1
1

Merge pull request #871 from rhc54/topic/dpm

More dynamic op cleanups
Этот коммит содержится в:
rhc54 2015-09-07 11:42:40 -07:00
родитель c404e98dce e6add86e4f
Коммит 3fa6c635a3
9 изменённых файлов: 309 добавлений и 178 удалений

Просмотреть файл

@ -112,8 +112,8 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
const char *port_string, bool send_first,
ompi_communicator_t **newcomm)
{
int k, size, rsize, rank, rc;
char **members = NULL, *nstring;
int k, size, rsize, rank, rc, rportlen=0;
char **members = NULL, *nstring, *rport=NULL, **pkeys=NULL;
bool dense, isnew;
opal_process_name_t pname;
opal_list_t ilist, mlist, rlist;
@ -169,6 +169,7 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
if (NULL == (proc_list[i] = ompi_group_peer_lookup(group,i))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
free(proc_list);
goto exit;
}
}
@ -195,13 +196,6 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
if (rank == root) {
/* the root for each side publishes their list of participants */
OBJ_CONSTRUCT(&ilist, opal_list_t);
/* publish a key-val containing the port string itself
* so other participants from our job can connect */
info = OBJ_NEW(opal_value_t);
info->key = opal_argv_join(members, ':');
info->type = OPAL_STRING;
info->data.string = strdup(port_string);
opal_list_append(&ilist, &info->super);
/* put my name at the front of the list of members - my
* name will therefore be on the list twice, but the
* other side's root needs to know the root from this side */
@ -220,141 +214,111 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
info->type = OPAL_STRING;
info->data.string = opal_argv_join(members, ':');
opal_list_append(&ilist, &info->super);
/* also save the key for later */
opal_argv_append_nosize(&pkeys, info->key);
/* publish them with "session" scope */
rc = opal_pmix.publish(&ilist);
OPAL_LIST_DESTRUCT(&ilist);
if (OPAL_SUCCESS != rc) {
opal_argv_free(members);
opal_argv_free(pkeys);
return OMPI_ERROR;
}
} else {
/* check to see if we have to get the port string for
* this connect - if we were started via a call to spawn,
* then the port string was given to us. Otherwise, the
* string will be NULL */
if (0 == strlen(port_string)) {
/* our root should have published it */
OBJ_CONSTRUCT(&ilist, opal_list_t);
pdat = OBJ_NEW(opal_pmix_pdata_t);
pdat->value.key = opal_argv_join(members, ':');
opal_list_append(&ilist, &pdat->super);
OBJ_CONSTRUCT(&mlist, opal_list_t);
/* if a non-blocking version of lookup isn't
* available, then use the blocking version */
if (NULL == opal_pmix.lookup_nb) {
rc = opal_pmix.lookup(&ilist, &mlist);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != rc) {
OMPI_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
return OMPI_ERROR;
}
} else {
char **keys = NULL;
struct lookup_caddy_t caddy;
opal_argv_append_nosize(&keys, pdat->value.key);
caddy.active = true;
caddy.pdat = pdat;
/* tell it to wait for the data to arrive */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_WAIT);
info->type = OPAL_BOOL;
info->data.flag = true;
opal_list_append(&mlist, &info->super);
/* give it a decent timeout as we don't know when
* the other side may call connect - it doesn't
* have to be simultaneous */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_TIMEOUT);
info->type = OPAL_INT;
info->data.integer = (size < 60) ? size : 60;
opal_list_append(&mlist, &info->super);
rc = opal_pmix.lookup_nb(keys, &mlist, lookup_cbfunc, &caddy);
if (OPAL_SUCCESS != rc) {
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&mlist);
opal_argv_free(keys);
opal_argv_free(members);
return OMPI_ERROR;
}
OMPI_WAIT_FOR_COMPLETION(caddy.active);
opal_argv_free(keys);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != caddy.status) {
OMPI_ERROR_LOG(caddy.status);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
return OMPI_ERROR;
}
}
port_string = strdup(pdat->value.data.string);
OPAL_LIST_DESTRUCT(&ilist);
/* lookup the other side's info - if a non-blocking form
* of lookup isn't available, then we use the blocking
* form and trust that the underlying system will WAIT
* until the other side publishes its data */
OBJ_CONSTRUCT(&ilist, opal_list_t);
pdat = OBJ_NEW(opal_pmix_pdata_t);
if (send_first) {
(void)asprintf(&pdat->value.key, "%s:accept", port_string);
} else {
(void)asprintf(&pdat->value.key, "%s:connect", port_string);
}
opal_list_append(&ilist, &pdat->super);
OBJ_CONSTRUCT(&mlist, opal_list_t);
/* if a non-blocking version of lookup isn't
* available, then use the blocking version */
if (NULL == opal_pmix.lookup_nb) {
rc = opal_pmix.lookup(&ilist, &mlist);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != rc) {
OMPI_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
goto exit;
}
} else {
char **keys = NULL;
struct lookup_caddy_t caddy;
opal_argv_append_nosize(&keys, pdat->value.key);
caddy.active = true;
caddy.pdat = pdat;
/* tell it to wait for the data to arrive */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_WAIT);
info->type = OPAL_BOOL;
info->data.flag = true;
opal_list_append(&mlist, &info->super);
/* give it a decent timeout as we don't know when
* the other side may call connect - it doesn't
* have to be simultaneous */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_TIMEOUT);
info->type = OPAL_INT;
info->data.integer = 60;
opal_list_append(&mlist, &info->super);
rc = opal_pmix.lookup_nb(keys, &mlist, lookup_cbfunc, &caddy);
if (OPAL_SUCCESS != rc) {
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&mlist);
opal_argv_free(keys);
opal_argv_free(members);
goto exit;
}
OMPI_WAIT_FOR_COMPLETION(caddy.active);
opal_argv_free(keys);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != caddy.status) {
OMPI_ERROR_LOG(caddy.status);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
goto exit;
}
}
/* save the result */
rport = strdup(pdat->value.data.string); // need this later
rportlen = strlen(rport) + 1; // retain the NULL terminator
OPAL_LIST_DESTRUCT(&ilist);
}
/* lookup the other side's info - if a non-blocking form
* of lookup isn't available, then we use the blocking
* form and trust that the underlying system will WAIT
* until the other side publishes its data */
OBJ_CONSTRUCT(&ilist, opal_list_t);
pdat = OBJ_NEW(opal_pmix_pdata_t);
if (send_first) {
(void)asprintf(&pdat->value.key, "%s:accept", port_string);
} else {
(void)asprintf(&pdat->value.key, "%s:connect", port_string);
/* if we aren't in a comm_spawn, the non-root members won't have
* a port_string - so let's make sure everyone knows the other
* side's participants */
/* bcast the list-length to all processes in the local comm */
rc = comm->c_coll.coll_bcast(&rportlen, 1, MPI_INT, root, comm,
comm->c_coll.coll_bcast_module);
if (OMPI_SUCCESS != rc) {
free(rport);
goto exit;
}
opal_list_append(&ilist, &pdat->super);
OBJ_CONSTRUCT(&mlist, opal_list_t);
/* if a non-blocking version of lookup isn't
* available, then use the blocking version */
if (NULL == opal_pmix.lookup_nb) {
rc = opal_pmix.lookup(&ilist, &mlist);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != rc) {
OMPI_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
return OMPI_ERROR;
}
} else {
char **keys = NULL;
struct lookup_caddy_t caddy;
opal_argv_append_nosize(&keys, pdat->value.key);
caddy.active = true;
caddy.pdat = pdat;
/* tell it to wait for the data to arrive */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_WAIT);
info->type = OPAL_BOOL;
info->data.flag = true;
opal_list_append(&mlist, &info->super);
/* give it a decent timeout as we don't know when
* the other side may call connect - it doesn't
* have to be simultaneous */
info = OBJ_NEW(opal_value_t);
info->key = strdup(OPAL_PMIX_TIMEOUT);
info->type = OPAL_INT;
info->data.integer = (size < 60) ? size : 60;
opal_list_append(&mlist, &info->super);
rc = opal_pmix.lookup_nb(keys, &mlist, lookup_cbfunc, &caddy);
if (OPAL_SUCCESS != rc) {
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&mlist);
opal_argv_free(keys);
opal_argv_free(members);
return OMPI_ERROR;
}
OMPI_WAIT_FOR_COMPLETION(caddy.active);
opal_argv_free(keys);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != caddy.status) {
OMPI_ERROR_LOG(caddy.status);
OPAL_LIST_DESTRUCT(&ilist);
opal_argv_free(members);
return OMPI_ERROR;
if (rank != root) {
/* non root processes need to allocate the buffer manually */
rport = (char*)malloc(rportlen);
if (NULL == rport) {
rc = OMPI_ERR_OUT_OF_RESOURCE;
goto exit;
}
}
/* now share the list of remote participants */
rc = comm->c_coll.coll_bcast(rport, rportlen, MPI_BYTE, root, comm,
comm->c_coll.coll_bcast_module);
if (OMPI_SUCCESS != rc) {
free(rport);
goto exit;
}
/* initiate a list of participants for the connect,
* starting with our own members, remembering to
@ -370,8 +334,9 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
if (OPAL_SUCCESS != (rc = opal_convert_string_to_process_name(&nm->name, members[i]))) {
OMPI_ERROR_LOG(rc);
opal_argv_free(members);
free(rport);
OPAL_LIST_DESTRUCT(&mlist);
return rc;
goto exit;
}
/* if the rank is wildcard, then we need to add all procs
* in that job to the list */
@ -390,7 +355,9 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
OMPI_ERROR_LOG(OMPI_ERR_BAD_PARAM);
OPAL_LIST_DESTRUCT(&mlist);
opal_argv_free(members);
return OMPI_ERR_BAD_PARAM;
free(rport);
rc = OMPI_ERR_BAD_PARAM;
goto exit;
}
++i;
} else {
@ -403,21 +370,22 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
/* the pdat object will contain a colon-delimited list
* of process names for the remote procs - convert it
* into an argv array */
members = opal_argv_split(pdat->value.data.string, ':');
OPAL_LIST_DESTRUCT(&ilist);
members = opal_argv_split(rport, ':');
free(rport);
/* the first entry is the root for the remote side */
if (OPAL_SUCCESS != (rc = opal_convert_string_to_process_name(&pname, members[0]))) {
OMPI_ERROR_LOG(rc);
opal_argv_free(members);
return rc;
goto exit;
}
/* check the name - it should never be a wildcard, so
* this is just checking for an error */
if (OPAL_VPID_WILDCARD == pname.vpid) {
OMPI_ERROR_LOG(OMPI_ERR_BAD_PARAM);
opal_argv_free(members);
return OMPI_ERR_BAD_PARAM;
rc = OMPI_ERR_BAD_PARAM;
goto exit;
}
/* add the list of remote procs to our list, and
@ -432,7 +400,7 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
opal_argv_free(members);
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&rlist);
return rc;
goto exit;
}
if (OPAL_VPID_WILDCARD == nm->name.vpid) {
jobid = nm->name.jobid;
@ -446,7 +414,8 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
opal_argv_free(members);
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&rlist);
return OMPI_ERR_BAD_PARAM;
rc = OMPI_ERR_BAD_PARAM;
goto exit;
}
rsize = strtoul(members[i+1], NULL, 10);
++i;
@ -485,10 +454,16 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
opal_argv_free(members);
/* tell the host RTE to connect us - this will download
* all known data for the nspace's of participating procs */
* all known data for the nspace's of participating procs
* so that add_procs will not result in a slew of lookups */
rc = opal_pmix.connect(&mlist);
OPAL_LIST_DESTRUCT(&mlist);
if (OPAL_SUCCESS != rc) {
OMPI_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&ilist);
OPAL_LIST_DESTRUCT(&rlist);
goto exit;
}
if (0 < opal_list_get_size(&ilist)) {
/* convert the list of new procs to a proc_t array */
new_proc_list = (ompi_proc_t**)calloc(opal_list_get_size(&ilist),
@ -598,6 +573,10 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
*/
exit:
if (NULL != pkeys) {
opal_pmix.unpublish(pkeys, NULL);
opal_argv_free(pkeys);
}
if (OMPI_SUCCESS != rc) {
if (MPI_COMM_NULL != newcomp && NULL != newcomp) {
OBJ_RETAIN(newcomp);
@ -695,13 +674,12 @@ int ompi_dpm_disconnect(ompi_communicator_t *comm)
return ret;
}
opal_pmix.fence(&coll, false);
/* ensure we tell the host RM to disconnect us */
opal_pmix.disconnect(&coll);
/* ensure we tell the host RM to disconnect us - this
* is a blocking operation that must include a fence */
ret = opal_pmix.disconnect(&coll);
OPAL_LIST_DESTRUCT(&coll);
return OMPI_SUCCESS;
return ret;
}
int ompi_dpm_spawn(int count, const char *array_of_commands[],
@ -1067,7 +1045,7 @@ int ompi_dpm_open_port(char *port_name)
r = opal_rand(&rnd);
opal_convert_process_name_to_string(&tmp, OMPI_PROC_MY_NAME);
snprintf(port_name, MPI_MAX_PORT_NAME, "%s:%u", tmp, r);
snprintf(port_name, MPI_MAX_PORT_NAME-1, "%s:%u", tmp, r);
free(tmp);
return OMPI_SUCCESS;
}

Просмотреть файл

@ -100,6 +100,9 @@ int MPI_Comm_join(int fd, MPI_Comm *intercomm)
send_first = true;
}
/* ensure the port name is NULL terminated */
memset(port_name, 0, MPI_MAX_PORT_NAME);
/* Assumption: socket_send should not block, even if the socket
is not configured to be non-blocking, because the message length are
so short. */

Просмотреть файл

@ -301,7 +301,7 @@ int pmix1_get(const opal_process_name_t *proc,
opal_output_verbose(1, opal_pmix_base_framework.framework_output,
"%s PMIx_client get on proc %s key %s",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(*proc), key);
(NULL == proc) ? "NULL" : OPAL_NAME_PRINT(*proc), key);
/* prep default response */
*val = NULL;
@ -371,7 +371,7 @@ int pmix1_getnb(const opal_process_name_t *proc, const char *key,
opal_output_verbose(1, opal_pmix_base_framework.framework_output,
"%s PMIx_client get_nb on proc %s key %s",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(*proc), key);
(NULL == proc) ? "NULL" : OPAL_NAME_PRINT(*proc), key);
/* create the caddy */
op = OBJ_NEW(pmix1_opcaddy_t);
@ -501,7 +501,7 @@ int pmix1_lookup(opal_list_t *data, opal_list_t *info)
++n;
}
} else {
pdata = NULL;
pinfo = NULL;
ninfo = 0;
}

Просмотреть файл

@ -569,6 +569,15 @@ static void pmix_server_dmdx_resp(int status, orte_process_name_t* sender,
}
}
static void opcon(orte_pmix_server_op_caddy_t *p)
{
p->procs = NULL;
p->info = NULL;
p->cbdata = NULL;
}
OBJ_CLASS_INSTANCE(orte_pmix_server_op_caddy_t,
opal_object_t,
opcon, NULL);
static void rqcon(pmix_server_req_t *p)
{

Просмотреть файл

@ -46,7 +46,8 @@
#include "orte/runtime/orte_globals.h"
#include "orte/mca/rml/rml.h"
#include "pmix_server_internal.h"
#include "orte/orted/pmix/pmix_server.h"
#include "orte/orted/pmix/pmix_server_internal.h"
void pmix_server_launch_resp(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
@ -327,6 +328,119 @@ int pmix_server_spawn_fn(opal_process_name_t *requestor,
return OPAL_SUCCESS;
}
static void _cnct(int sd, short args, void *cbdata);
static void _cnlk(int status, opal_list_t *data, void *cbdata)
{
orte_pmix_server_op_caddy_t *cd = (orte_pmix_server_op_caddy_t*)cbdata;
int rc, cnt;
opal_pmix_pdata_t *pdat;
orte_job_t *jdata;
opal_buffer_t buf;
/* if we failed to get the required data, then just inform
* the embedded server that the connect cannot succeed */
if (ORTE_SUCCESS != status || NULL == data) {
if (NULL != cd->cbfunc) {
rc = status;
goto release;
}
}
/* register the returned data with the embedded PMIx server */
pdat = (opal_pmix_pdata_t*)opal_list_get_first(data);
if (OPAL_BYTE_OBJECT != pdat->value.type) {
rc = ORTE_ERR_BAD_PARAM;
goto release;
}
/* the data will consist of a packed buffer with the job data in it */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
opal_dss.load(&buf, pdat->value.data.bo.bytes, pdat->value.data.bo.size);
pdat->value.data.bo.bytes = NULL;
pdat->value.data.bo.size = 0;
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(&buf, &jdata, &cnt, ORTE_JOB))) {
OBJ_DESTRUCT(&buf);
goto release;
}
OBJ_DESTRUCT(&buf);
if (ORTE_SUCCESS != (rc = orte_pmix_server_register_nspace(jdata))) {
OBJ_RELEASE(jdata);
goto release;
}
OBJ_RELEASE(jdata); // no reason to keep this around
/* restart the cnct processor */
ORTE_PMIX_OPERATION(cd->procs, cd->info, _cnct, cd->cbfunc, cd->cbdata);
OBJ_RELEASE(cd);
release:
if (NULL != cd->cbfunc) {
cd->cbfunc(rc, cd->cbdata);
}
OBJ_RELEASE(cd);
}
static void _cnct(int sd, short args, void *cbdata)
{
orte_pmix_server_op_caddy_t *cd = (orte_pmix_server_op_caddy_t*)cbdata;
orte_namelist_t *nm;
char **keys = NULL, *key;
orte_job_t *jdata;
int rc = ORTE_SUCCESS;
/* at some point, we need to add bookeeping to track which
* procs are "connected" so we know who to notify upon
* termination or failure. For now, we have to ensure
* that we have registered all participating nspaces so
* the embedded PMIx server can provide them to the client.
* Otherwise, the client will receive an error as it won't
* be able to resolve any of the required data for the
* missing nspaces */
/* cycle thru the procs */
OPAL_LIST_FOREACH(nm, cd->procs, orte_namelist_t) {
/* see if we have the job object for this job */
if (NULL == (jdata = orte_get_job_data_object(nm->name.jobid))) {
/* we don't know about this job. If our "global" data
* server is just our HNP, then we have no way of finding
* out about it, and all we can do is return an error */
if (orte_pmix_server_globals.server.jobid == ORTE_PROC_MY_HNP->jobid &&
orte_pmix_server_globals.server.vpid == ORTE_PROC_MY_HNP->vpid) {
rc = ORTE_ERR_NOT_SUPPORTED;
goto release;
}
/* ask the global data server for the data - if we get it,
* then we can complete the request */
key = opal_convert_jobid_to_string(nm->name.jobid);
opal_argv_append_nosize(&keys, key);
free(key);
if (ORTE_SUCCESS != (rc = pmix_server_lookup_fn(&nm->name, keys, cd->info, _cnlk, cd))) {
opal_argv_free(keys);
goto release;
}
opal_argv_free(keys);
/* the callback function on this lookup will return us to this
* routine so we can continue the process */
return;
}
/* we know about the job - check to ensure it has been
* registered with the local PMIx server */
if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_NSPACE_REGISTERED, NULL, OPAL_BOOL)) {
/* it hasn't been registered yet, so register it now */
if (ORTE_SUCCESS != (rc = orte_pmix_server_register_nspace(jdata))) {
goto release;
}
}
}
release:
if (NULL != cd->cbfunc) {
cd->cbfunc(rc, cd->cbdata);
}
OBJ_RELEASE(cd);
}
int pmix_server_connect_fn(opal_list_t *procs, opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata)
{
@ -335,26 +449,52 @@ int pmix_server_connect_fn(opal_list_t *procs, opal_list_t *info,
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)opal_list_get_size(procs));
/* for now, just ack the call */
if (NULL != cbfunc) {
cbfunc(OPAL_SUCCESS, cbdata);
/* protect ourselves */
if (NULL == procs || 0 == opal_list_get_size(procs)) {
return ORTE_ERR_BAD_PARAM;
}
/* must thread shift this as we will be accessing global data */
ORTE_PMIX_OPERATION(procs, info, _cnct, cbfunc, cbdata);
return ORTE_SUCCESS;
}
return OPAL_SUCCESS;
static void mdxcbfunc(int status,
const char *data, size_t ndata, void *cbdata,
opal_pmix_release_cbfunc_t relcbfunc, void *relcbdata)
{
orte_pmix_server_op_caddy_t *cd = (orte_pmix_server_op_caddy_t*)cbdata;
/* ack the call */
if (NULL != cd->cbfunc) {
cd->cbfunc(status, cd->cbdata);
}
OBJ_RELEASE(cd);
}
int pmix_server_disconnect_fn(opal_list_t *procs, opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata)
{
orte_pmix_server_op_caddy_t *cd;
int rc;
opal_output_verbose(2, orte_pmix_server_globals.output,
"%s disconnect called with %d procs",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)opal_list_get_size(procs));
/* for now, just ack the call */
if (NULL != cbfunc) {
cbfunc(OPAL_SUCCESS, cbdata);
/* at some point, we need to add bookeeping to track which
* procs are "connected" so we know who to notify upon
* termination or failure. For now, just execute a fence
* Note that we do not need to thread-shift here as the
* fence function will do it for us */
cd = OBJ_NEW(orte_pmix_server_op_caddy_t);
cd->cbfunc = cbfunc;
cd->cbdata = cbdata;
if (ORTE_SUCCESS != (rc = pmix_server_fencenb_fn(procs, info, NULL, 0,
mdxcbfunc, cd))) {
OBJ_RELEASE(cd);
}
return OPAL_SUCCESS;
return rc;
}

Просмотреть файл

@ -72,10 +72,8 @@ OBJ_CLASS_DECLARATION(pmix_server_req_t);
typedef struct {
opal_object_t super;
opal_event_t ev;
orte_job_t *jdata;
orte_process_name_t proc;
int status;
orte_proc_t *object;
opal_list_t *procs;
opal_list_t *info;
opal_pmix_op_cbfunc_t cbfunc;
void *cbdata;
} orte_pmix_server_op_caddy_t;
@ -115,21 +113,18 @@ do { \
opal_event_active(&(_req->ev), OPAL_EV_WRITE, 1); \
} while(0);
#define ORTE_PMIX_OPERATION(n, r, ob, s, fn, cf, cb) \
do { \
orte_pmix_server_op_caddy_t *_cd; \
_cd = OBJ_NEW(orte_pmix_server_op_caddy_t); \
/* convert the namespace to jobid and create name */ \
orte_util_convert_string_to_jobid(&(_cd->proc.jobid), (n)); \
_cd->proc.vpid = (r); \
_cd->object = (ob); \
_cd->cbfunc = (cf); \
_cd->cbdata = (cb); \
_cd->status = (s); \
opal_event_set(orte_event_base, &(_cd->ev), -1, \
OPAL_EV_WRITE, (fn), _cd); \
opal_event_set_priority(&(_cd->ev), ORTE_MSG_PRI); \
opal_event_active(&(_cd->ev), OPAL_EV_WRITE, 1); \
#define ORTE_PMIX_OPERATION(p, i, fn, cf, cb) \
do { \
orte_pmix_server_op_caddy_t *_cd; \
_cd = OBJ_NEW(orte_pmix_server_op_caddy_t); \
_cd->procs = (p); \
_cd->info = (i); \
_cd->cbfunc = (cf); \
_cd->cbdata = (cb); \
opal_event_set(orte_event_base, &(_cd->ev), -1, \
OPAL_EV_WRITE, (fn), _cd); \
opal_event_set_priority(&(_cd->ev), ORTE_MSG_PRI); \
opal_event_active(&(_cd->ev), OPAL_EV_WRITE, 1); \
} while(0);

Просмотреть файл

@ -387,6 +387,9 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
opal_list_append(pmap, &kv->super);
}
/* mark the job as registered */
orte_set_attribute(&jdata->attributes, ORTE_JOB_NSPACE_REGISTERED, ORTE_ATTR_LOCAL, NULL, OPAL_BOOL);
/* pass it down */
if (OPAL_SUCCESS != opal_pmix.server_register_nspace(jdata->jobid,
jdata->num_local_procs,

Просмотреть файл

@ -259,6 +259,8 @@ const char *orte_attr_key_to_str(orte_attribute_key_t key)
return "JOB-ROOM-NUM";
case ORTE_JOB_LAUNCH_PROXY:
return "JOB-LAUNCH-PROXY";
case ORTE_JOB_NSPACE_REGISTERED:
return "JOB-NSPACE-REGISTERED";
case ORTE_PROC_NOBARRIER:
return "PROC-NOBARRIER";

Просмотреть файл

@ -129,6 +129,7 @@ typedef uint16_t orte_job_flags_t;
#define ORTE_JOB_NOTIFICATIONS (ORTE_JOB_START_KEY + 38) // string - comma-separated list of desired notifications+methods
#define ORTE_JOB_ROOM_NUM (ORTE_JOB_START_KEY + 39) // int - number of remote request's hotel room
#define ORTE_JOB_LAUNCH_PROXY (ORTE_JOB_START_KEY + 40) // opal_process_name_t - name of spawn requestor
#define ORTE_JOB_NSPACE_REGISTERED (ORTE_JOB_START_KEY + 41) // bool - job has been registered with embedded PMIx server
#define ORTE_JOB_MAX_KEY 300