ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Fix ricochet effect - move failed procs to next on list instead of loadbalancing

Просмотр исходного кода 
This commit was SVN r24413.
Этот коммит содержится в:
Ralph Castain 2011-02-18 13:11:55 +00:00
родитель b98a2917ff
Коммит 7b35ada7fc
1 изменённых файлов: 23 добавлений и 54 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

67
orte/mca/rmaps/resilient/rmaps_resilient.c
Просмотреть файл

@ -414,15 +414,18 @@ static int get_new_node(orte_proc_t *proc,
{
orte_node_t *nd, *oldnode, *node;
int rc;
orte_vpid_t totprocs;
opal_list_t node_list;
opal_list_item_t *item;
opal_list_item_t *item, *next;
orte_std_cntr_t num_slots;
/* if no ftgrps are available, then just map it on the lightest loaded
* node known to the system, avoiding the current node if possible and
* taking into account any limitations specified by user in hostfile
* and -host options
/* if no ftgrps are available, then just put it on the next node
* on the list - obviously, this is a rather unintelligent decision.
* However, we want to ensure that we don't just keep bouncing
* back/forth between the same two nodes.
*
* Note: if the list only has oldnode on it, then this installs
* the proc back on its original node - this is better than not
* restarting at all
*/
*ndret = NULL;
nd = NULL;
@ -446,13 +449,12 @@ static int get_new_node(orte_proc_t *proc,
}
/*
* Cycle thru the list to find the current node and remove it
* Cycle thru the list to find the current node
*
*/
nd = NULL;
for (item = opal_list_get_first(&node_list);
item != opal_list_get_end(&node_list);
item = opal_list_get_next(item)) {
item = opal_list_get_first(&node_list);
while (item != opal_list_get_end(&node_list)) {
next = opal_list_get_next(item);
node = (orte_node_t*)item;
OPAL_OUTPUT_VERBOSE((7, orte_rmaps_base.rmaps_output,
"%s CHECKING NODE %s[%s] AGAINST NODE %s[%s]",
@ -462,52 +464,19 @@ static int get_new_node(orte_proc_t *proc,
oldnode->name,
(NULL == oldnode->daemon) ? "?" : ORTE_VPID_PRINT(oldnode->daemon->name.vpid)));
if (node == oldnode) {
/* remove it from the list */
opal_list_remove_item(&node_list, item);
/* maintain acctg */
OBJ_RELEASE(item);
if (next == opal_list_get_end(&node_list)) {
nd = (orte_node_t*)opal_list_get_first(&node_list);
} else {
nd = (orte_node_t*)next;
}
break;
}
}
/* if the list is empty, then go ahead and install it back on
* its original node as this is better than not restarting
* at all
*/
if (0 == opal_list_get_size(&node_list)) {
nd = oldnode;
OPAL_OUTPUT_VERBOSE((1, orte_rmaps_base.rmaps_output,
"%s rmaps:resilient: Placing process back on original node %s daemon %s (no ftgrp)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
nd->name, ORTE_NAME_PRINT((&nd->daemon->name))));
} else {
/* put it on the lightest loaded node on the list */
totprocs = 1000000;
nd = NULL;
for (item = opal_list_get_first(&node_list);
item != opal_list_get_end(&node_list);
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
if (node->num_procs < totprocs) {
nd = node;
totprocs = node->num_procs;
}
}
if (NULL == nd) {
OPAL_OUTPUT_VERBOSE((1, orte_rmaps_base.rmaps_output,
"%s ERROR - NO NODES AVAILABLE",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* this can only happen if no nodes are available - quietly return */
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
rc = ORTE_ERR_OUT_OF_RESOURCE;
goto error;
}
OPAL_OUTPUT_VERBOSE((1, orte_rmaps_base.rmaps_output,
"%s rmaps:resilient: Placing process on node %s daemon %s (no ftgrp)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(nd == oldnode) ? "OLDNODE" : nd->name,
ORTE_NAME_PRINT((&nd->daemon->name))));
}
error:
while (NULL != (item = opal_list_remove_first(&node_list))) {