int orte_util_add_hostfile_nodes(opal_list_t *nodes,
char *hostfile)
{
opal_list_t exclude;
opal_list_item_t *item, *itm;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_debug_output,
"%s hostfile: checking hostfile %s for nodes",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
/* parse the hostfile and add the contents to the list */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, nodes, &exclude, false))) {
goto cleanup;
}
/* parse the nodes to check for any relative node directives */
for (item = opal_list_get_first(nodes);
item != opal_list_get_end(nodes);
item = opal_list_get_next(item)) {
orte_node_t *node=(orte_node_t*)item;
if ('+' == node->name[0]) {
orte_show_help("help-hostfile.txt", "hostfile:relative-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
}
/* remove from the list of nodes those that are in the exclude list */
while(NULL != (item = opal_list_remove_first(&exclude))) {
orte_node_t *exnode = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
orte_node_t *node=(orte_node_t*)itm;
if (0 == strcmp(exnode->name, node->name)) {
/* match - remove it */
opal_list_remove_item(nodes, itm);
OBJ_RELEASE(itm);
break;
}
}
OBJ_RELEASE(item);
}
cleanup:
OBJ_DESTRUCT(&exclude);
return rc;
}
int orte_util_get_ordered_host_list(opal_list_t *nodes,
char *hostfile)
{
opal_list_t exclude;
opal_list_item_t *item, *itm, *item2, *item1;
char *cptr;
int num_empty, i, nodeidx, startempty=0;
bool want_all_empty=false;
orte_node_t *node_from_pool, *newnode;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_ras_base_framework.framework_output,
"%s hostfile: creating ordered list of hosts from hostfile %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
/* parse the hostfile and add the contents to the list, keeping duplicates */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, nodes, &exclude, true))) {
goto cleanup;
}
/* parse the nodes to process any relative node directives */
item2 = opal_list_get_first(nodes);
while (item2 != opal_list_get_end(nodes)) {
orte_node_t *node=(orte_node_t*)item2;
/* save the next location in case this one gets removed */
item1 = opal_list_get_next(item2);
if ('+' != node->name[0]) {
item2 = item1;
continue;
}
/* see if we specified empty nodes */
if ('e' == node->name[1] ||
'E' == node->name[1]) {
/* request for empty nodes - do they want
* all of them?
*/
if (NULL != (cptr = strchr(node->name, ':'))) {
/* the colon indicates a specific # are requested */
cptr++; /* step past : */
num_empty = strtol(cptr, NULL, 10);
} else {
/* want them all - set num_empty to max */
num_empty = INT_MAX;
want_all_empty = true;
}
/* insert empty nodes into newnodes list in place of the current item.
* since item1 is the next item, we insert in front of it
*/
if (!orte_hnp_is_allocated && 0 == startempty) {
startempty = 1;
}
for (i=startempty; 0 < num_empty && i < orte_node_pool->size; i++) {
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
continue;
}
if (0 == node_from_pool->slots_inuse) {
newnode = OBJ_NEW(orte_node_t);
newnode->name = strdup(node_from_pool->name);
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node->slots < node_from_pool->slots) {
newnode->slots = node->slots;
} else {
newnode->slots = node_from_pool->slots;
}
opal_list_insert_pos(nodes, item1, &newnode->super);
/* track number added */
--num_empty;
}
}
/* bookmark where we stopped in case they ask for more */
startempty = i;
/* did they get everything they wanted? */
if (!want_all_empty && 0 < num_empty) {
orte_show_help("help-hostfile.txt", "hostfile:not-enough-empty",
true, num_empty);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* since we have expanded the provided node, remove
* it from list
*/
opal_list_remove_item(nodes, item2);
OBJ_RELEASE(item2);
} else if ('n' == node->name[1] ||
'N' == node->name[1]) {
/* they want a specific relative node #, so
* look it up on global pool
*/
nodeidx = strtol(&node->name[2], NULL, 10);
/* if the HNP is not allocated, then we need to
* adjust the index as the node pool is offset
* by one
*/
if (!orte_hnp_is_allocated) {
nodeidx++;
}
/* see if that location is filled */
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, nodeidx))) {
/* this is an error */
orte_show_help("help-hostfile.txt", "hostfile:relative-node-not-found",
true, nodeidx, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* create the node object */
newnode = OBJ_NEW(orte_node_t);
newnode->name = strdup(node_from_pool->name);
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node->slots < node_from_pool->slots) {
newnode->slots = node->slots;
} else {
newnode->slots = node_from_pool->slots;
}
/* insert it before item1 */
opal_list_insert_pos(nodes, item1, &newnode->super);
/* since we have expanded the provided node, remove
* it from list
*/
opal_list_remove_item(nodes, item2);
OBJ_RELEASE(item2);
} else {
/* invalid relative node syntax */
orte_show_help("help-hostfile.txt", "hostfile:invalid-relative-node-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* move to next */
item2 = item1;
}
/* remove from the list of nodes those that are in the exclude list */
while(NULL != (item = opal_list_remove_first(&exclude))) {
orte_node_t *exnode = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
orte_node_t *node=(orte_node_t*)itm;
if (0 == strcmp(exnode->name, node->name)) {
/* match - remove it */
opal_list_remove_item(nodes, itm);
OBJ_RELEASE(itm);
/* have to cycle through the entire list as we could
* have duplicates
*/
}
}
OBJ_RELEASE(item);
}
cleanup:
OBJ_DESTRUCT(&exclude);
return rc;
}
/* Parse the provided hostfile and filter the nodes that are
* on the input list, removing those that
* are not found in the hostfile
*/
int orte_util_filter_hostfile_nodes(opal_list_t *nodes,
char *hostfile,
bool remove)
{
opal_list_t newnodes, exclude;
opal_list_item_t *item1, *item2, *next, *item3;
orte_node_t *node_from_list, *node_from_file, *node_from_pool, *node3;
int rc = ORTE_SUCCESS;
char *cptr;
int num_empty, nodeidx;
bool want_all_empty = false;
opal_list_t keep;
bool found;
OPAL_OUTPUT_VERBOSE((1, orte_ras_base_framework.framework_output,
"%s hostfile: filtering nodes through hostfile %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
/* parse the hostfile and create local list of findings */
OBJ_CONSTRUCT(&newnodes, opal_list_t);
OBJ_CONSTRUCT(&exclude, opal_list_t);
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, &newnodes, &exclude, false))) {
OBJ_DESTRUCT(&newnodes);
OBJ_DESTRUCT(&exclude);
return rc;
}
/* if the hostfile was empty, then treat it as a no-op filter */
if (0 == opal_list_get_size(&newnodes)) {
OBJ_DESTRUCT(&newnodes);
OBJ_DESTRUCT(&exclude);
/* indicate that the hostfile was empty */
return ORTE_ERR_TAKE_NEXT_OPTION;
}
/* remove from the list of newnodes those that are in the exclude list
* since we could have added duplicate names above due to the */
while (NULL != (item1 = opal_list_remove_first(&exclude))) {
node_from_file = (orte_node_t*)item1;
/* check for matches on nodes */
for (item2 = opal_list_get_first(&newnodes);
item2 != opal_list_get_end(&newnodes);
item2 = opal_list_get_next(item2)) {
orte_node_t *node = (orte_node_t*)item2;
if (0 == strcmp(node_from_file->name, node->name)) {
/* match - remove it */
opal_output(0, "HOST %s ON EXCLUDE LIST - REMOVING", node->name);
opal_list_remove_item(&newnodes, item2);
OBJ_RELEASE(item2);
break;
}
}
OBJ_RELEASE(item1);
}
/* now check our nodes and keep or mark those that match. We can
* destruct our hostfile list as we go since this won't be needed
*/
OBJ_CONSTRUCT(&keep, opal_list_t);
while (NULL != (item2 = opal_list_remove_first(&newnodes))) {
node_from_file = (orte_node_t*)item2;
next = opal_list_get_next(item2);
/* see if this is a relative node syntax */
if ('+' == node_from_file->name[0]) {
/* see if we specified empty nodes */
if ('e' == node_from_file->name[1] ||
'E' == node_from_file->name[1]) {
/* request for empty nodes - do they want
* all of them?
*/
if (NULL != (cptr = strchr(node_from_file->name, ':'))) {
/* the colon indicates a specific # are requested */
cptr++; /* step past : */
num_empty = strtol(cptr, NULL, 10);
} else {
/* want them all - set num_empty to max */
num_empty = INT_MAX;
want_all_empty = true;
}
/* search the list of nodes provided to us and find those
* that are empty
*/
item1 = opal_list_get_first(nodes);
while (0 < num_empty && item1 != opal_list_get_end(nodes)) {
node_from_list = (orte_node_t*)item1;
next = opal_list_get_next(item1); /* keep our place */
if (0 == node_from_list->slots_inuse) {
/* check to see if this node is explicitly called
* out later - if so, don't use it here
*/
for (item3 = opal_list_get_first(&newnodes);
item3 != opal_list_get_end(&newnodes);
item3 = opal_list_get_next(item3)) {
node3 = (orte_node_t*)item3;
if (0 == strcmp(node3->name, node_from_list->name)) {
/* match - don't use it */
goto skipnode;
}
}
if (remove) {
/* remove item from list */
opal_list_remove_item(nodes, item1);
/* xfer to keep list */
opal_list_append(&keep, item1);
} else {
/* mark as included */
node_from_list->mapped = true;
}
--num_empty;
}
skipnode:
item1 = next;
}
/* did they get everything they wanted? */
if (!want_all_empty && 0 < num_empty) {
orte_show_help("help-hostfile.txt", "hostfile:not-enough-empty",
true, num_empty);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
} else if ('n' == node_from_file->name[1] ||
'N' == node_from_file->name[1]) {
/* they want a specific relative node #, so
* look it up on global pool
*/
nodeidx = strtol(&node_from_file->name[2], NULL, 10);
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, nodeidx))) {
/* this is an error */
orte_show_help("help-hostfile.txt", "hostfile:relative-node-not-found",
true, nodeidx, node_from_file->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* search the list of nodes provided to us and find it */
for (item1 = opal_list_get_first(nodes);
item1 != opal_list_get_end(nodes);
item1 = opal_list_get_next(nodes)) {
node_from_list = (orte_node_t*)item1;
if (0 == strcmp(node_from_list->name, node_from_pool->name)) {
if (remove) {
/* match - remove item from list */
opal_list_remove_item(nodes, item1);
/* xfer to keep list */
opal_list_append(&keep, item1);
} else {
/* mark as included */
node_from_list->mapped = true;
}
break;
}
}
} else {
/* invalid relative node syntax */
orte_show_help("help-hostfile.txt", "hostfile:invalid-relative-node-syntax",
true, node_from_file->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
} else {
/* we are looking for a specific node on the list
* search the provided list of nodes to see if this
* one is found
*/
found = false;
for (item1 = opal_list_get_first(nodes);
item1 != opal_list_get_end(nodes);
item1 = opal_list_get_next(item1)) {
node_from_list = (orte_node_t*)item1;
/* since the name in the hostfile might not match
* our local name, and yet still be intended to match,
* we have to check for local interfaces
*/
if (0 == strcmp(node_from_file->name, node_from_list->name) ||
(0 == strcmp(node_from_file->name, "localhost") &&
0 == strcmp(node_from_list->name, orte_process_info.nodename)) ||
(opal_ifislocal(node_from_list->name) &&
opal_ifislocal(node_from_file->name))) {
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node_from_file->slots < node_from_list->slots) {
node_from_list->slots = node_from_file->slots;
}
if (remove) {
/* remove the node from the list */
opal_list_remove_item(nodes, item1);
/* xfer it to keep list */
opal_list_append(&keep, item1);
} else {
/* mark as included */
node_from_list->mapped = true;
}
found = true;
break;
}
}
/* if the host in the newnode list wasn't found,
* then that is an error we need to report to the
* user and abort
*/
if (!found) {
orte_show_help("help-hostfile.txt", "hostfile:extra-node-not-found",
true, hostfile, node_from_file->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
}
/* cleanup the newnode list */
OBJ_RELEASE(item2);
}
/* if we still have entries on our hostfile list, then
* there were requested hosts that were not in our allocation.
* This is an error - report it to the user and return an error
*/
if (0 != opal_list_get_size(&newnodes)) {
orte_show_help("help-hostfile.txt", "not-all-mapped-alloc",
true, hostfile);
while (NULL != (item1 = opal_list_remove_first(&newnodes))) {
OBJ_RELEASE(item1);
}
OBJ_DESTRUCT(&newnodes);
return ORTE_ERR_SILENT;
}
if (!remove) {
/* all done */
OBJ_DESTRUCT(&newnodes);
return ORTE_SUCCESS;
}
/* clear the rest of the nodes list */
while (NULL != (item1 = opal_list_remove_first(nodes))) {
OBJ_RELEASE(item1);
}
/* the nodes list has been cleared - rebuild it in order */
while (NULL != (item1 = opal_list_remove_first(&keep))) {
opal_list_append(nodes, item1);
}
cleanup:
OBJ_DESTRUCT(&newnodes);
return rc;
}
int orte_util_add_hostfile_nodes(opal_list_t *nodes,
char *hostfile)
{
opal_list_t exclude, adds;
opal_list_item_t *item, *itm;
int rc;
orte_node_t *nd, *node;
bool found;
OPAL_OUTPUT_VERBOSE((1, orte_ras_base_framework.framework_output,
"%s hostfile: checking hostfile %s for nodes",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
OBJ_CONSTRUCT(&adds, opal_list_t);
/* parse the hostfile and add any new contents to the list */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, &adds, &exclude, false))) {
goto cleanup;
}
/* check for any relative node directives */
for (item = opal_list_get_first(&adds);
item != opal_list_get_end(&adds);
item = opal_list_get_next(item)) {
node=(orte_node_t*)item;
if ('+' == node->name[0]) {
orte_show_help("help-hostfile.txt", "hostfile:relative-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
}
/* remove from the list of nodes those that are in the exclude list */
while (NULL != (item = opal_list_remove_first(&exclude))) {
nd = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(&adds);
itm != opal_list_get_end(&adds);
itm = opal_list_get_next(itm)) {
node = (orte_node_t*)itm;
if (0 == strcmp(nd->name, node->name)) {
/* match - remove it */
opal_list_remove_item(&adds, itm);
OBJ_RELEASE(itm);
break;
}
}
OBJ_RELEASE(item);
}
/* transfer across all unique nodes */
while (NULL != (item = opal_list_remove_first(&adds))) {
nd = (orte_node_t*)item;
found = false;
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
node = (orte_node_t*)itm;
if (0 == strcmp(nd->name, node->name)) {
found = true;
break;
}
}
if (!found) {
opal_list_append(nodes, &nd->super);
} else {
OBJ_RELEASE(item);
}
}
cleanup:
OPAL_LIST_DESTRUCT(&exclude);
OPAL_LIST_DESTRUCT(&adds);
return rc;
}
int orte_util_add_hostfile_nodes(opal_list_t *nodes,
bool *override_oversubscribed,
char *hostfile)
{
opal_list_t exclude;
opal_list_item_t *item, *itm;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_debug_output,
"%s hostfile: checking hostfile %s for nodes",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
/* parse the hostfile and add the contents to the list */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, nodes, &exclude, false))) {
goto cleanup;
}
/* parse the nodes to check for any relative node directives */
for (item = opal_list_get_first(nodes);
item != opal_list_get_end(nodes);
item = opal_list_get_next(item)) {
orte_node_t *node=(orte_node_t*)item;
if ('+' == node->name[0]) {
orte_show_help("help-hostfile.txt", "hostfile:relative-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
}
/* remove from the list of nodes those that are in the exclude list */
while(NULL != (item = opal_list_remove_first(&exclude))) {
orte_node_t *exnode = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
orte_node_t *node=(orte_node_t*)itm;
if (0 == strcmp(exnode->name, node->name)) {
/* match - remove it */
opal_list_remove_item(nodes, itm);
OBJ_RELEASE(itm);
break;
}
}
OBJ_RELEASE(item);
}
/* indicate that ORTE should override any oversubscribed conditions
* based on local hardware limits since the user (a) might not have
* provided us any info on the #slots for a node, and (b) the user
* might have been wrong! If we don't check the number of local physical
* processors, then we could be too aggressive on our sched_yield setting
* and cause performance problems.
*/
*override_oversubscribed = true;
cleanup:
OBJ_DESTRUCT(&exclude);
return rc;
}
