From 62631b3fa3f172b716f9860291bb0da075003b3d Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Fri, 8 Jan 2016 15:51:47 -0800
Subject: [PATCH] Add active feature list infrastructure
---
src/common/node_conf.c | 152 ++++++++++++++++++++++++++++++++++--
src/common/node_conf.h | 13 ++-
src/slurmctld/node_mgr.c | 76 +++++++++++++-----
src/slurmctld/read_config.c | 12 ++-
4 files changed, 221 insertions(+), 32 deletions(-)
diff --git a/src/common/node_conf.c b/src/common/node_conf.c
index 4340c21a866..1544686b7e1 100644
--- a/src/common/node_conf.c
+++ b/src/common/node_conf.c
@@ -89,13 +89,17 @@ time_t last_node_update = (time_t) 0; /* time of last update */
struct node_record *node_record_table_ptr = NULL; /* node records */
xhash_t* node_hash_table = NULL;
int node_record_count = 0; /* count in node_record_table_ptr */
-
uint16_t *cr_node_num_cores = NULL;
uint32_t *cr_node_cores_offset = NULL;
-static void _add_config_feature(char *feature, bitstr_t *node_bitmap);
+/* Local function defiitions */
+static void _add_config_feature(List feature_list, char *feature,
+ bitstr_t *node_bitmap);
+static void _add_config_feature_inx(List feature_list, char *feature,
+ int node_inx);
static int _build_single_nodeline_info(slurm_conf_node_t *node_ptr,
struct config_record *config_ptr);
+static void _copy_feature_list(void);
static int _delete_config_record (void);
#if _DEBUG
static void _dump_hash (void);
@@ -110,14 +114,15 @@ 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)
+static void _add_config_feature(List feature_list, char *feature,
+ bitstr_t *node_bitmap)
{
node_feature_t *feature_ptr;
ListIterator feature_iter;
bool match = false;
/* If feature already in avail_feature_list, just update the bitmap */
- feature_iter = list_iterator_create(avail_feature_list);
+ feature_iter = list_iterator_create(feature_list);
while ((feature_ptr = (node_feature_t *) list_next(feature_iter))) {
if (strcmp(feature, feature_ptr->name))
continue;
@@ -132,10 +137,37 @@ static void _add_config_feature(char *feature, bitstr_t *node_bitmap)
feature_ptr->magic = FEATURE_MAGIC;
feature_ptr->name = xstrdup(feature);
feature_ptr->node_bitmap = bit_copy(node_bitmap);
- list_append(avail_feature_list, feature_ptr);
+ list_append(feature_list, feature_ptr);
}
}
+static void _add_config_feature_inx(List feature_list, char *feature,
+ int node_inx)
+{
+ node_feature_t *feature_ptr;
+ ListIterator feature_iter;
+ bool match = false;
+
+ /* If feature already in avail_feature_list, just update the bitmap */
+ feature_iter = list_iterator_create(feature_list);
+ while ((feature_ptr = (node_feature_t *) list_next(feature_iter))) {
+ if (strcmp(feature, feature_ptr->name))
+ continue;
+ bit_set(feature_ptr->node_bitmap, node_inx);
+ match = true;
+ break;
+ }
+ list_iterator_destroy(feature_iter);
+
+ if (!match) { /* Need to create new avail_feature_list record */
+ feature_ptr = xmalloc(sizeof(node_feature_t));
+ feature_ptr->magic = FEATURE_MAGIC;
+ feature_ptr->name = xstrdup(feature);
+ feature_ptr->node_bitmap = bit_alloc(node_record_count);
+ bit_set(feature_ptr->node_bitmap, node_inx);
+ list_append(feature_list, feature_ptr);
+ }
+}
/*
* _build_single_nodeline_info - From the slurm.conf reader, build table,
@@ -692,8 +724,111 @@ extern int build_all_nodeline_info (bool set_bitmap)
return max_rc;
}
-/* Given a config_record with it's bitmap already set,
- * build avail_feature_list */
+/* Rebuild active_feature_list for given node bitmap */
+extern void build_active_feature_list(bitstr_t *node_bitmap,
+ char *active_features)
+{
+ node_feature_t *feature_ptr;
+ ListIterator feature_iter;
+ char *tmp_str, *token, *last = NULL;
+
+ /* Clear these nodes from the feature_list record,
+ * then restore as needed */
+ feature_iter = list_iterator_create(active_feature_list);
+ bit_not(node_bitmap);
+ while ((feature_ptr = (node_feature_t *) list_next(feature_iter))) {
+ bit_and(feature_ptr->node_bitmap, node_bitmap);
+ }
+ list_iterator_destroy(feature_iter);
+ bit_not(node_bitmap);
+
+ if (active_features) {
+ tmp_str = xstrdup(active_features);
+ token = strtok_r(tmp_str, ",", &last);
+ while (token) {
+ _add_config_feature(active_feature_list, token,
+ node_bitmap);
+ token = strtok_r(NULL, ",", &last);
+ }
+ xfree(tmp_str);
+ }
+}
+
+/* Clear active_feature_list,
+ * then copy avail_feature_list into active_feature_list */
+static void _copy_feature_list(void)
+{
+ node_feature_t *active_feature_ptr, *avail_feature_ptr;
+ ListIterator feature_iter;
+
+ (void) list_delete_all(active_feature_list, &_list_find_feature, NULL);
+
+ feature_iter = list_iterator_create(avail_feature_list);
+ while ((avail_feature_ptr = (node_feature_t *)list_next(feature_iter))){
+ active_feature_ptr = xmalloc(sizeof(node_feature_t));
+ active_feature_ptr->magic = FEATURE_MAGIC;
+ active_feature_ptr->name = xstrdup(avail_feature_ptr->name);
+ active_feature_ptr->node_bitmap =
+ bit_copy(avail_feature_ptr->node_bitmap);
+ list_append(active_feature_list, active_feature_ptr);
+ }
+ list_iterator_destroy(feature_iter);
+}
+
+/* Rebuild active_feature_list for given node index,
+ * IN node_inx - Node index, if -1 then copy alloc_feature_list into
+ * acitve_feature_list, if -2 then log state
+ */
+extern void build_active_feature_list2(int node_inx, char *active_features)
+{
+ node_feature_t *feature_ptr;
+ ListIterator feature_iter;
+ char *tmp_str, *token, *last = NULL;
+
+ if (node_inx == -1) {
+ _copy_feature_list();
+ return;
+ }
+ if (node_inx == -2) {
+#if _DEBUG
+ feature_iter = list_iterator_create(active_feature_list);
+ while ((feature_ptr = (node_feature_t *)
+ list_next(feature_iter))) {
+ info("ACTIVE FEATURE: NAME:%s CNT:%d",
+ feature_ptr->name,
+ bit_set_count(feature_ptr->node_bitmap));
+ }
+ list_iterator_destroy(feature_iter);
+#endif
+ return;
+ }
+
+ if ((node_inx < 0) || (node_inx >= node_record_count)) {
+ error("%s: Invalid node_inx:%d", __func__, node_inx);
+ return;
+ }
+
+ /* Clear this node from the feature_list record,
+ * then restore as needed */
+ feature_iter = list_iterator_create(active_feature_list);
+ while ((feature_ptr = (node_feature_t *) list_next(feature_iter))) {
+ bit_clear(feature_ptr->node_bitmap, node_inx);
+ }
+ list_iterator_destroy(feature_iter);
+
+ if (active_features) {
+ tmp_str = xstrdup(active_features);
+ token = strtok_r(tmp_str, ",", &last);
+ while (token) {
+ _add_config_feature_inx(active_feature_list, token,
+ node_inx);
+ token = strtok_r(NULL, ",", &last);
+ }
+ xfree(tmp_str);
+ }
+}
+
+/* Rebuild avail_feature_list for given node configuration structure */
extern void build_avail_feature_list(struct config_record *config_ptr)
{
node_feature_t *feature_ptr;
@@ -714,7 +849,8 @@ extern void build_avail_feature_list(struct config_record *config_ptr)
tmp_str = xstrdup(config_ptr->feature);
token = strtok_r(tmp_str, ",", &last);
while (token) {
- _add_config_feature(token, config_ptr->node_bitmap);
+ _add_config_feature(avail_feature_list, token,
+ config_ptr->node_bitmap);
token = strtok_r(NULL, ",", &last);
}
xfree(tmp_str);
diff --git a/src/common/node_conf.h b/src/common/node_conf.h
index bef76371e69..78efa433b68 100644
--- a/src/common/node_conf.h
+++ b/src/common/node_conf.h
@@ -240,8 +240,17 @@ extern int build_all_nodeline_info (bool set_bitmap);
*/
extern int build_all_frontend_info (bool is_slurmd_context);
-/* Given a config_record with it's bitmap already set,
- * build avail_feature_list */
+/* Rebuild active_feature_list for given node bitmap */
+extern void build_active_feature_list(bitstr_t *node_bitmap,
+ char *active_features);
+
+/* Rebuild active_feature_list for given node index,
+ * IN node_inx - Node index, if -1 then copy alloc_feature_list into
+ * acitve_feature_list, if -2 then log state
+ */
+extern void build_active_feature_list2(int node_inx, char *active_features);
+
+/* Rebuild avail_feature_list for given node configuration structure */
extern void build_avail_feature_list(struct config_record *config_ptr);
/*
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index d76bda53087..780a9489705 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -111,7 +111,10 @@ static void _pack_node(struct node_record *dump_node_ptr, Buf buffer,
static void _sync_bitmaps(struct node_record *node_ptr, int job_count);
static void _update_config_ptr(bitstr_t *bitmap,
struct config_record *config_ptr);
-static int _update_node_features(char *node_names, char *features);
+static int _update_node_active_features(char *node_names,
+ char *active_features);
+static int _update_node_avail_features(char *node_names,
+ char *avail_features);
static int _update_node_gres(char *node_names, char *gres);
static int _update_node_weight(char *node_names, uint32_t weight);
static bool _valid_node_state_change(uint32_t old, uint32_t new);
@@ -1347,7 +1350,8 @@ int update_node ( update_node_msg_t * update_node_msg )
node_ptr->features =
xstrdup(update_node_msg->features);
}
- /* _update_node_features() logs and updates config */
+ /* _update_node_avail_features() logs and updates
+ * avail_feature_list */
}
if (update_node_msg->features_act &&
@@ -1363,7 +1367,8 @@ int update_node ( update_node_msg_t * update_node_msg )
node_ptr->features_act =
xstrdup(update_node_msg->features_act);
}
- /* _update_node_features() logs and updates config */
+ /* _update_node_active_features() logs and updates
+ * active_feature_list */
}
if (update_node_msg->gres) {
@@ -1614,9 +1619,15 @@ int update_node ( update_node_msg_t * update_node_msg )
FREE_NULL_HOSTLIST(hostname_list);
last_node_update = now;
+ if ((error_code == 0) && (update_node_msg->features_act)) {
+ error_code = _update_node_active_features(
+ update_node_msg->node_names,
+ update_node_msg->features_act);
+ }
if ((error_code == 0) && (update_node_msg->features)) {
- error_code = _update_node_features(update_node_msg->node_names,
- update_node_msg->features);
+ error_code = _update_node_avail_features(
+ update_node_msg->node_names,
+ update_node_msg->features);
}
if ((error_code == 0) && (update_node_msg->gres)) {
error_code = _update_node_gres(update_node_msg->node_names,
@@ -1668,8 +1679,8 @@ extern void restore_node_features(int recover)
error("Node %s Features(%s) differ from slurm.conf",
node_ptr->name, node_ptr->features);
if (recover == 2) {
- _update_node_features(node_ptr->name,
- node_ptr->features);
+ _update_node_avail_features(node_ptr->name,
+ node_ptr->features);
} else {
xfree(node_ptr->features);
node_ptr->features = xstrdup(node_ptr->
@@ -1677,7 +1688,7 @@ extern void restore_node_features(int recover)
feature);
}
}
-
+//FIXME ??
/* We lose the gres information updated manually and always
* use the information from slurm.conf */
(void) gres_plugin_node_reconfig(node_ptr->name,
@@ -1762,7 +1773,6 @@ static int _update_node_weight(char *node_names, uint32_t weight)
new_config_ptr->node_bitmap = bit_copy(tmp_bitmap);
new_config_ptr->nodes = bitmap2node_name(tmp_bitmap);
- build_avail_feature_list(new_config_ptr);
_update_config_ptr(tmp_bitmap, new_config_ptr);
/* Update remaining records */
@@ -1783,13 +1793,37 @@ static int _update_node_weight(char *node_names, uint32_t weight)
}
/*
- * _update_node_features - Update features associated with nodes
- * build new config list records as needed
+ * _update_node_active_features - Update active features associated with nodes
* IN node_names - List of nodes to update
- * IN features - New features value
+ * IN active_features - New active features value
* RET: SLURM_SUCCESS or error code
*/
-static int _update_node_features(char *node_names, char *features)
+static int _update_node_active_features(char *node_names, char *active_features)
+{
+ bitstr_t *node_bitmap = NULL;
+ int rc;
+
+ rc = node_name2bitmap(node_names, false, &node_bitmap);
+ if (rc) {
+ info("%s: invalid node_name (%s)", __func__, node_names);
+ return rc;
+ }
+ build_active_feature_list(node_bitmap, active_features);
+ FREE_NULL_BITMAP(node_bitmap);
+
+ info("%s: nodes %s active features set to: %s",
+ __func__, node_names, active_features);
+ return SLURM_SUCCESS;
+}
+
+/*
+ * _update_node_avail_features - Update available features associated with
+ * nodes, build new config list records as needed
+ * IN node_names - List of nodes to update
+ * IN avail_features - New available features value
+ * RET: SLURM_SUCCESS or error code
+ */
+static int _update_node_avail_features(char *node_names, char *avail_features)
{
bitstr_t *node_bitmap = NULL, *tmp_bitmap;
ListIterator config_iterator;
@@ -1799,7 +1833,7 @@ static int _update_node_features(char *node_names, char *features)
rc = node_name2bitmap(node_names, false, &node_bitmap);
if (rc) {
- info("_update_node_features: invalid node_name");
+ info("%s: invalid node_name (%s)", __func__, node_names);
return rc;
}
@@ -1821,8 +1855,8 @@ static int _update_node_features(char *node_names, char *features)
} else if (tmp_cnt == config_cnt) {
/* all nodes changed, update in situ */
xfree(config_ptr->feature);
- if (features && features[0])
- config_ptr->feature = xstrdup(features);
+ if (avail_features && avail_features[0])
+ config_ptr->feature = xstrdup(avail_features);
build_avail_feature_list(config_ptr);
} else {
/* partial update, split config_record */
@@ -1830,8 +1864,10 @@ static int _update_node_features(char *node_names, char *features)
if (first_new == NULL)
first_new = new_config_ptr;
xfree(new_config_ptr->feature);
- if (features && features[0])
- new_config_ptr->feature = xstrdup(features);
+ if (avail_features && avail_features[0]) {
+ new_config_ptr->feature =
+ xstrdup(avail_features);
+ }
new_config_ptr->node_bitmap = bit_copy(tmp_bitmap);
new_config_ptr->nodes = bitmap2node_name(tmp_bitmap);
@@ -1850,8 +1886,8 @@ static int _update_node_features(char *node_names, char *features)
list_iterator_destroy(config_iterator);
FREE_NULL_BITMAP(node_bitmap);
- info("_update_node_features: nodes %s features set to: %s",
- node_names, features);
+ info("%s: nodes %s available features set to: %s",
+ __func__, node_names, avail_features);
return SLURM_SUCCESS;
}
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 196eff948f0..b2d9d03fad0 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -339,8 +339,8 @@ static int _build_bitmaps(void)
/* scan all nodes and identify which are up, idle and
* their configuration, resync DRAINED vs. DRAINING state */
- for (i=0, node_ptr=node_record_table_ptr;
- i<node_record_count; i++, node_ptr++) {
+ for (i = 0, node_ptr = node_record_table_ptr;
+ i < node_record_count; i++, node_ptr++) {
uint32_t drain_flag, job_cnt;
if (node_ptr->name[0] == '\0')
@@ -366,12 +366,20 @@ static int _build_bitmaps(void)
bit_set(node_ptr->config_ptr->node_bitmap, i);
}
+ /* Build active and available feature lists used for scheduling */
config_iterator = list_iterator_create(config_list);
while ((config_ptr = (struct config_record *)
list_next(config_iterator))) {
build_avail_feature_list(config_ptr);
}
list_iterator_destroy(config_iterator);
+ build_active_feature_list2(-1, NULL); /* Copy avail list to active */
+ for (i = 0, node_ptr = node_record_table_ptr;
+ i < node_record_count; i++, node_ptr++) {
+ if (node_ptr->features_act)
+ build_active_feature_list2(i, node_ptr->features_act);
+ }
+ build_active_feature_list2(-2, NULL); /* Log active list */
return error_code;
}
--
GitLab