static void xcast_recv(int status, orte_process_name_t* sender,
opal_buffer_t* buffer, orte_rml_tag_t tg,
void* cbdata)
{
opal_list_item_t *item;
orte_namelist_t *nm;
int ret, cnt;
opal_buffer_t *relay, *rly;
orte_daemon_cmd_flag_t command = ORTE_DAEMON_NULL_CMD;
opal_buffer_t wireup;
opal_byte_object_t *bo;
int8_t flag;
orte_job_t *jdata;
orte_proc_t *rec;
opal_list_t coll;
orte_grpcomm_signature_t *sig;
orte_rml_tag_t tag;
OPAL_OUTPUT_VERBOSE((1, orte_grpcomm_base_framework.framework_output,
"%s grpcomm:direct:xcast:recv: with %d bytes",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)buffer->bytes_used));
/* we need a passthru buffer to send to our children */
rly = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(rly, buffer);
/* get the signature that we do not need */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &sig, &cnt, ORTE_SIGNATURE))) {
ORTE_ERROR_LOG(ret);
ORTE_FORCED_TERMINATE(ret);
return;
}
OBJ_RELEASE(sig);
/* get the target tag */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &tag, &cnt, ORTE_RML_TAG))) {
ORTE_ERROR_LOG(ret);
ORTE_FORCED_TERMINATE(ret);
return;
}
/* setup a buffer we can pass to ourselves - this just contains
* the initial message, minus the headers inserted by xcast itself */
relay = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(relay, buffer);
/* setup the relay list */
OBJ_CONSTRUCT(&coll, opal_list_t);
/* if this is headed for the daemon command processor,
* then we first need to check for add_local_procs
* as that command includes some needed wireup info */
if (ORTE_RML_TAG_DAEMON == tag) {
/* peek at the command */
cnt=1;
if (ORTE_SUCCESS == (ret = opal_dss.unpack(buffer, &command, &cnt, ORTE_DAEMON_CMD))) {
/* if it is add_procs, then... */
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(buffer, &bo, &cnt, OPAL_BYTE_OBJECT))) {
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();
/* see if we have wiring info as well */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &flag, &cnt, OPAL_INT8))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
if (ORTE_DAEMON_ADD_LOCAL_PROCS == command) {
OBJ_RELEASE(relay);
relay = OBJ_NEW(opal_buffer_t);
/* repack the command */
if (OPAL_SUCCESS != (ret = opal_dss.pack(relay, &command, 1, ORTE_DAEMON_CMD))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
if (0 == flag) {
/* copy the remainder of the payload */
opal_dss.copy_payload(relay, buffer);
/* no - just return */
goto relay;
}
}
/* unpack the byte object */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &bo, &cnt, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
if (0 < bo->size) {
/* load it into a buffer */
OBJ_CONSTRUCT(&wireup, opal_buffer_t);
opal_dss.load(&wireup, bo->bytes, bo->size);
/* pass it for processing */
if (ORTE_SUCCESS != (ret = orte_routed.init_routes(ORTE_PROC_MY_NAME->jobid, &wireup))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&wireup);
goto relay;
}
/* done with the wireup buffer - dump it */
OBJ_DESTRUCT(&wireup);
}
free(bo);
if (ORTE_DAEMON_ADD_LOCAL_PROCS == command) {
/* copy the remainder of the payload */
opal_dss.copy_payload(relay, buffer);
}
}
} else {
ORTE_ERROR_LOG(ret);
goto CLEANUP;
}
}
relay:
/* get the list of next recipients from the routed module */
orte_routed.get_routing_list(&coll);
/* if list is empty, no relay is required */
if (opal_list_is_empty(&coll)) {
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base_framework.framework_output,
"%s grpcomm:direct:send_relay - recipient list is empty!",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
OBJ_RELEASE(rly);
goto CLEANUP;
}
/* send the message to each recipient on list, deconstructing it as we go */
while (NULL != (item = opal_list_remove_first(&coll))) {
nm = (orte_namelist_t*)item;
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base_framework.framework_output,
"%s grpcomm:direct:send_relay sending relay msg of %d bytes to %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (int)rly->bytes_used,
ORTE_NAME_PRINT(&nm->name)));
OBJ_RETAIN(rly);
/* check the state of the recipient - no point
* sending to someone not alive
*/
jdata = orte_get_job_data_object(nm->name.jobid);
if (NULL == (rec = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, nm->name.vpid))) {
opal_output(0, "%s grpcomm:direct:send_relay proc %s not found - cannot relay",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(&nm->name));
OBJ_RELEASE(rly);
OBJ_RELEASE(item);
continue;
}
if (ORTE_PROC_STATE_RUNNING < rec->state) {
opal_output(0, "%s grpcomm:direct:send_relay proc %s not running - cannot relay",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(&nm->name));
OBJ_RELEASE(rly);
OBJ_RELEASE(item);
continue;
}
if (ORTE_SUCCESS != (ret = orte_rml.send_buffer_nb(&nm->name, rly, ORTE_RML_TAG_XCAST,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(rly);
OBJ_RELEASE(item);
continue;
}
OBJ_RELEASE(item);
}
OBJ_RELEASE(rly); // retain accounting
CLEANUP:
/* cleanup */
OBJ_DESTRUCT(&coll);
/* now send the relay buffer to myself for processing */
if (ORTE_DAEMON_DVM_NIDMAP_CMD != command) {
if (ORTE_SUCCESS != (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_NAME, relay, tag,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(relay);
}
}
}
void orte_grpcomm_base_xcast_recv(int status, orte_process_name_t* sender,
opal_buffer_t* buffer, orte_rml_tag_t tag,
void* cbdata)
{
opal_list_item_t *item;
orte_namelist_t *nm;
int ret, cnt;
opal_buffer_t *relay, *rly;
orte_daemon_cmd_flag_t command;
opal_buffer_t wireup;
opal_byte_object_t *bo;
int8_t flag;
orte_grpcomm_collective_t coll;
OPAL_OUTPUT_VERBOSE((1, orte_grpcomm_base.output,
"%s grpcomm:xcast:recv:send_relay",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* setup the relay message */
relay = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(relay, buffer);
/* peek at the command */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &command, &cnt, ORTE_DAEMON_CMD))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
/* if it is add_procs, then... */
if (ORTE_DAEMON_ADD_LOCAL_PROCS == command) {
/* extract the byte object holding the daemonmap */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &bo, &cnt, OPAL_BYTE_OBJECT))) {
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 bo
*/
if (ORTE_PROC_IS_HNP) {
/* no need - already have the info */
if (NULL != bo->bytes) {
free(bo->bytes);
}
} else {
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base.output,
"%s grpcomm:base: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();
/* see if we have wiring info as well */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &flag, &cnt, OPAL_INT8))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
if (0 == flag) {
/* no - just return */
goto relay;
}
/* unpack the byte object */
cnt=1;
if (ORTE_SUCCESS != (ret = opal_dss.unpack(buffer, &bo, &cnt, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(ret);
goto relay;
}
if (0 < bo->size) {
/* load it into a buffer */
OBJ_CONSTRUCT(&wireup, opal_buffer_t);
opal_dss.load(&wireup, bo->bytes, bo->size);
/* pass it for processing */
if (ORTE_SUCCESS != (ret = orte_routed.init_routes(ORTE_PROC_MY_NAME->jobid, &wireup))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&wireup);
goto relay;
}
/* done with the wireup buffer - dump it */
OBJ_DESTRUCT(&wireup);
}
}
relay:
/* setup the relay list */
OBJ_CONSTRUCT(&coll, orte_grpcomm_collective_t);
/* get the list of next recipients from the routed module */
orte_routed.get_routing_list(ORTE_GRPCOMM_XCAST, &coll);
/* if list is empty, no relay is required */
if (opal_list_is_empty(&coll.targets)) {
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base.output,
"%s orte:daemon:send_relay - recipient list is empty!",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto CLEANUP;
}
/* send the message to each recipient on list, deconstructing it as we go */
while (NULL != (item = opal_list_remove_first(&coll.targets))) {
nm = (orte_namelist_t*)item;
OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base.output,
"%s orte:daemon:send_relay sending relay msg to %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&nm->name)));
rly = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(rly, relay);
if (0 > (ret = orte_rml.send_buffer_nb(&nm->name, rly, ORTE_RML_TAG_XCAST, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(rly);
continue;
}
}
CLEANUP:
/* cleanup */
OBJ_DESTRUCT(&coll);
/* now send it to myself for processing */
if (0 > (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_NAME, relay,
ORTE_RML_TAG_DAEMON, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(relay);
}
}
