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

Next step in reducing launch time: begin reducing the size of the launch message itself. Start by expressing the daemon map as a set of three regular expression strings. On an 8k cluster, this reduces the nidmap contribution from over 200kBytes to 21 bytes in size.

Просмотр исходного кода 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
Этот коммит содержится в:
Ralph Castain 2017-01-23 09:14:20 -08:00
родитель f3920828ed
Коммит 86ab751c5e
9 изменённых файлов: 498 добавлений и 210 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

33
orte/mca/grpcomm/direct/grpcomm_direct.c
Просмотреть файл

@ -374,35 +374,18 @@ static void xcast_recv(int status, orte_process_name_t* sender,
orte_orteds_term_ordered = true;
} else if (ORTE_DAEMON_ADD_LOCAL_PROCS == command ||
ORTE_DAEMON_DVM_NIDMAP_CMD == command) {
/* extract the byte object holding the daemonmap */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(data, &bo, &cnt, OPAL_BYTE_OBJECT))) {
/* update our local nidmap, if required - the decode function
* knows what to do
*/
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base_framework.framework_output,
"%s grpcomm:direct:xcast updating daemon nidmap",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if (ORTE_SUCCESS != (ret = orte_util_decode_daemon_nodemap(data))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
/* update our local nidmap, if required - the decode function
* knows what to do - it will also free the bytes in the byte object
*/
if (ORTE_PROC_IS_HNP) {
/* no need - already have the info */
if (NULL != bo) {
if (NULL != bo->bytes) {
free(bo->bytes);
}
free(bo);
}
} else {
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base_framework.framework_output,
"%s grpcomm:direct:xcast updating daemon nidmap",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if (ORTE_SUCCESS != (ret = orte_util_decode_daemon_nodemap(bo))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
}
/* update the routing plan */
orte_routed.update_routing_plan(rtmod);

14
orte/mca/odls/base/odls_base_default_fns.c
Просмотреть файл

@ -77,7 +77,6 @@
#include "orte/util/session_dir.h"
#include "orte/util/proc_info.h"
#include "orte/util/nidmap.h"
#include "orte/util/regex.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "orte/runtime/orte_wait.h"
@ -138,21 +137,12 @@ int orte_odls_base_default_get_add_procs_data(opal_buffer_t *buffer,
return rc;
}
/* construct a nodemap - only want updated items */
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(&bo, true))) {
/* construct a nodemap of the daemons */
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(buffer))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* store it */
boptr = &bo;
if (ORTE_SUCCESS != (rc = opal_dss.pack(buffer, &boptr, 1, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* release the data since it has now been copied into our buffer */
free(bo.bytes);
/* if we are not using static ports, we need to send the wireup info */
if (!orte_static_ports) {
/* pack a flag indicating wiring info is provided */

41
orte/mca/plm/base/plm_base_launch_support.c
Просмотреть файл

@ -1059,41 +1059,12 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
"%s plm:base:orted_report_launch attempting to assign daemon %s to node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&dname), nodename));
for (idx=0; idx < orte_node_pool->size; idx++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, idx))) {
continue;
}
if (ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_LOC_VERIFIED)) {
/* already assigned */
continue;
}
if (0 == strcmp(nodename, node->name)) {
/* flag that we verified the location */
ORTE_FLAG_SET(node, ORTE_NODE_FLAG_LOC_VERIFIED);
if (node == daemon->node) {
/* it wound up right where it should */
break;
}
/* remove the prior association */
if (NULL != daemon->node) {
OBJ_RELEASE(daemon->node);
}
if (NULL != node->daemon) {
OBJ_RELEASE(node->daemon);
}
/* associate this daemon with the node */
node->daemon = daemon;
OBJ_RETAIN(daemon);
/* associate this node with the daemon */
daemon->node = node;
OBJ_RETAIN(node);
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:orted_report_launch assigning daemon %s to node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&daemon->name), node->name));
break;
}
}
/* to "relocate" the daemon, we just update the name of
* the node object pointed to by this daemon */
free(daemon->node->name);
daemon->node->name = strdup(nodename);
/* mark that it was verified */
ORTE_FLAG_SET(node, ORTE_NODE_FLAG_LOC_VERIFIED);
}
node = daemon->node;

34
orte/mca/plm/rsh/plm_rsh_module.c
Просмотреть файл

@ -14,7 +14,7 @@
* reserved.
* Copyright (c) 2008-2009 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2011 IBM Corporation. All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015-2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -780,7 +780,6 @@ static int remote_spawn(opal_buffer_t *launch)
int rc=ORTE_SUCCESS;
bool failed_launch = true;
orte_std_cntr_t n;
opal_byte_object_t *bo;
orte_process_name_t target;
orte_plm_rsh_caddy_t *caddy;
orte_job_t *daemons;
@ -802,23 +801,8 @@ static int remote_spawn(opal_buffer_t *launch)
goto cleanup;
}
/* extract the byte object holding the nidmap */
n=1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(launch, &bo, &n, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* update our nidmap - this will free data in the byte object */
if (ORTE_SUCCESS != (rc = orte_util_decode_daemon_nodemap(bo))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* ensure the routing plan is updated */
rtmod = orte_rml.get_routed(orte_coll_conduit);
orte_routed.update_routing_plan(rtmod);
/* get the updated routing list */
rtmod = orte_rml.get_routed(orte_coll_conduit);
OBJ_CONSTRUCT(&coll, opal_list_t);
orte_routed.get_routing_list(rtmod, &coll);
@ -1124,7 +1108,6 @@ static void launch_daemons(int fd, short args, void *cbdata)
/* if we are tree launching, find our children and create the launch cmd */
if (!mca_plm_rsh_component.no_tree_spawn) {
orte_daemon_cmd_flag_t command = ORTE_DAEMON_TREE_SPAWN;
opal_byte_object_t bo, *boptr;
orte_job_t *jdatorted;
/* get the tree spawn buffer */
@ -1142,21 +1125,12 @@ static void launch_daemons(int fd, short args, void *cbdata)
goto cleanup;
}
/* construct a nodemap of all daemons we know about */
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(&bo, false))) {
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(orte_tree_launch_cmd))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
goto cleanup;
}
/* store it */
boptr = &bo;
if (ORTE_SUCCESS != (rc = opal_dss.pack(orte_tree_launch_cmd, &boptr, 1, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
free(bo.bytes);
goto cleanup;
}
/* release the data since it has now been copied into our buffer */
free(bo.bytes);
/* get the orted job data object */
if (NULL == (jdatorted = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);

14
orte/mca/state/dvm/state_dvm.c
Просмотреть файл

@ -1,5 +1,5 @@
/*
* Copyright (c) 2015-2016 Intel, Inc. All rights reserved
* Copyright (c) 2015-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -260,22 +260,12 @@ static void vm_ready(int fd, short args, void *cbdata)
return;
}
/* construct a nodemap with everything in it */
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(&bo, false))) {
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(buf))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return;
}
/* store it */
boptr = &bo;
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &boptr, 1, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return;
}
/* release the data since it has now been copied into our buffer */
free(bo.bytes);
/* pack a flag indicating wiring info is provided */
flag = 1;
opal_dss.pack(buf, &flag, 1, OPAL_INT8);

559
orte/util/nidmap.c
Просмотреть файл

@ -12,7 +12,7 @@
* All rights reserved.
* Copyright (c) 2012-2014 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -184,134 +184,508 @@ int orte_util_build_daemon_nidmap(char **nodes)
return rc;
}
int orte_util_encode_nodemap(opal_byte_object_t *boptr, bool update)
int orte_util_encode_nodemap(opal_buffer_t *buffer)
{
orte_node_t *node;
int32_t i;
char *node;
char prefix[ORTE_MAX_NODE_PREFIX];
int i, j, n, len, startnum, nodenum, numdigits;
bool found, fullname;
char *suffix, *sfx;
orte_regex_node_t *ndreg;
orte_regex_range_t *range, *rng, *idrng;
opal_list_t nodeids, nodenms, dvpids;
opal_list_item_t *item, *itm2;
char **regexargs = NULL, *tmp, *tmp2;
orte_node_t *nptr;
int rc;
opal_buffer_t buf;
orte_job_t *daemons;
orte_proc_t *dmn;
/* if the daemon job has not been updated, then there is
* nothing to send
*/
daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
if (update && !ORTE_FLAG_TEST(daemons, ORTE_JOB_FLAG_UPDATED)) {
boptr->bytes = NULL;
boptr->size = 0;
return ORTE_SUCCESS;
}
/* setup the list of results */
OBJ_CONSTRUCT(&nodeids, opal_list_t);
OBJ_CONSTRUCT(&nodenms, opal_list_t);
OBJ_CONSTRUCT(&dvpids, opal_list_t);
/* setup a buffer for tmp use */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
/* send the number of nodes */
if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &daemons->num_procs, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
return rc;
}
for (i=0; i < daemons->procs->size; i++) {
if (NULL == (dmn = (orte_proc_t*)opal_pointer_array_get_item(daemons->procs, i))) {
rng = NULL;
idrng = NULL;
for (n=0; n < orte_node_pool->size; n++) {
if (NULL == (nptr = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, n))) {
continue;
}
/* if the daemon doesn't have a node, that's an error */
if (NULL == (node = dmn->node)) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
/* if no daemon has been assigned, then this node is not being used */
if (NULL == nptr->daemon) {
continue;
}
if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &dmn->name.vpid, 1, ORTE_VPID))) {
ORTE_ERROR_LOG(rc);
return rc;
/* deal with the nodeids - these are just the index into
* the node array. We pass these to ensure coherence across
* the virtual machine */
if (NULL == idrng) {
/* just starting */
idrng = OBJ_NEW(orte_regex_range_t);
idrng->start = n;
idrng->cnt = 1;
opal_list_append(&nodeids, &idrng->super);
} else {
/* is this the next in line */
if (n == (idrng->start + idrng->cnt)) {
idrng->cnt++;
} else {
/* need to start another range */
idrng = OBJ_NEW(orte_regex_range_t);
idrng->start = n;
opal_list_append(&nodeids, &idrng->super);
}
}
/* pack the node */
if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &node, 1, ORTE_NODE))) {
ORTE_ERROR_LOG(rc);
return rc;
/* deal with the daemon vpid - see if it is next in the
* current range */
if (NULL == rng) {
/* just starting */
rng = OBJ_NEW(orte_regex_range_t);
rng->start = nptr->daemon->name.vpid;
rng->cnt = 1;
opal_list_append(&dvpids, &rng->super);
} else {
/* is this the next in line */
if (nptr->daemon->name.vpid == (orte_vpid_t)(rng->start + rng->cnt)) {
rng->cnt++;
} else {
/* need to start another range */
rng = OBJ_NEW(orte_regex_range_t);
rng->start = nptr->daemon->name.vpid;
opal_list_append(&dvpids, &rng->super);
}
}
node = nptr->name;
/* determine this node's prefix by looking for first non-alpha char */
fullname = false;
len = strlen(node);
startnum = -1;
memset(prefix, 0, ORTE_MAX_NODE_PREFIX);
numdigits = 0;
for (i=0, j=0; i < len; i++) {
if (!isalpha(node[i])) {
/* found a non-alpha char */
if (!isdigit(node[i])) {
/* if it is anything but a digit, we just use
* the entire name
*/
fullname = true;
break;
}
/* count the size of the numeric field - but don't
* add the digits to the prefix
*/
numdigits++;
if (startnum < 0) {
/* okay, this defines end of the prefix */
startnum = i;
}
continue;
}
if (startnum < 0) {
prefix[j++] = node[i];
}
}
if (fullname || startnum < 0) {
/* can't compress this name - just add it to the list */
ndreg = OBJ_NEW(orte_regex_node_t);
ndreg->prefix = strdup(node);
opal_list_append(&nodenms, &ndreg->super);
continue;
}
/* convert the digits and get any suffix */
nodenum = strtol(&node[startnum], &sfx, 10);
if (NULL != sfx) {
suffix = strdup(sfx);
} else {
suffix = NULL;
}
/* is this node name already on our list? */
found = false;
for (item = opal_list_get_first(&nodenms);
!found && item != opal_list_get_end(&nodenms);
item = opal_list_get_next(item)) {
ndreg = (orte_regex_node_t*)item;
if (0 < strlen(prefix) && NULL == ndreg->prefix) {
continue;
}
if (0 == strlen(prefix) && NULL != ndreg->prefix) {
continue;
}
if (0 < strlen(prefix) && NULL != ndreg->prefix
&& 0 != strcmp(prefix, ndreg->prefix)) {
continue;
}
if (NULL == suffix && NULL != ndreg->suffix) {
continue;
}
if (NULL != suffix && NULL == ndreg->suffix) {
continue;
}
if (NULL != suffix && NULL != ndreg->suffix &&
0 != strcmp(suffix, ndreg->suffix)) {
continue;
}
if (numdigits != ndreg->num_digits) {
continue;
}
/* found a match - flag it */
found = true;
/* get the last range on this nodeid - we do this
* to preserve order
*/
range = (orte_regex_range_t*)opal_list_get_last(&ndreg->ranges);
if (NULL == range) {
/* first range for this nodeid */
range = OBJ_NEW(orte_regex_range_t);
range->start = nodenum;
range->cnt = 1;
opal_list_append(&ndreg->ranges, &range->super);
break;
}
/* see if the node number is out of sequence */
if (nodenum != (range->start + range->cnt)) {
/* start a new range */
range = OBJ_NEW(orte_regex_range_t);
range->start = nodenum;
range->cnt = 1;
opal_list_append(&ndreg->ranges, &range->super);
break;
}
/* everything matches - just increment the cnt */
range->cnt++;
break;
}
if (!found) {
/* need to add it */
ndreg = OBJ_NEW(orte_regex_node_t);
if (0 < strlen(prefix)) {
ndreg->prefix = strdup(prefix);
}
if (NULL != suffix) {
ndreg->suffix = strdup(suffix);
}
ndreg->num_digits = numdigits;
opal_list_append(&nodenms, &ndreg->super);
/* record the first range for this nodeid - we took
* care of names we can't compress above
*/
range = OBJ_NEW(orte_regex_range_t);
range->start = nodenum;
range->cnt = 1;
opal_list_append(&ndreg->ranges, &range->super);
}
if (NULL != suffix) {
free(suffix);
}
}
/* transfer the payload to the byte object */
opal_dss.unload(&buf, (void**)&boptr->bytes, &boptr->size);
OBJ_DESTRUCT(&buf);
/* begin constructing the regular expression */
while (NULL != (item = opal_list_remove_first(&nodenms))) {
ndreg = (orte_regex_node_t*)item;
/* if no ranges, then just add the name */
if (0 == opal_list_get_size(&ndreg->ranges)) {
if (NULL != ndreg->prefix) {
/* solitary node */
asprintf(&tmp, "%s", ndreg->prefix);
opal_argv_append_nosize(&regexargs, tmp);
free(tmp);
}
OBJ_RELEASE(ndreg);
continue;
}
/* start the regex for this nodeid with the prefix */
if (NULL != ndreg->prefix) {
asprintf(&tmp, "%s[%d:", ndreg->prefix, ndreg->num_digits);
} else {
asprintf(&tmp, "[%d:", ndreg->num_digits);
}
/* add the ranges */
while (NULL != (itm2 = opal_list_remove_first(&ndreg->ranges))) {
range = (orte_regex_range_t*)itm2;
if (1 == range->cnt) {
asprintf(&tmp2, "%s%d,", tmp, range->start);
} else {
asprintf(&tmp2, "%s%d-%d,", tmp, range->start, range->start + range->cnt - 1);
}
free(tmp);
tmp = tmp2;
OBJ_RELEASE(range);
}
/* replace the final comma */
tmp[strlen(tmp)-1] = ']';
if (NULL != ndreg->suffix) {
/* add in the suffix, if provided */
asprintf(&tmp2, "%s%s", tmp, ndreg->suffix);
free(tmp);
tmp = tmp2;
}
opal_argv_append_nosize(&regexargs, tmp);
free(tmp);
OBJ_RELEASE(ndreg);
}
/* assemble final result */
tmp = opal_argv_join(regexargs, ',');
/* cleanup */
opal_argv_free(regexargs);
OBJ_DESTRUCT(&nodenms);
/* pack the string */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buffer, &tmp, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&dvpids);
return rc;
}
if (NULL != tmp) {
free(tmp);
}
/* do the same for the indices */
tmp = NULL;
while (NULL != (item = opal_list_remove_first(&nodeids))) {
rng = (orte_regex_range_t*)item;
if (1 < rng->cnt) {
if (NULL == tmp) {
asprintf(&tmp, "%d-%d", rng->start, rng->start + rng->cnt - 1);
} else {
asprintf(&tmp2, "%s,%d-%d", tmp, rng->start, rng->start + rng->cnt - 1);
free(tmp);
tmp = tmp2;
}
} else {
if (NULL == tmp) {
asprintf(&tmp, "%d", rng->start);
} else {
asprintf(&tmp2, "%s,%d", tmp, rng->start);
free(tmp);
tmp = tmp2;
}
}
OBJ_RELEASE(rng);
}
OPAL_LIST_DESTRUCT(&nodeids);
/* pack the string */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buffer, &tmp, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&dvpids);
return rc;
}
if (NULL != tmp) {
free(tmp);
}
/* do the same for the vpids */
tmp = NULL;
while (NULL != (item = opal_list_remove_first(&dvpids))) {
rng = (orte_regex_range_t*)item;
if (1 < rng->cnt) {
if (NULL == tmp) {
asprintf(&tmp, "%d-%d", rng->start, rng->start + rng->cnt - 1);
} else {
asprintf(&tmp2, "%s,%d-%d", tmp, rng->start, rng->start + rng->cnt - 1);
free(tmp);
tmp = tmp2;
}
} else {
if (NULL == tmp) {
asprintf(&tmp, "%d", rng->start);
} else {
asprintf(&tmp2, "%s,%d", tmp, rng->start);
free(tmp);
tmp = tmp2;
}
}
OBJ_RELEASE(rng);
}
OPAL_LIST_DESTRUCT(&dvpids);
/* pack the string */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buffer, &tmp, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&dvpids);
return rc;
}
if (NULL != tmp) {
free(tmp);
}
return ORTE_SUCCESS;
}
/* decode a nodemap for a daemon */
int orte_util_decode_daemon_nodemap(opal_byte_object_t *bo)
int orte_util_decode_daemon_nodemap(opal_buffer_t *buffer)
{
int n;
orte_vpid_t vpid;
orte_node_t *node, *nptr;
opal_buffer_t buf;
int rc=ORTE_SUCCESS;
int k, m, n, rc, start, endpt;
orte_node_t *node;
size_t num_nodes;
orte_job_t *daemons;
orte_proc_t *dptr;
orte_vpid_t num_daemons;
char **nodes, *indices, *dvpids;
char *ndnames, *rmndr, **tmp;
int *nodeids, *dids;
if (NULL == bo->bytes || 0 == bo->size) {
/* nothing to unpack */
/* unpack the node regex */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &ndnames, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* it is okay for this to be NULL */
if (NULL == ndnames) {
return ORTE_SUCCESS;
}
/* xfer the byte object to a buffer for unpacking */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
opal_dss.load(&buf, bo->bytes, bo->size);
/* unpack the number of procs */
n=1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &num_daemons, &n, ORTE_VPID))) {
/* unpack the index regex */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &indices, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
free(ndnames);
return rc;
}
/* this is not allowed to be NULL */
if (NULL == indices) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
free(ndnames);
return ORTE_ERR_BAD_PARAM;
}
/* transfer the data to the nodes */
daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
daemons->num_procs = num_daemons;
n=1;
while (OPAL_SUCCESS == (rc = opal_dss.unpack(&buf, &vpid, &n, ORTE_VPID))) {
/* unpack and store the node */
n=1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &node, &n, ORTE_NODE))) {
ORTE_ERROR_LOG(rc);
return rc;
/* unpack the daemon vpid regex */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &dvpids, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
free(ndnames);
free(indices);
return rc;
}
/* this is not allowed to be NULL */
if (NULL == dvpids) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
free(ndnames);
free(indices);
return ORTE_ERR_BAD_PARAM;
}
/* if we are the HNP, then we just discard these strings as we already
* have a complete picture - but we needed to unpack them in order to
* maintain sync in the unpacking order */
if (ORTE_PROC_IS_HNP) {
free(ndnames);
free(indices);
free(dvpids);
return ORTE_SUCCESS;
}
/* decompress the regex */
nodes = NULL;
if (ORTE_SUCCESS != (rc = orte_regex_extract_node_names(ndnames, &nodes))) {
free(ndnames);
free(indices);
free(dvpids);
return rc;
}
free(ndnames);
if (NULL == nodes) {
/* should not happen */
free(indices);
free(dvpids);
return ORTE_ERR_NOT_FOUND;
}
/* decompress the index ranges */
num_nodes = opal_argv_count(nodes);
nodeids = (int*)malloc(num_nodes * sizeof(int));
tmp = opal_argv_split(indices, ',');
k = 0;
for (n=0; NULL != tmp[n]; n++) {
/* convert the number - since it might be a range,
* save the remainder pointer */
nodeids[k] = strtoul(tmp[n], &rmndr, 10);
if (NULL != rmndr) {
/* it must be a range - find the endpoint */
++rmndr;
endpt = strtoul(rmndr, NULL, 10);
start = nodeids[k] + 1;
for (m=0; m < endpt; m++) {
++k;
nodeids[k] = start + m;
}
}
/* set the nodeid */
node->index = vpid;
/* do we already have this node? */
nptr = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, vpid);
/* set the new node object into the array */
opal_pointer_array_set_item(orte_node_pool, vpid, node);
if (NULL == (dptr = (orte_proc_t*)opal_pointer_array_get_item(daemons->procs, vpid))) {
++k;
}
opal_argv_free(tmp);
free(indices);
/* decompress the vpids */
dids = (int*)malloc(num_nodes * sizeof(int));
tmp = opal_argv_split(dvpids, ',');
k = 0;
for (n=0; NULL != tmp[n]; n++) {
/* convert the number - since it might be a range,
* save the remainder pointer */
dids[k] = strtoul(tmp[n], &rmndr, 10);
if (NULL != rmndr) {
/* it must be a range - find the endpoint */
++rmndr;
endpt = strtoul(rmndr, NULL, 10);
start = dids[k] + 1;
for (m=0; m < endpt; m++) {
++k;
dids[k] = start + m;
}
}
++k;
}
opal_argv_free(tmp);
free(dvpids);
daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
/* since this is a complete picture of all the nodes, reset the
* counters and regenerate them */
orte_process_info.num_daemons = 0;
orte_process_info.num_nodes = 0;
daemons->num_procs = 0;
/* update the node array */
for (n=0; NULL != nodes[n]; n++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, nodeids[n]))) {
node = OBJ_NEW(orte_node_t);
node->name = nodes[n];
node->index = nodeids[n];
opal_pointer_array_set_item(orte_node_pool, node->index, node);
} else if (NULL != node->daemon) {
OBJ_RELEASE(node->daemon);
node->daemon = NULL;
}
++orte_process_info.num_nodes;
if (NULL == (dptr = (orte_proc_t*)opal_pointer_array_get_item(daemons->procs, dids[n]))) {
/* create a daemon object for this node */
dptr = OBJ_NEW(orte_proc_t);
dptr->name.jobid = ORTE_PROC_MY_NAME->jobid;
dptr->name.vpid = vpid;
dptr->name.vpid = dids[n];
ORTE_FLAG_SET(dptr, ORTE_PROC_FLAG_ALIVE); // assume the daemon is alive until discovered otherwise
opal_pointer_array_set_item(daemons->procs, vpid, dptr);
}
if (NULL != node->daemon) {
OBJ_RELEASE(node->daemon);
opal_pointer_array_set_item(daemons->procs, dids[n], dptr);
} else if (NULL != dptr->node) {
OBJ_RELEASE(dptr->node);
dptr->node = NULL;
}
++daemons->num_procs;
/* link the node to the daemon */
OBJ_RETAIN(dptr);
node->daemon = dptr;
if (NULL != dptr->node) {
OBJ_RELEASE(dptr->node);
}
/* link the node to the daemon */
OBJ_RETAIN(node);
if (NULL != nptr) {
OBJ_RELEASE(nptr);
}
dptr->node = node;
}
if (ORTE_ERR_UNPACK_READ_PAST_END_OF_BUFFER != rc) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&buf);
return rc;
}
rc = ORTE_SUCCESS;
/* we cannot use opal_argv_free here as this would release
* all the node names themselves. Instead, we just free the
* array of string pointers, leaving the strings alone */
free(nodes);
free(nodeids);
free(dids);
/* unpdate num procs */
orte_process_info.num_procs = daemons->num_procs;
if (orte_process_info.max_procs < orte_process_info.num_procs) {
@ -334,6 +708,5 @@ int orte_util_decode_daemon_nodemap(opal_byte_object_t *bo)
}
}
OBJ_DESTRUCT(&buf);
return rc;
return ORTE_SUCCESS;
}

11
orte/util/nidmap.h
Просмотреть файл

@ -9,7 +9,7 @@
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -41,8 +41,13 @@ BEGIN_C_DECLS
#define ORTE_CONTIG_NODE_CMD 0x01
#define ORTE_NON_CONTIG_NODE_CMD 0x02
ORTE_DECLSPEC int orte_util_encode_nodemap(opal_byte_object_t *boptr, bool update);
ORTE_DECLSPEC int orte_util_decode_daemon_nodemap(opal_byte_object_t *bo);
/* create a regular expression describing the nodes in the
* allocation */
ORTE_DECLSPEC int orte_util_encode_nodemap(opal_buffer_t *buffer);
/* decode a regular expression created by the encode function
* into the orte_node_pool array */
ORTE_DECLSPEC int orte_util_decode_daemon_nodemap(opal_buffer_t *buffer);
ORTE_DECLSPEC int orte_util_build_daemon_nidmap(char **nodes);

1
orte/util/regex.c
Просмотреть файл

@ -10,6 +10,7 @@
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2013 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow

1
orte/util/regex.h
Просмотреть файл

@ -9,6 +9,7 @@
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow