ptr;
@@ -152,7 +152,7 @@ static int _build_single_nodeline_info(slurm_conf_node_t *node_ptr,
int state_val = NODE_STATE_UNKNOWN;
if (node_ptr->state != NULL) {
- state_val = state_str2int(node_ptr->state);
+ state_val = state_str2int(node_ptr->state, node_ptr->nodenames);
if (state_val == NO_VAL)
goto cleanup;
}
@@ -178,9 +178,9 @@ static int _build_single_nodeline_info(slurm_conf_node_t *node_ptr,
/* some sanity checks */
#ifdef HAVE_FRONT_END
- if ((hostlist_count(hostname_list) != 1) ||
- (hostlist_count(address_list) != 1)) {
- error("Only one hostname and address allowed "
+ if ((hostlist_count(hostname_list) > 1) ||
+ (hostlist_count(address_list) > 1)) {
+ error("Only one NodeHostName and NodeAddr are allowed "
"in FRONT_END mode");
goto cleanup;
}
@@ -249,8 +249,9 @@ cleanup:
static int _delete_config_record (void)
{
last_node_update = time (NULL);
- (void) list_delete_all (config_list, &_list_find_config, NULL);
- (void) list_delete_all (feature_list, &_list_find_feature, NULL);
+ (void) list_delete_all (config_list, &_list_find_config, NULL);
+ (void) list_delete_all (feature_list, &_list_find_feature, NULL);
+ (void) list_delete_all (front_end_list, &list_find_frontend, NULL);
return SLURM_SUCCESS;
}
@@ -463,6 +464,91 @@ static int _list_find_feature (void *feature_entry, void *key)
return 0;
}
+/* Log the contents of a frontend record */
+static void _dump_front_end(slurm_conf_frontend_t *fe_ptr)
+{
+ debug("fe name:%s addr:%s port:%u state:%u reason:%s",
+ fe_ptr->frontends, fe_ptr->addresses,
+ fe_ptr->port, fe_ptr->node_state, fe_ptr->reason);
+}
+
+/*
+ * build_all_frontend_info - get a array of slurm_conf_frontend_t structures
+ * from the slurm.conf reader, build table, and set values
+ * RET 0 if no error, error code otherwise
+ */
+extern int build_all_frontend_info (void)
+{
+ slurm_conf_frontend_t **ptr_array;
+ slurm_conf_frontend_t *fe_single, *fe_line;
+ int i, count, max_rc = SLURM_SUCCESS;
+
+ count = slurm_conf_frontend_array(&ptr_array);
+#ifdef HAVE_FRONT_END
+ if (count == 0) {
+ verbose("No FrontendName information available!");
+ if (node_record_count == 0)
+ fatal("No nodes configured");
+ fe_single = xmalloc(sizeof(slurm_conf_frontend_t));
+ if (list_append(front_end_list, fe_single) == NULL)
+ fatal("list_append: malloc failure");
+ if (node_record_table_ptr->comm_name) {
+ fe_single->frontends = xstrdup(node_record_table_ptr->
+ comm_name);
+ fe_single->addresses = xstrdup(node_record_table_ptr->
+ comm_name);
+ } else {
+ fe_single->frontends = xstrdup(node_record_table_ptr->
+ name);
+ fe_single->addresses = xstrdup(node_record_table_ptr->
+ name);
+ }
+ fe_single->port = node_record_table_ptr->port;
+ fe_single->node_state = NODE_STATE_IDLE;
+ _dump_front_end(fe_single);
+ }
+
+ for (i = 0; i < count; i++) {
+ hostlist_t hl_name, hl_addr;
+ char *fe_name, *fe_addr;
+
+ fe_line = ptr_array[i];
+ hl_name = hostlist_create(fe_line->frontends);
+ if (hl_name == NULL)
+ fatal("Invalid FrontendName:%s", fe_line->frontends);
+ hl_addr = hostlist_create(fe_line->addresses);
+ if (hl_addr == NULL)
+ fatal("Invalid FrontendAddr:%s", fe_line->addresses);
+ if (hostlist_count(hl_name) != hostlist_count(hl_addr)) {
+ fatal("Inconsistent node count between "
+ "FrontendName(%s) and FrontendAddr(%s)",
+ fe_line->frontends, fe_line->addresses);
+ }
+ while ((fe_name = hostlist_shift(hl_name))) {
+ fe_addr = hostlist_shift(hl_addr);
+ fe_single = xmalloc(sizeof(slurm_conf_frontend_t));
+ if (list_append(front_end_list, fe_single) == NULL)
+ fatal("list_append: malloc failure");
+ fe_single->frontends = xstrdup(fe_name);
+ fe_single->addresses = xstrdup(fe_addr);
+ free(fe_name);
+ free(fe_addr);
+ fe_single->port = fe_line->port;
+ if (fe_line->reason && fe_line->reason[0])
+ fe_single->reason = xstrdup(fe_line->reason);
+ fe_single->node_state = fe_line->node_state;
+ _dump_front_end(fe_single);
+ }
+ hostlist_destroy(hl_addr);
+ hostlist_destroy(hl_name);
+ }
+#else
+ if (count != 0)
+ fatal("FrontendName information configured!");
+#endif
+ return max_rc;
+}
+
/*
* build_all_nodeline_info - get a array of slurm_conf_node_t structures
* from the slurm.conf reader, build table, and set values
@@ -704,9 +790,11 @@ extern int init_node_conf (void)
if (config_list) /* delete defunct configuration entries */
(void) _delete_config_record ();
else {
- config_list = list_create (_list_delete_config);
- feature_list = list_create (_list_delete_feature);
- if ((config_list == NULL) || (feature_list == NULL))
+ config_list = list_create (_list_delete_config);
+ feature_list = list_create (_list_delete_feature);
+ front_end_list = list_create (destroy_frontend);
+ if ((config_list == NULL) || (feature_list == NULL) ||
+ (front_end_list == NULL))
fatal("list_create malloc failure");
}
@@ -725,6 +813,8 @@ extern void node_fini2 (void)
config_list = NULL;
list_destroy(feature_list);
feature_list = NULL;
+ list_destroy(front_end_list);
+ front_end_list = NULL;
}
node_ptr = node_record_table_ptr;
@@ -838,7 +928,7 @@ extern void rehash_node (void)
}
/* Convert a node state string to it's equivalent enum value */
-extern int state_str2int(const char *state_str)
+extern int state_str2int(const char *state_str, char *node_name)
{
int state_val = NO_VAL;
int i;
@@ -858,7 +948,7 @@ extern int state_str2int(const char *state_str)
state_val = NODE_STATE_IDLE | NODE_STATE_FAIL;
}
if (state_val == NO_VAL) {
- error("invalid node state %s", state_str);
+ error("node %s has invalid state %s", node_name, state_str);
errno = EINVAL;
}
return state_val;
diff --git a/src/common/node_conf.h b/src/common/node_conf.h
index 7807359061b087ef7fc92b66717c4f5f1aecc38b..325823851e313ea28f37b48c14dfbedf0000cdd4 100644
--- a/src/common/node_conf.h
+++ b/src/common/node_conf.h
@@ -81,6 +81,8 @@ struct config_record {
};
extern List config_list; /* list of config_record entries */
+extern List front_end_list; /* list of slurm_conf_frontend_t entries */
+
struct features_record {
uint32_t magic; /* magic cookie to test data integrity */
char *name; /* name of a feature */
@@ -167,8 +169,10 @@ extern time_t last_node_update; /* time of last node record update */
*/
char * bitmap2node_name (bitstr_t *bitmap);
+extern int build_all_frontend_info (void);
+
/*
- * _build_all_nodeline_info - get a array of slurm_conf_node_t structures
+ * build_all_nodeline_info - get a array of slurm_conf_node_t structures
* from the slurm.conf reader, build table, and set values
* IN set_bitmap - if true, set node_bitmap in config record (used by slurmd)
* RET 0 if no error, error code otherwise
@@ -241,6 +245,6 @@ extern void purge_node_rec (struct node_record *node_ptr);
extern void rehash_node (void);
/* Convert a node state string to it's equivalent enum value */
-extern int state_str2int(const char *state_str);
+extern int state_str2int(const char *state_str, char *node_name);
#endif /* !_HAVE_NODE_CONF_H */
diff --git a/src/common/read_config.c b/src/common/read_config.c
index f37bb3e461a89fc28d692f1489e0b31b2c869875..50025f2e09f9d0da0a889b4a07321bd7e282f70d 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -61,21 +61,23 @@
#include <slurm/slurm.h>
#include "src/common/hostlist.h"
-#include "src/common/slurm_protocol_defs.h"
-#include "src/common/slurm_protocol_api.h"
#include "src/common/log.h"
#include "src/common/macros.h"
+#include "src/common/node_conf.h"
+#include "src/common/parse_config.h"
#include "src/common/parse_spec.h"
+#include "src/common/parse_time.h"
#include "src/common/read_config.h"
-#include "src/common/xmalloc.h"
-#include "src/common/xstring.h"
+#include "src/common/slurm_protocol_api.h"
+#include "src/common/slurm_protocol_defs.h"
#include "src/common/slurm_rlimits_info.h"
-#include "src/common/parse_config.h"
-#include "src/common/parse_time.h"
#include "src/common/slurm_selecttype_info.h"
-#include "src/common/util-net.h"
-#include "src/common/uid.h"
#include "src/common/strlcpy.h"
+#include "src/common/uid.h"
+#include "src/common/util-net.h"
+#include "src/common/xmalloc.h"
+#include "src/common/xstring.h"
+
/*
** Define slurm-specific aliases for use by plugins, see slurm_xlator.h
@@ -93,6 +95,7 @@ static s_p_hashtbl_t *conf_hashtbl = NULL;
static slurm_ctl_conf_t *conf_ptr = &slurmctld_conf;
static bool conf_initialized = false;
+static s_p_hashtbl_t *default_frontend_tbl;
static s_p_hashtbl_t *default_nodename_tbl;
static s_p_hashtbl_t *default_partition_tbl;
@@ -118,10 +121,13 @@ bool nodehash_initialized = false;
static names_ll_t *host_to_node_hashtbl[NAME_HASH_LEN] = {NULL};
static names_ll_t *node_to_host_hashtbl[NAME_HASH_LEN] = {NULL};
+static void _destroy_nodename(void *ptr);
+static int _parse_frontend(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover);
static int _parse_nodename(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover);
-static void _destroy_nodename(void *ptr);
static bool _is_valid_path(char *path, char *msg);
static int _parse_partitionname(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
@@ -282,6 +288,7 @@ s_p_options_t slurm_conf_options[] = {
{"VSizeFactor", S_P_UINT16},
{"WaitTime", S_P_UINT16},
+ {"FrontendName", S_P_ARRAY, _parse_frontend, destroy_frontend},
{"NodeName", S_P_ARRAY, _parse_nodename, _destroy_nodename},
{"PartitionName", S_P_ARRAY, _parse_partitionname,
_destroy_partitionname},
@@ -393,6 +400,80 @@ static void _set_node_prefix(const char *nodenames)
}
#endif /* SYSTEM_DIMENSIONS > 1 */
+static int _parse_frontend(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover)
+{
+ s_p_hashtbl_t *tbl, *dflt;
+ slurm_conf_frontend_t *n;
+ char *node_state = NULL;
+ static s_p_options_t _frontend_options[] = {
+ {"FrontendAddr", S_P_STRING},
+ {"Port", S_P_UINT16},
+ {"Reason", S_P_STRING},
+ {"State", S_P_STRING},
+ {NULL}
+ };
+
+#ifndef HAVE_FRONT_END
+ fatal("Use of FrontendName in slurm.conf without SLURM being "
+ "configured/built with the --have-front-end option");
+#endif
+
+ tbl = s_p_hashtbl_create(_frontend_options);
+ s_p_parse_line(tbl, *leftover, leftover);
+ /* s_p_dump_values(tbl, _frontend_options); */
+
+ if (strcasecmp(value, "DEFAULT") == 0) {
+ char *tmp;
+ if (s_p_get_string(&tmp, "FrontendAddr", tbl)) {
+ error("FrontendAddr not allowed with "
+ "FrontendName=DEFAULT");
+ xfree(tmp);
+ s_p_hashtbl_destroy(tbl);
+ return -1;
+ }
+
+ if (default_frontend_tbl != NULL)
+ s_p_hashtbl_destroy(default_frontend_tbl);
+ default_frontend_tbl = tbl;
+
+ return 0;
+ } else {
+ n = xmalloc(sizeof(slurm_conf_frontend_t));
+ dflt = default_frontend_tbl;
+
+ n->frontends = xstrdup(value);
+
+ if (!s_p_get_string(&n->addresses, "FrontendAddr", tbl))
+ n->addresses = xstrdup(n->frontends);
+
+ if (!s_p_get_uint16(&n->port, "Port", tbl) &&
+ !s_p_get_uint16(&n->port, "Port", dflt)) {
+ /* This gets resolved in slurm_conf_get_port()
+ * and slurm_conf_get_addr(). For now just
+ * leave with a value of zero */
+ n->port = 0;
+ }
+
+ if (!s_p_get_string(&n->reason, "Reason", tbl))
+ s_p_get_string(&n->reason, "Reason", dflt);
+
+ if (!s_p_get_string(&node_state, "State", tbl) &&
+ !s_p_get_string(&node_state, "State", dflt))
+ n->node_state = NODE_STATE_UNKNOWN;
+ else {
+ n->node_state = state_str2int(node_state, value);
+ xfree(node_state);
+ }
+
+ *dest = (void *)n;
+
+ return 1;
+ }
+
+ /* should not get here */
+}
static int _parse_nodename(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
@@ -459,10 +540,21 @@ static int _parse_nodename(void **dest, slurm_parser_enum_t type,
_set_node_prefix(n->nodenames);
#endif
+#ifdef HAVE_FRONT_END
+ if (!s_p_get_string(&n->hostnames, "NodeHostname", tbl))
+ xfree(n->hostnames);
+ if (!s_p_get_string(&n->addresses, "NodeAddr", tbl))
+ xfree(n->addresses);
+ else {
+ verbose("Use FrontendNode/FrondendAddr configuration "
+ "options instead of NodeAddr");
+ }
+#else
if (!s_p_get_string(&n->hostnames, "NodeHostname", tbl))
n->hostnames = xstrdup(n->nodenames);
if (!s_p_get_string(&n->addresses, "NodeAddr", tbl))
n->addresses = xstrdup(n->hostnames);
+#endif
if (!s_p_get_uint16(&n->cores, "CoresPerSocket", tbl)
&& !s_p_get_uint16(&n->cores, "CoresPerSocket", dflt)) {
@@ -581,6 +673,35 @@ static int _parse_nodename(void **dest, slurm_parser_enum_t type,
/* should not get here */
}
+/* Destroy a front_end record built by slurm_conf_frontend_array() */
+extern void destroy_frontend(void *ptr)
+{
+ slurm_conf_frontend_t *n = (slurm_conf_frontend_t *) ptr;
+ xfree(n->frontends);
+ xfree(n->addresses);
+ xfree(n->reason);
+ xfree(ptr);
+}
+
+/*
+ * list_find_frontend - find an entry in the front_end list, see list.h for
+ * documentation
+ * IN key - is feature name or NULL for all features
+ * RET 1 if found, 0 otherwise
+ */
+extern int list_find_frontend (void *front_end_entry, void *key)
+{
+ slurm_conf_frontend_t *front_end_ptr;
+
+ if (key == NULL)
+ return 1;
+
+ front_end_ptr = (slurm_conf_frontend_t *) front_end_entry;
+ if (strcmp(front_end_ptr->frontends, (char *) key) == 0)
+ return 1;
+ return 0;
+}
+
static void _destroy_nodename(void *ptr)
{
slurm_conf_node_t *n = (slurm_conf_node_t *)ptr;
@@ -594,6 +715,22 @@ static void _destroy_nodename(void *ptr)
xfree(ptr);
}
+int slurm_conf_frontend_array(slurm_conf_frontend_t **ptr_array[])
+{
+ int count;
+ slurm_conf_frontend_t **ptr;
+
+ if (s_p_get_array((void ***)&ptr, &count, "FrontendName",
+ conf_hashtbl)) {
+ *ptr_array = ptr;
+ return count;
+ } else {
+ *ptr_array = NULL;
+ return 0;
+ }
+}
+
+
int slurm_conf_nodename_array(slurm_conf_node_t **ptr_array[])
{
int count;
@@ -608,7 +745,6 @@ int slurm_conf_nodename_array(slurm_conf_node_t **ptr_array[])
}
}
-
static int _parse_partitionname(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
@@ -939,10 +1075,14 @@ static void _init_name_hashtbl()
static int _get_hash_idx(const char *s)
{
- int i;
+ int i = 0;
+
+ if (s) {
+ while (*s) {
+ i += (int)*s++;
+ }
+ }
- i = 0;
- while (*s) i += (int)*s++;
return i % NAME_HASH_LEN;
}
@@ -961,8 +1101,8 @@ static void _push_to_hashtbls(char *alias, char *hostname,
/* Ensure only one slurmd configured on each host */
p = host_to_node_hashtbl[hostname_idx];
while (p) {
- if (strcmp(p->hostname, hostname)==0) {
- error("Duplicated NodeHostname %s in the config file",
+ if (strcmp(p->hostname, hostname) == 0) {
+ error("Duplicated NodeHostName %s in the config file",
hostname);
return;
}
@@ -1029,7 +1169,7 @@ static int _register_conf_node_aliases(slurm_conf_node_t *node_ptr)
char *address = NULL;
int error_code = SLURM_SUCCESS;
- if (node_ptr->nodenames == NULL || *node_ptr->nodenames == '\0')
+ if ((node_ptr->nodenames == NULL) || (node_ptr->nodenames[0] == '\0'))
return -1;
if ((alias_list = hostlist_create(node_ptr->nodenames)) == NULL) {
@@ -1058,10 +1198,11 @@ static int _register_conf_node_aliases(slurm_conf_node_t *node_ptr)
/* some sanity checks */
#ifdef HAVE_FRONT_END
- if (hostlist_count(hostname_list) != 1
- || hostlist_count(address_list) != 1) {
- error("Only one hostname and address allowed "
+ if ((hostlist_count(hostname_list) > 1) ||
+ (hostlist_count(address_list) > 1)) {
+ error("Only one NodeHostName and NodeAddr allowed "
"in FRONT_END mode");
+ error("Use FrontendNode/FrondendAddr configuration options");
goto cleanup;
}
@@ -1120,15 +1261,14 @@ static void _init_slurmd_nodehash(void)
else
nodehash_initialized = true;
- if(!conf_initialized) {
+ if (!conf_initialized) {
_init_slurm_conf(NULL);
conf_initialized = true;
}
count = slurm_conf_nodename_array(&ptr_array);
- if (count == 0) {
+ if (count == 0)
return;
- }
for (i = 0; i < count; i++) {
_register_conf_node_aliases(ptr_array[i]);
@@ -1179,6 +1319,30 @@ extern char *slurm_conf_get_hostname(const char *node_name)
*/
extern char *slurm_conf_get_nodename(const char *node_hostname)
{
+ names_ll_t *p;
+ char *alias = NULL;
+#ifdef HAVE_FRONT_END
+ slurm_conf_frontend_t *front_end_ptr = NULL;
+
+ slurm_conf_lock();
+ _init_slurmd_nodehash();
+
+ if (!front_end_list) {
+ error("front_end_list is NULL");
+ } else {
+ front_end_ptr = list_find_first(front_end_list,
+ list_find_frontend,
+ (char *) node_hostname);
+ }
+ if (front_end_ptr) {
+ p = host_to_node_hashtbl[0];
+ if (p)
+ alias = xstrdup(p->alias);
+ }
+ slurm_conf_unlock();
+
+ return alias;
+#else
int idx;
names_ll_t *p;
@@ -1189,15 +1353,15 @@ extern char *slurm_conf_get_nodename(const char *node_hostname)
p = host_to_node_hashtbl[idx];
while (p) {
if (strcmp(p->hostname, node_hostname) == 0) {
- char *alias = xstrdup(p->alias);
- slurm_conf_unlock();
- return alias;
+ alias = xstrdup(p->alias);
+ break
}
p = p->next_hostname;
}
slurm_conf_unlock();
- return NULL;
+ return alias
+#endif
}
/*
@@ -1677,6 +1841,10 @@ static void
_destroy_slurm_conf(void)
{
s_p_hashtbl_destroy(conf_hashtbl);
+ if (default_frontend_tbl != NULL) {
+ s_p_hashtbl_destroy(default_frontend_tbl);
+ default_frontend_tbl = NULL;
+ }
if (default_nodename_tbl != NULL) {
s_p_hashtbl_destroy(default_nodename_tbl);
default_nodename_tbl = NULL;
diff --git a/src/common/read_config.h b/src/common/read_config.h
index ffe660bccf3266e1a663690bb4985e77aa5a7149..7c2f600b3d06c6ed1dc826661555dda2fadab4d5 100644
--- a/src/common/read_config.h
+++ b/src/common/read_config.h
@@ -43,6 +43,7 @@
#ifndef _READ_CONFIG_H
#define _READ_CONFIG_H
+#include "src/common/list.h"
#include "src/common/slurm_protocol_defs.h"
#include "src/common/slurm_protocol_socket_common.h"
#include "src/common/parse_config.h"
@@ -151,6 +152,16 @@ extern char *default_plugstack;
#define DEFAULT_UNKILLABLE_TIMEOUT 60 /* seconds */
#define DEFAULT_MAX_TASKS_PER_NODE 128
+typedef struct slurm_conf_frontend {
+ char *frontends; /* frontend node name */
+ char *addresses; /* frontend node address */
+ uint16_t port; /* frontend specific port */
+ char *reason; /* reason for down frontend node */
+ uint16_t node_state; /* enum node_states, ORed with
+ * NODE_STATE_NO_RESPOND if not
+ * responding */
+} slurm_conf_frontend_t;
+
typedef struct slurm_conf_node {
char *nodenames;
char *hostnames;
@@ -210,6 +221,17 @@ typedef struct {
char *value;
} config_key_pair_t;
+/* Destroy a front_end record built by slurm_conf_frontend_array() */
+extern void destroy_frontend(void *ptr);
+
+/*
+ * list_find_frontend - find an entry in the front_end list, see list.h for
+ * documentation
+ * IN key - is feature name or NULL for all features
+ * RET 1 if found, 0 otherwise
+ */
+extern int list_find_frontend (void *front_end_entry, void *key);
+
/*
* slurm_conf_init - load the slurm configuration from the a file.
* IN file_name - name of the slurm configuration file to be read
@@ -257,6 +279,15 @@ extern slurm_ctl_conf_t *slurm_conf_lock(void);
extern void slurm_conf_unlock(void);
+
+/*
+ * Set "ptr_array" with the pointer to an array of pointers to
+ * slurm_conf_frontend_t structures.
+ *
+ * Return value is the length of the array.
+ */
+extern int slurm_conf_frontend_array(slurm_conf_frontend_t **ptr_array[]);
+
/*
* Set "ptr_array" with the pointer to an array of pointers to
* slurm_conf_node_t structures.
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 5379b35615c7c286ccd05fcf0e100e8fdcbc9572..faf49f9cf4e179a79d60459eb06b86d7670b7376 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -724,6 +724,7 @@ static void _pack_node (struct node_record *dump_node_ptr, Buf buffer,
*/
void set_slurmd_addr (void)
{
+#ifndef HAVE_FRONT_END
int i;
struct node_record *node_ptr = node_record_table_ptr;
DEF_TIMERS;
@@ -747,7 +748,7 @@ void set_slurmd_addr (void)
}
END_TIMER2("set_slurmd_addr");
- return;
+#endif
}
/*
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index cce432765c93058fd5157fb89207faa7ce5655b8..2276e3f17cec42258d9ca6ce964b8579fc944d5b 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -326,7 +326,7 @@ static int _handle_downnodes_line(slurm_conf_downnodes_t *down)
int state_val = NODE_STATE_DOWN;
if (down->state != NULL) {
- state_val = state_str2int(down->state);
+ state_val = state_str2int(down->state, down->nodenames);
if (state_val == NO_VAL) {
error("Invalid State \"%s\"", down->state);
goto cleanup;
@@ -398,6 +398,7 @@ static int _build_all_nodeline_info(void)
/* Load the node table here */
rc = build_all_nodeline_info(false);
+ rc = MAX(build_all_frontend_info(), rc);
/* Now perform operations on the node table as needed by slurmctld */
#ifdef HAVE_BG
diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c
index 89378901b4e689a493e19f1b525e621c481e0e3f..8e36dd61fcc5677c93bf3f7c74cf3cea2781e5d5 100644
--- a/src/slurmd/slurmd/slurmd.c
+++ b/src/slurmd/slurmd/slurmd.c
@@ -203,7 +203,7 @@ main (int argc, char *argv[])
conf->argc = &argc;
slurmd_uid = slurm_get_slurmd_user_id();
curr_uid = getuid();
- if(curr_uid != slurmd_uid) {
+ if (curr_uid != slurmd_uid) {
struct passwd *pw = NULL;
char *slurmd_user = NULL;
char *curr_user = NULL;
@@ -761,7 +761,6 @@ _read_config(void)
conf->block_map_size = 0;
- _update_logging();
_update_nice();
get_procs(&conf->actual_cpus);
@@ -785,11 +784,11 @@ _read_config(void)
conf->threads = conf->conf_threads;
}
- if(cf->fast_schedule &&
- ((conf->cpus != conf->actual_cpus) ||
- (conf->sockets != conf->actual_sockets) ||
- (conf->cores != conf->actual_cores) ||
- (conf->threads != conf->actual_threads))) {
+ if (cf->fast_schedule &&
+ ((conf->cpus != conf->actual_cpus) ||
+ (conf->sockets != conf->actual_sockets) ||
+ (conf->cores != conf->actual_cores) ||
+ (conf->threads != conf->actual_threads))) {
info("Node configuration differs from hardware\n"
" Procs=%u:%u(hw) Sockets=%u:%u(hw)\n"
" CoresPerSocket=%u:%u(hw) ThreadsPerCore=%u:%u(hw)",
@@ -847,6 +846,7 @@ _reconfigure(void)
bool did_change;
_reconfig = 0;
+ _update_logging();
_read_config();
/*
@@ -855,7 +855,6 @@ _reconfigure(void)
slurm_topo_build_config();
_set_topo_info();
- /* _update_logging(); */
_print_conf();
/*
@@ -1181,13 +1180,21 @@ _slurmd_init(void)
* an alternate location for the slurm config file.
*/
_process_cmdline(*conf->argc, *conf->argv);
+ _update_logging();
+
+ /*
+ * Build nodes table like in slurmctld
+ * This is required by the topology stack
+ */
+ init_node_conf();
+ build_all_nodeline_info(true);
+ build_all_frontend_info();
/*
* Read global slurm config file, ovverride necessary values from
* defaults and command line.
- *
*/
- _read_config();
+ _read_config(); // SETUP LOGGING EARLIER
cpu_cnt = MAX(conf->conf_cpus, conf->block_map_size);
if ((gres_plugin_init() != SLURM_SUCCESS) ||
(gres_plugin_node_config_load(cpu_cnt) != SLURM_SUCCESS))
@@ -1195,25 +1202,12 @@ _slurmd_init(void)
if (slurm_topo_init() != SLURM_SUCCESS)
return SLURM_FAILURE;
- /*
- * Build nodes table like in slurmctld
- * This is required by the topology stack
- */
- init_node_conf();
- build_all_nodeline_info(true);
-
/*
* Get and set slurmd topology information
*/
slurm_topo_build_config();
_set_topo_info();
- /*
- * Update location of log messages (syslog, stderr, logfile, etc.),
- * print current configuration (if in debug mode), and
- * load appropriate plugin(s).
- */
- /* _update_logging(); */
_print_conf();
if (slurm_proctrack_init() != SLURM_SUCCESS)
return SLURM_FAILURE;