static void
apply_master_prefs(resource_t * rsc)
{
int score, new_score;
GListPtr gIter = rsc->children;
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
if (clone_data->applied_master_prefs) {
/* Make sure we only do this once */
return;
}
clone_data->applied_master_prefs = TRUE;
for (; gIter != NULL; gIter = gIter->next) {
GHashTableIter iter;
node_t *node = NULL;
resource_t *child_rsc = (resource_t *) gIter->data;
g_hash_table_iter_init(&iter, child_rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if (can_run_resources(node) == FALSE) {
/* This node will never be promoted to master,
* so don't apply the master score as that may
* lead to clone shuffling
*/
continue;
}
score = master_score(child_rsc, node, 0);
if (score > 0) {
new_score = merge_weights(node->weight, score);
if (new_score != node->weight) {
crm_trace("\t%s: Updating preference for %s (%d->%d)",
child_rsc->id, node->details->uname, node->weight, new_score);
node->weight = new_score;
}
}
new_score = max(child_rsc->priority, score);
if (new_score != child_rsc->priority) {
crm_trace("\t%s: Updating priority (%d->%d)",
child_rsc->id, child_rsc->priority, new_score);
child_rsc->priority = new_score;
}
}
}
}
static node_t *
can_run_instance(resource_t * rsc, node_t * node, int limit)
{
node_t *local_node = NULL;
if (node == NULL && rsc->allowed_nodes) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&local_node)) {
can_run_instance(rsc, local_node, limit);
}
return NULL;
}
if (can_run_resources(node) == FALSE) {
goto bail;
} else if (is_set(rsc->flags, pe_rsc_orphan)) {
goto bail;
}
local_node = parent_node_instance(rsc, node);
if (local_node == NULL) {
crm_warn("%s cannot run on %s: node not allowed", rsc->id, node->details->uname);
goto bail;
} else if (local_node->weight < 0) {
common_update_score(rsc, node->details->id, local_node->weight);
pe_rsc_trace(rsc, "%s cannot run on %s: Parent node weight doesn't allow it.",
rsc->id, node->details->uname);
} else if (local_node->count < limit) {
pe_rsc_trace(rsc, "%s can run on %s (already running %d)",
rsc->id, node->details->uname, local_node->count);
return local_node;
} else {
pe_rsc_trace(rsc, "%s cannot run on %s: node full (%d >= %d)",
rsc->id, node->details->uname, local_node->count, limit);
}
bail:
if (node) {
common_update_score(rsc, node->details->id, -INFINITY);
}
return NULL;
}
static node_t *
can_be_master(resource_t * rsc)
{
node_t *node = NULL;
node_t *local_node = NULL;
resource_t *parent = uber_parent(rsc);
clone_variant_data_t *clone_data = NULL;
#if 0
enum rsc_role_e role = RSC_ROLE_UNKNOWN;
role = rsc->fns->state(rsc, FALSE);
crm_info("%s role: %s", rsc->id, role2text(role));
#endif
if (rsc->children) {
GListPtr gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
if (can_be_master(child) == NULL) {
crm_trace( "Child %s of %s can't be promoted", child->id, rsc->id);
return NULL;
}
}
}
node = rsc->fns->location(rsc, NULL, FALSE);
if (node == NULL) {
crm_trace( "%s cannot be master: not allocated", rsc->id);
return NULL;
} else if (is_not_set(rsc->flags, pe_rsc_managed)) {
if (rsc->fns->state(rsc, TRUE) == RSC_ROLE_MASTER) {
crm_notice("Forcing unmanaged master %s to remain promoted on %s",
rsc->id, node->details->uname);
} else {
return NULL;
}
} else if (rsc->priority < 0) {
crm_trace( "%s cannot be master: preference: %d", rsc->id, rsc->priority);
return NULL;
} else if (can_run_resources(node) == FALSE) {
crm_trace( "Node cant run any resources: %s", node->details->uname);
return NULL;
}
get_clone_variant_data(clone_data, parent);
local_node = pe_hash_table_lookup(parent->allowed_nodes, node->details->id);
if (local_node == NULL) {
crm_err("%s cannot run on %s: node not allowed", rsc->id, node->details->uname);
return NULL;
} else if (local_node->count < clone_data->master_node_max
|| is_not_set(rsc->flags, pe_rsc_managed)) {
return local_node;
} else {
crm_trace( "%s cannot be master on %s: node full",
rsc->id, node->details->uname);
}
return NULL;
}
gint
sort_clone_instance(gconstpointer a, gconstpointer b, gpointer data_set)
{
int rc = 0;
node_t *node1 = NULL;
node_t *node2 = NULL;
node_t *current_node1 = NULL;
node_t *current_node2 = NULL;
unsigned int nnodes1 = 0;
unsigned int nnodes2 = 0;
gboolean can1 = TRUE;
gboolean can2 = TRUE;
const resource_t *resource1 = (const resource_t *)a;
const resource_t *resource2 = (const resource_t *)b;
CRM_ASSERT(resource1 != NULL);
CRM_ASSERT(resource2 != NULL);
/* allocation order:
* - active instances
* - instances running on nodes with the least copies
* - active instances on nodes that can't support them or are to be fenced
* - failed instances
* - inactive instances
*/
current_node1 = pe__find_active_on(resource1, &nnodes1, NULL);
current_node2 = pe__find_active_on(resource2, &nnodes2, NULL);
if (nnodes1 && nnodes2) {
if (nnodes1 < nnodes2) {
crm_trace("%s < %s: running_on", resource1->id, resource2->id);
return -1;
} else if (nnodes1 > nnodes2) {
crm_trace("%s > %s: running_on", resource1->id, resource2->id);
return 1;
}
}
node1 = current_node1;
node2 = current_node2;
if (node1) {
node_t *match = pe_hash_table_lookup(resource1->allowed_nodes, node1->details->id);
if (match == NULL || match->weight < 0) {
crm_trace("%s: current location is unavailable", resource1->id);
node1 = NULL;
can1 = FALSE;
}
}
if (node2) {
node_t *match = pe_hash_table_lookup(resource2->allowed_nodes, node2->details->id);
if (match == NULL || match->weight < 0) {
crm_trace("%s: current location is unavailable", resource2->id);
node2 = NULL;
can2 = FALSE;
}
}
if (can1 != can2) {
if (can1) {
crm_trace("%s < %s: availability of current location", resource1->id, resource2->id);
return -1;
}
crm_trace("%s > %s: availability of current location", resource1->id, resource2->id);
return 1;
}
if (resource1->priority < resource2->priority) {
crm_trace("%s < %s: priority", resource1->id, resource2->id);
return 1;
} else if (resource1->priority > resource2->priority) {
crm_trace("%s > %s: priority", resource1->id, resource2->id);
return -1;
}
if (node1 == NULL && node2 == NULL) {
crm_trace("%s == %s: not active", resource1->id, resource2->id);
return 0;
}
if (node1 != node2) {
if (node1 == NULL) {
crm_trace("%s > %s: active", resource1->id, resource2->id);
return 1;
} else if (node2 == NULL) {
crm_trace("%s < %s: active", resource1->id, resource2->id);
return -1;
}
}
can1 = can_run_resources(node1);
can2 = can_run_resources(node2);
if (can1 != can2) {
if (can1) {
crm_trace("%s < %s: can", resource1->id, resource2->id);
return -1;
}
crm_trace("%s > %s: can", resource1->id, resource2->id);
return 1;
}
node1 = parent_node_instance(resource1, node1);
node2 = parent_node_instance(resource2, node2);
if (node1 != NULL && node2 == NULL) {
crm_trace("%s < %s: not allowed", resource1->id, resource2->id);
return -1;
} else if (node1 == NULL && node2 != NULL) {
crm_trace("%s > %s: not allowed", resource1->id, resource2->id);
return 1;
}
if (node1 == NULL || node2 == NULL) {
crm_trace("%s == %s: not allowed", resource1->id, resource2->id);
return 0;
}
if (node1->count < node2->count) {
crm_trace("%s < %s: count", resource1->id, resource2->id);
return -1;
} else if (node1->count > node2->count) {
crm_trace("%s > %s: count", resource1->id, resource2->id);
return 1;
}
can1 = did_fail(resource1);
can2 = did_fail(resource2);
if (can1 != can2) {
if (can1) {
crm_trace("%s > %s: failed", resource1->id, resource2->id);
return 1;
}
crm_trace("%s < %s: failed", resource1->id, resource2->id);
return -1;
}
if (node1 && node2) {
int lpc = 0;
int max = 0;
node_t *n = NULL;
GListPtr gIter = NULL;
GListPtr list1 = NULL;
GListPtr list2 = NULL;
GHashTable *hash1 =
g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free);
GHashTable *hash2 =
g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free);
n = node_copy(current_node1);
g_hash_table_insert(hash1, (gpointer) n->details->id, n);
n = node_copy(current_node2);
g_hash_table_insert(hash2, (gpointer) n->details->id, n);
if(resource1->parent) {
for (gIter = resource1->parent->rsc_cons; gIter; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
crm_trace("Applying %s to %s", constraint->id, resource1->id);
hash1 = native_merge_weights(constraint->rsc_rh, resource1->id, hash1,
constraint->node_attribute,
(float)constraint->score / INFINITY, 0);
}
for (gIter = resource1->parent->rsc_cons_lhs; gIter; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
crm_trace("Applying %s to %s", constraint->id, resource1->id);
hash1 = native_merge_weights(constraint->rsc_lh, resource1->id, hash1,
constraint->node_attribute,
(float)constraint->score / INFINITY, pe_weights_positive);
}
}
if(resource2->parent) {
for (gIter = resource2->parent->rsc_cons; gIter; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
crm_trace("Applying %s to %s", constraint->id, resource2->id);
hash2 = native_merge_weights(constraint->rsc_rh, resource2->id, hash2,
constraint->node_attribute,
(float)constraint->score / INFINITY, 0);
}
for (gIter = resource2->parent->rsc_cons_lhs; gIter; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
crm_trace("Applying %s to %s", constraint->id, resource2->id);
hash2 = native_merge_weights(constraint->rsc_lh, resource2->id, hash2,
constraint->node_attribute,
(float)constraint->score / INFINITY, pe_weights_positive);
}
}
/* Current location score */
node1 = g_hash_table_lookup(hash1, current_node1->details->id);
node2 = g_hash_table_lookup(hash2, current_node2->details->id);
if (node1->weight < node2->weight) {
if (node1->weight < 0) {
crm_trace("%s > %s: current score: %d %d", resource1->id, resource2->id, node1->weight, node2->weight);
rc = -1;
goto out;
} else {
crm_trace("%s < %s: current score: %d %d", resource1->id, resource2->id, node1->weight, node2->weight);
rc = 1;
goto out;
}
} else if (node1->weight > node2->weight) {
crm_trace("%s > %s: current score: %d %d", resource1->id, resource2->id, node1->weight, node2->weight);
rc = -1;
goto out;
}
/* All location scores */
list1 = g_hash_table_get_values(hash1);
list2 = g_hash_table_get_values(hash2);
list1 = sort_nodes_by_weight(list1, current_node1, data_set);
list2 = sort_nodes_by_weight(list2, current_node2, data_set);
max = g_list_length(list1);
if (max < g_list_length(list2)) {
max = g_list_length(list2);
}
for (; lpc < max; lpc++) {
node1 = g_list_nth_data(list1, lpc);
node2 = g_list_nth_data(list2, lpc);
if (node1 == NULL) {
crm_trace("%s < %s: colocated score NULL", resource1->id, resource2->id);
rc = 1;
break;
} else if (node2 == NULL) {
crm_trace("%s > %s: colocated score NULL", resource1->id, resource2->id);
rc = -1;
break;
}
if (node1->weight < node2->weight) {
crm_trace("%s < %s: colocated score", resource1->id, resource2->id);
rc = 1;
break;
} else if (node1->weight > node2->weight) {
crm_trace("%s > %s: colocated score", resource1->id, resource2->id);
rc = -1;
break;
}
}
/* Order by reverse uname - same as sort_node_weight() does? */
out:
g_hash_table_destroy(hash1); /* Free mem */
g_hash_table_destroy(hash2); /* Free mem */
g_list_free(list1);
g_list_free(list2);
if (rc != 0) {
return rc;
}
}
rc = strcmp(resource1->id, resource2->id);
crm_trace("%s %c %s: default", resource1->id, rc < 0 ? '<' : '>', resource2->id);
return rc;
}
void
distribute_children(resource_t *rsc, GListPtr children, GListPtr nodes,
int max, int per_host_max, pe_working_set_t * data_set)
{
int loop_max = 0;
int allocated = 0;
int available_nodes = 0;
/* count now tracks the number of clones currently allocated */
for(GListPtr nIter = nodes; nIter != NULL; nIter = nIter->next) {
pe_node_t *node = nIter->data;
node->count = 0;
if (can_run_resources(node)) {
available_nodes++;
}
}
if(available_nodes) {
loop_max = max / available_nodes;
}
if (loop_max < 1) {
loop_max = 1;
}
pe_rsc_debug(rsc, "Allocating up to %d %s instances to a possible %d nodes (at most %d per host, %d optimal)",
max, rsc->id, available_nodes, per_host_max, loop_max);
/* Pre-allocate as many instances as we can to their current location */
for (GListPtr gIter = children; gIter != NULL && allocated < max; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
if (child->running_on && is_set(child->flags, pe_rsc_provisional)
&& is_not_set(child->flags, pe_rsc_failed)) {
node_t *child_node = pe__current_node(child);
node_t *local_node = parent_node_instance(child, child_node);
pe_rsc_trace(rsc, "Checking pre-allocation of %s to %s (%d remaining of %d)",
child->id, child_node->details->uname, max - allocated, max);
if (can_run_resources(child_node) == FALSE || child_node->weight < 0) {
pe_rsc_trace(rsc, "Not pre-allocating because %s can not run %s",
child_node->details->uname, child->id);
} else if(local_node && local_node->count >= loop_max) {
pe_rsc_trace(rsc,
"Not pre-allocating because %s already allocated optimal instances",
child_node->details->uname);
} else if (color_instance(child, child_node, max < available_nodes, per_host_max, data_set)) {
pe_rsc_trace(rsc, "Pre-allocated %s to %s", child->id,
child_node->details->uname);
allocated++;
}
}
}
pe_rsc_trace(rsc, "Done pre-allocating (%d of %d)", allocated, max);
for (GListPtr gIter = children; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
if (child->running_on != NULL) {
node_t *child_node = pe__current_node(child);
node_t *local_node = parent_node_instance(child, child_node);
if (local_node == NULL) {
crm_err("%s is running on %s which isn't allowed",
child->id, child_node->details->uname);
}
}
if (is_not_set(child->flags, pe_rsc_provisional)) {
} else if (allocated >= max) {
pe_rsc_debug(rsc, "Child %s not allocated - limit reached %d %d", child->id, allocated, max);
resource_location(child, NULL, -INFINITY, "clone_color:limit_reached", data_set);
} else {
if (color_instance(child, NULL, max < available_nodes, per_host_max, data_set)) {
allocated++;
}
}
}
pe_rsc_debug(rsc, "Allocated %d %s instances of a possible %d",
allocated, rsc->id, max);
}