int orte_ess_env_put(orte_std_cntr_t num_procs,
orte_std_cntr_t num_local_procs,
char ***env)
{
char* param;
char* value;
/* tell the ESS to select the PMI component - but don't override
* anything that may have been provided elsewhere
*/
if(OPAL_SUCCESS != mca_base_var_env_name ("ess", ¶m)) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
opal_setenv(param, "pmi", false, env);
free(param);
/* since we want to pass the name as separate components, make sure
* that the "name" environmental variable is cleared!
*/
if(OPAL_SUCCESS != mca_base_var_env_name ("orte_ess_name", ¶m)) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
opal_unsetenv(param, env);
free(param);
asprintf(&value, "%ld", (long) num_procs);
if(OPAL_SUCCESS != mca_base_var_env_name ("orte_ess_num_procs", ¶m)) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
opal_setenv(param, value, true, env);
free(param);
/* although the num_procs is the comm_world size, users
* would appreciate being given a public environmental variable
* that also represents this value - something MPI specific - so
* do that here.
*
* AND YES - THIS BREAKS THE ABSTRACTION BARRIER TO SOME EXTENT.
* We know - just live with it
*/
opal_setenv("OMPI_COMM_WORLD_SIZE", value, true, env);
free(value);
/* users would appreciate being given a public environmental variable
* that also represents this value - something MPI specific - so
* do that here.
*
* AND YES - THIS BREAKS THE ABSTRACTION BARRIER TO SOME EXTENT.
* We know - just live with it
*/
asprintf(&value, "%ld", (long) num_local_procs);
opal_setenv("OMPI_COMM_WORLD_LOCAL_SIZE", value, true, env);
free(value);
return ORTE_SUCCESS;
}
int opal_cr_refresh_environ(int prev_pid) {
char *file_name;
#if OPAL_ENABLE_CRDEBUG == 1
char *tmp;
#endif
struct stat file_status;
if( 0 >= prev_pid ) {
prev_pid = getpid();
}
/*
* Make sure the file exists. If it doesn't then this means 2 things:
* 1) We have already executed this function, and
* 2) The file has been deleted on the previous round.
*/
asprintf(&file_name, "%s/%s-%d", opal_tmp_directory(), OPAL_CR_BASE_ENV_NAME, prev_pid);
if (NULL == file_name) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
if(0 != stat(file_name, &file_status) ){
free(file_name);
return OPAL_SUCCESS;
}
#if OPAL_ENABLE_CRDEBUG == 1
mca_base_var_env_name ("opal_cr_enable_crdebug", &tmp);
opal_unsetenv(tmp, &environ);
free (tmp);
#endif
extract_env_vars(prev_pid, file_name);
#if OPAL_ENABLE_CRDEBUG == 1
MPIR_debug_with_checkpoint = 0;
(void) mca_base_var_register ("opal", "opal", "cr", "enable_crdebug",
"Enable checkpoint/restart debugging",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_8, MCA_BASE_VAR_SCOPE_ALL_EQ,
&MPIR_debug_with_checkpoint);
opal_output_verbose(10, opal_cr_output,
"opal_cr: init: C/R Debugging Enabled [%s] (refresh)\n",
(MPIR_debug_with_checkpoint ? "True": "False"));
#endif
free(file_name);
return OPAL_SUCCESS;
}
int opal_cr_refresh_environ(int prev_pid) {
int val;
char *file_name = NULL;
struct stat file_status;
if( 0 >= prev_pid ) {
prev_pid = getpid();
}
/*
* Make sure the file exists. If it doesn't then this means 2 things:
* 1) We have already executed this function, and
* 2) The file has been deleted on the previous round.
*/
asprintf(&file_name, "%s/%s-%d", opal_tmp_directory(), OPAL_CR_BASE_ENV_NAME, prev_pid);
if(0 != stat(file_name, &file_status) ){
return OPAL_SUCCESS;
}
#if OPAL_ENABLE_CRDEBUG == 1
opal_unsetenv(mca_base_param_env_var("opal_cr_enable_crdebug"), &environ);
#endif
extract_env_vars(prev_pid, file_name);
#if OPAL_ENABLE_CRDEBUG == 1
mca_base_param_reg_int_name("opal_cr", "enable_crdebug",
"Enable checkpoint/restart debugging",
false, false,
0,
&val);
MPIR_debug_with_checkpoint = OPAL_INT_TO_BOOL(val);
opal_output_verbose(10, opal_cr_output,
"opal_cr: init: C/R Debugging Enabled [%s] (refresh)\n",
(MPIR_debug_with_checkpoint ? "True": "False"));
#else
val = 0; /* Silence Compiler warning */
#endif
if( NULL != file_name ){
free(file_name);
file_name = NULL;
}
return OPAL_SUCCESS;
}
int orte_daemon(int argc, char *argv[])
{
int ret = 0;
opal_cmd_line_t *cmd_line = NULL;
char *rml_uri;
int i;
opal_buffer_t *buffer;
char hostname[100];
char *tmp_env_var = NULL;
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
/* initialize the singleton died pipe to an illegal value so we can detect it was set */
orted_globals.singleton_died_pipe = -1;
/* init the failure orted vpid to an invalid value */
orted_globals.fail = ORTE_VPID_INVALID;
/* setup to check common command line options that just report and die */
cmd_line = OBJ_NEW(opal_cmd_line_t);
if (OPAL_SUCCESS != opal_cmd_line_create(cmd_line, orte_cmd_line_opts)) {
OBJ_RELEASE(cmd_line);
exit(1);
}
mca_base_cmd_line_setup(cmd_line);
if (ORTE_SUCCESS != (ret = opal_cmd_line_parse(cmd_line, false,
argc, argv))) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
fprintf(stderr, "Usage: %s [OPTION]...\n%s\n", argv[0], args);
free(args);
OBJ_RELEASE(cmd_line);
return ret;
}
/*
* Since this process can now handle MCA/GMCA parameters, make sure to
* process them.
*/
mca_base_cmd_line_process_args(cmd_line, &environ, &environ);
/* Ensure that enough of OPAL is setup for us to be able to run */
/*
* NOTE: (JJH)
* We need to allow 'mca_base_cmd_line_process_args()' to process command
* line arguments *before* calling opal_init_util() since the command
* line could contain MCA parameters that affect the way opal_init_util()
* functions. AMCA parameters are one such option normally received on the
* command line that affect the way opal_init_util() behaves.
* It is "safe" to call mca_base_cmd_line_process_args() before
* opal_init_util() since mca_base_cmd_line_process_args() does *not*
* depend upon opal_init_util() functionality.
*/
if (OPAL_SUCCESS != opal_init_util(&argc, &argv)) {
fprintf(stderr, "OPAL failed to initialize -- orted aborting\n");
exit(1);
}
/* save the environment for launch purposes. This MUST be
* done so that we can pass it to any local procs we
* spawn - otherwise, those local procs won't see any
* non-MCA envars that were set in the enviro when the
* orted was executed - e.g., by .csh
*/
orte_launch_environ = opal_argv_copy(environ);
/* purge any ess flag set in the environ when we were launched */
opal_unsetenv("OMPI_MCA_ess", &orte_launch_environ);
/* if orte_daemon_debug is set, let someone know we are alive right
* away just in case we have a problem along the way
*/
if (orted_globals.debug) {
gethostname(hostname, 100);
fprintf(stderr, "Daemon was launched on %s - beginning to initialize\n", hostname);
}
/* check for help request */
if (orted_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
orte_show_help("help-orted.txt", "orted:usage", false,
argv[0], args);
free(args);
return 1;
}
#if defined(HAVE_SETSID) && !defined(__WINDOWS__)
/* see if we were directed to separate from current session */
if (orted_globals.set_sid) {
setsid();
}
#endif /* !defined(__WINDOWS__) */
/* see if they want us to spin until they can connect a debugger to us */
i=0;
while (orted_spin_flag) {
i++;
if (1000 < i) i=0;
}
#if OPAL_ENABLE_FT_CR == 1
/* Mark as a tool program */
tmp_env_var = mca_base_param_env_var("opal_cr_is_tool");
opal_setenv(tmp_env_var,
"1",
true, &environ);
free(tmp_env_var);
#endif
tmp_env_var = NULL; /* Silence compiler warning */
/* if mapreduce set, flag it */
if (orted_globals.mapreduce) {
orte_map_reduce = true;
}
/* Set the flag telling OpenRTE that I am NOT a
* singleton, but am "infrastructure" - prevents setting
* up incorrect infrastructure that only a singleton would
* require.
*/
if (orted_globals.hnp) {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
ORTE_ERROR_LOG(ret);
return ret;
}
} else {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_DAEMON))) {
ORTE_ERROR_LOG(ret);
return ret;
}
}
/* finalize the OPAL utils. As they are opened again from orte_init->opal_init
* we continue to have a reference count on them. So we have to finalize them twice...
*/
opal_finalize_util();
if ((int)ORTE_VPID_INVALID != orted_globals.fail) {
orted_globals.abort=false;
/* some vpid was ordered to fail. The value can be positive
* or negative, depending upon the desired method for failure,
* so need to check both here
*/
if (0 > orted_globals.fail) {
orted_globals.fail = -1*orted_globals.fail;
orted_globals.abort = true;
}
/* are we the specified vpid? */
if ((int)ORTE_PROC_MY_NAME->vpid == orted_globals.fail) {
/* if the user specified we delay, then setup a timer
* and have it kill us
*/
if (0 < orted_globals.fail_delay) {
ORTE_TIMER_EVENT(orted_globals.fail_delay, 0, shutdown_callback, ORTE_SYS_PRI);
} else {
opal_output(0, "%s is executing clean %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_globals.abort ? "abort" : "abnormal termination");
/* do -not- call finalize as this will send a message to the HNP
* indicating clean termination! Instead, just forcibly cleanup
* the local session_dir tree and exit
*/
orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
/* if we were ordered to abort, do so */
if (orted_globals.abort) {
abort();
}
/* otherwise, return with non-zero status */
ret = ORTE_ERROR_DEFAULT_EXIT_CODE;
goto DONE;
}
}
}
/* detach from controlling terminal
* otherwise, remain attached so output can get to us
*/
if(!orte_debug_flag &&
!orte_debug_daemons_flag &&
orted_globals.daemonize) {
opal_daemon_init(NULL);
}
/* insert our contact info into our process_info struct so we
* have it for later use and set the local daemon field to our name
*/
orte_process_info.my_daemon_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_DAEMON->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_DAEMON->vpid = ORTE_PROC_MY_NAME->vpid;
/* if I am also the hnp, then update that contact info field too */
if (ORTE_PROC_IS_HNP) {
orte_process_info.my_hnp_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_HNP->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_HNP->vpid = ORTE_PROC_MY_NAME->vpid;
}
/* setup the primary daemon command receive function */
ret = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);
if (ret != ORTE_SUCCESS && ret != ORTE_ERR_NOT_IMPLEMENTED) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* output a message indicating we are alive, our name, and our pid
* for debugging purposes
*/
if (orte_debug_daemons_flag) {
fprintf(stderr, "Daemon %s checking in as pid %ld on host %s\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.pid,
orte_process_info.nodename);
}
/* We actually do *not* want the orted to voluntarily yield() the
processor more than necessary. The orted already blocks when
it is doing nothing, so it doesn't use any more CPU cycles than
it should; but when it *is* doing something, we do not want it
to be unnecessarily delayed because it voluntarily yielded the
processor in the middle of its work.
For example: when a message arrives at the orted, we want the
OS to wake up the orted in a timely fashion (which most OS's
seem good about doing) and then we want the orted to process
the message as fast as possible. If the orted yields and lets
aggressive MPI applications get the processor back, it may be a
long time before the OS schedules the orted to run again
(particularly if there is no IO event to wake it up). Hence,
routed OOB messages (for example) may be significantly delayed
before being delivered to MPI processes, which can be
problematic in some scenarios (e.g., COMM_SPAWN, BTL's that
require OOB messages for wireup, etc.). */
opal_progress_set_yield_when_idle(false);
/* Change the default behavior of libevent such that we want to
continually block rather than blocking for the default timeout
and then looping around the progress engine again. There
should be nothing in the orted that cannot block in libevent
until "something" happens (i.e., there's no need to keep
cycling through progress because the only things that should
happen will happen in libevent). This is a minor optimization,
but what the heck... :-) */
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE);
/* if requested, obtain and report a new process name and my uri to the indicated pipe */
if (orted_globals.uri_pipe > 0) {
orte_job_t *jdata;
orte_proc_t *proc;
orte_node_t *node;
orte_app_context_t *app;
char *tmp, *nptr, *sysinfo;
int32_t ljob;
/* setup the singleton's job */
jdata = OBJ_NEW(orte_job_t);
orte_plm_base_create_jobid(jdata);
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* must create a map for it (even though it has no
* info in it) so that the job info will be picked
* up in subsequent pidmaps or other daemons won't
* know how to route
*/
jdata->map = OBJ_NEW(orte_job_map_t);
/* setup an app_context for the singleton */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup("singleton");
app->num_procs = 1;
opal_pointer_array_add(jdata->apps, app);
#if 0
/* run our local allocator to read the available
* allocation in case this singleton decides to
* comm_spawn other procs
*/
if (ORTE_SUCCESS != (ret = orte_ras.allocate(jdata))) {
ORTE_ERROR_LOG(ret);
/* don't quit as this would cause the singleton
* to hang!
*/
}
#endif
/* setup a proc object for the singleton - since we
* -must- be the HNP, and therefore we stored our
* node on the global node pool, and since the singleton
* -must- be on the same node as us, indicate that
*/
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = jdata->jobid;
proc->name.vpid = 0;
proc->alive = true;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
/* obviously, it is on my node */
node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0);
proc->node = node;
OBJ_RETAIN(node); /* keep accounting straight */
opal_pointer_array_add(jdata->procs, proc);
jdata->num_procs = 1;
/* and obviously it is one of my local procs */
OBJ_RETAIN(proc);
opal_pointer_array_add(orte_local_children, proc);
jdata->num_local_procs = 1;
/* set the trivial */
proc->local_rank = 0;
proc->node_rank = 0;
proc->app_rank = 0;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->alive = true;
proc->app_idx = 0;
proc->local_proc = true;
#if OPAL_HAVE_HWLOC
proc->bind_idx = 0;
#endif
/* the singleton will use the first three collectives
* for its modex/barriers
*/
orte_grpcomm_base.coll_id += 3;
/* need to setup a pidmap for it */
jdata->pmap = (opal_byte_object_t*)malloc(sizeof(opal_byte_object_t));
if (ORTE_SUCCESS != (ret = orte_util_encode_pidmap(jdata->pmap))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* if we don't yet have a daemon map, then we have to generate one
* to pass back to it
*/
if (NULL == orte_odls_globals.dmap) {
orte_odls_globals.dmap = (opal_byte_object_t*)malloc(sizeof(opal_byte_object_t));
/* construct a nodemap */
if (ORTE_SUCCESS != (ret = orte_util_encode_nodemap(orte_odls_globals.dmap))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* create a string that contains our uri + the singleton's name + sysinfo */
orte_util_convert_process_name_to_string(&nptr, &proc->name);
orte_util_convert_sysinfo_to_string(&sysinfo, orte_local_cpu_type, orte_local_cpu_model);
asprintf(&tmp, "%s[%s][%s]", orte_process_info.my_daemon_uri, nptr, sysinfo);
free(nptr);
free(sysinfo);
/* pass that info to the singleton */
#ifndef __WINDOWS__
write(orted_globals.uri_pipe, tmp, strlen(tmp)+1); /* need to add 1 to get the NULL */
#else
send(orted_globals.uri_pipe, tmp, strlen(tmp)+1, 0); /* need to add 1 to get the NULL */
#endif
/* cleanup */
free(tmp);
}
/* if we were given a pipe to monitor for singleton termination, set that up */
if (orted_globals.singleton_died_pipe > 0) {
/* register shutdown handler */
pipe_handler = (opal_event_t*)malloc(sizeof(opal_event_t));
opal_event_set(orte_event_base, pipe_handler,
orted_globals.singleton_died_pipe,
OPAL_EV_READ,
pipe_closed,
pipe_handler);
opal_event_add(pipe_handler, NULL);
}
/* If I have a parent, then save his contact info so
* any messages we send can flow thru him.
*/
mca_base_param_reg_string_name("orte", "parent_uri",
"URI for the parent if tree launch is enabled.",
true, false, NULL, &rml_uri);
if (NULL != rml_uri) {
orte_process_name_t parent;
/* set the contact info into the hash table */
if (ORTE_SUCCESS != (ret = orte_rml.set_contact_info(rml_uri))) {
ORTE_ERROR_LOG(ret);
free(rml_uri);
goto DONE;
}
ret = orte_rml_base_parse_uris(rml_uri, &parent, NULL );
if( ORTE_SUCCESS != ret ) {
ORTE_ERROR_LOG(ret);
free(rml_uri);
goto DONE;
}
free(rml_uri);
/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(ORTE_PROC_MY_HNP, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(&parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* if we are not the HNP...the only time we will be an HNP
* is if we are launched by a singleton to provide support
* for it
*/
if (!ORTE_PROC_IS_HNP) {
/* send the information to the orted report-back point - this function
* will process the data, but also counts the number of
* orteds that reported back so the launch procedure can continue.
* We need to do this at the last possible second as the HNP
* can turn right around and begin issuing orders to us
*/
buffer = OBJ_NEW(opal_buffer_t);
/* insert our name for rollup purposes */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* for now, always include our contact info, even if we are using
* static ports. Eventually, this will be removed
*/
rml_uri = orte_rml.get_contact_info();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &rml_uri, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* include our node name */
opal_dss.pack(buffer, &orte_process_info.nodename, 1, OPAL_STRING);
#if OPAL_HAVE_HWLOC
/* add the local topology */
if (NULL != opal_hwloc_topology &&
(1 == ORTE_PROC_MY_NAME->vpid || orte_hetero_nodes)) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &opal_hwloc_topology, 1, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(ret);
}
}
#endif
if ((orte_static_ports || orte_use_common_port) && !orted_globals.tree_spawn) {
/* use the rollup collective to send our data to the HNP
* so we minimize the HNP bottleneck
*/
orte_grpcomm_collective_t *coll;
coll = OBJ_NEW(orte_grpcomm_collective_t);
/* get the list of contributors we need from the routed module */
orte_routed.get_routing_list(ORTE_GRPCOMM_COLL_PEERS, coll);
/* add the collective to our list */
opal_list_append(&orte_grpcomm_base.active_colls, &coll->super);
/* send the buffer to ourselves to start the collective */
if (0 > (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_NAME, buffer,
ORTE_RML_TAG_ROLLUP, 0,
rml_cbfunc, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
} else {
/* send directly to the HNP's callback */
if (0 > (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, buffer,
ORTE_RML_TAG_ORTED_CALLBACK, 0,
rml_cbfunc, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
}
if (orte_debug_daemons_flag) {
opal_output(0, "%s orted: up and running - waiting for commands!", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
}
/* loop the event lib until an exit event is detected */
while (orte_event_base_active) {
opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
}
/* ensure all local procs are dead */
orte_odls.kill_local_procs(NULL);
DONE:
/* update the exit status, in case it wasn't done */
ORTE_UPDATE_EXIT_STATUS(orte_exit_status);
/* cleanup and leave */
orte_finalize();
if (orte_debug_flag) {
fprintf(stderr, "exiting with status %d\n", orte_exit_status);
}
exit(orte_exit_status);
}
static int setup_fork(orte_job_t *jdata,
orte_app_context_t *app)
{
int i;
char *param;
bool oversubscribed;
orte_node_t *node;
char **envcpy, **nps, **firstranks;
char *npstring, *firstrankstring;
char *num_app_ctx;
bool takeus = false;
/* see if we are included */
for (i=0; NULL != jdata->personality[i]; i++) {
if (0 == strcmp(jdata->personality[i], "ompi")) {
takeus = true;
break;
}
}
if (!takeus) {
return ORTE_ERR_TAKE_NEXT_OPTION;
}
/* see if the mapper thinks we are oversubscribed */
oversubscribed = false;
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, ORTE_PROC_MY_NAME->vpid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_OVERSUBSCRIBED)) {
oversubscribed = true;
}
/* setup base environment: copy the current environ and merge
in the app context environ */
if (NULL != app->env) {
/* manually free original context->env to avoid a memory leak */
char **tmp = app->env;
envcpy = opal_environ_merge(orte_launch_environ, app->env);
if (NULL != tmp) {
opal_argv_free(tmp);
}
} else {
envcpy = opal_argv_copy(orte_launch_environ);
}
app->env = envcpy;
/* special case handling for --prefix: this is somewhat icky,
but at least some users do this. :-\ It is possible that
when using --prefix, the user will also "-x PATH" and/or
"-x LD_LIBRARY_PATH", which would therefore clobber the
work that was done in the prior pls to ensure that we have
the prefix at the beginning of the PATH and
LD_LIBRARY_PATH. So examine the context->env and see if we
find PATH or LD_LIBRARY_PATH. If found, that means the
prior work was clobbered, and we need to re-prefix those
variables. */
param = NULL;
orte_get_attribute(&app->attributes, ORTE_APP_PREFIX_DIR, (void**)¶m, OPAL_STRING);
for (i = 0; NULL != param && NULL != app->env && NULL != app->env[i]; ++i) {
char *newenv;
/* Reset PATH */
if (0 == strncmp("PATH=", app->env[i], 5)) {
asprintf(&newenv, "%s/bin:%s", param, app->env[i] + 5);
opal_setenv("PATH", newenv, true, &app->env);
free(newenv);
}
/* Reset LD_LIBRARY_PATH */
else if (0 == strncmp("LD_LIBRARY_PATH=", app->env[i], 16)) {
asprintf(&newenv, "%s/lib:%s", param, app->env[i] + 16);
opal_setenv("LD_LIBRARY_PATH", newenv, true, &app->env);
free(newenv);
}
}
if (NULL != param) {
free(param);
}
/* pass my contact info to the local proc so we can talk */
opal_setenv("OMPI_MCA_orte_local_daemon_uri", orte_process_info.my_daemon_uri, true, &app->env);
/* pass the hnp's contact info to the local proc in case it
* needs it
*/
if (NULL != orte_process_info.my_hnp_uri) {
opal_setenv("OMPI_MCA_orte_hnp_uri", orte_process_info.my_hnp_uri, true, &app->env);
}
/* setup yield schedule - do not override any user-supplied directive! */
if (oversubscribed) {
opal_setenv("OMPI_MCA_mpi_yield_when_idle", "1", false, &app->env);
} else {
opal_setenv("OMPI_MCA_mpi_yield_when_idle", "0", false, &app->env);
}
/* set the app_context number into the environment */
asprintf(¶m, "%ld", (long)app->idx);
opal_setenv("OMPI_MCA_orte_app_num", param, true, &app->env);
free(param);
/* although the total_slots_alloc is the universe size, users
* would appreciate being given a public environmental variable
* that also represents this value - something MPI specific - so
* do that here. Also required by the ompi_attributes code!
*
* AND YES - THIS BREAKS THE ABSTRACTION BARRIER TO SOME EXTENT.
* We know - just live with it
*/
asprintf(¶m, "%ld", (long)jdata->total_slots_alloc);
opal_setenv("OMPI_UNIVERSE_SIZE", param, true, &app->env);
free(param);
/* pass the number of nodes involved in this job */
asprintf(¶m, "%ld", (long)(jdata->map->num_nodes));
opal_setenv("OMPI_MCA_orte_num_nodes", param, true, &app->env);
free(param);
/* pass a param telling the child what type and model of cpu we are on,
* if we know it. If hwloc has the value, use what it knows. Otherwise,
* see if we were explicitly given it and use that value.
*/
hwloc_obj_t obj;
char *htmp;
if (NULL != opal_hwloc_topology) {
obj = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL != (htmp = (char*)hwloc_obj_get_info_by_name(obj, "CPUType")) ||
NULL != (htmp = orte_local_cpu_type)) {
opal_setenv("OMPI_MCA_orte_cpu_type", htmp, true, &app->env);
}
if (NULL != (htmp = (char*)hwloc_obj_get_info_by_name(obj, "CPUModel")) ||
NULL != (htmp = orte_local_cpu_model)) {
opal_setenv("OMPI_MCA_orte_cpu_model", htmp, true, &app->env);
}
} else {
if (NULL != orte_local_cpu_type) {
opal_setenv("OMPI_MCA_orte_cpu_type", orte_local_cpu_type, true, &app->env);
}
if (NULL != orte_local_cpu_model) {
opal_setenv("OMPI_MCA_orte_cpu_model", orte_local_cpu_model, true, &app->env);
}
}
/* get shmem's best component name so we can provide a hint to the shmem
* framework. the idea here is to have someone figure out what component to
* select (via the shmem framework) and then have the rest of the
* components in shmem obey that decision. for more details take a look at
* the shmem framework in opal.
*/
if (NULL != (param = opal_shmem_base_best_runnable_component_name())) {
opal_setenv("OMPI_MCA_shmem_RUNTIME_QUERY_hint", param, true, &app->env);
free(param);
}
/* Set an info MCA param that tells the launched processes that
* any binding policy was applied by us (e.g., so that
* MPI_INIT doesn't try to bind itself)
*/
opal_setenv("OMPI_MCA_orte_bound_at_launch", "1", true, &app->env);
/* tell the ESS to avoid the singleton component - but don't override
* anything that may have been provided elsewhere
*/
opal_setenv("OMPI_MCA_ess", "^singleton", false, &app->env);
/* ensure that the spawned process ignores direct launch components,
* but do not overrride anything we were given */
opal_setenv("OMPI_MCA_pmix", "^s1,s2,cray", false, &app->env);
/* since we want to pass the name as separate components, make sure
* that the "name" environmental variable is cleared!
*/
opal_unsetenv("OMPI_MCA_orte_ess_name", &app->env);
asprintf(¶m, "%ld", (long)jdata->num_procs);
opal_setenv("OMPI_MCA_orte_ess_num_procs", param, true, &app->env);
/* although the num_procs is the comm_world size, users
* would appreciate being given a public environmental variable
* that also represents this value - something MPI specific - so
* do that here.
*
* AND YES - THIS BREAKS THE ABSTRACTION BARRIER TO SOME EXTENT.
* We know - just live with it
*/
opal_setenv("OMPI_COMM_WORLD_SIZE", param, true, &app->env);
free(param);
/* users would appreciate being given a public environmental variable
* that also represents this value - something MPI specific - so
* do that here.
*
* AND YES - THIS BREAKS THE ABSTRACTION BARRIER TO SOME EXTENT.
* We know - just live with it
*/
asprintf(¶m, "%ld", (long)jdata->num_local_procs);
opal_setenv("OMPI_COMM_WORLD_LOCAL_SIZE", param, true, &app->env);
free(param);
/* forcibly set the local tmpdir base to match ours */
opal_setenv("OMPI_MCA_orte_tmpdir_base", orte_process_info.tmpdir_base, true, &app->env);
/* MPI-3 requires we provide some further info to the procs,
* so we pass them as envars to avoid introducing further
* ORTE calls in the MPI layer
*/
asprintf(&num_app_ctx, "%lu", (unsigned long)jdata->num_apps);
/* build some common envars we need to pass for MPI-3 compatibility */
nps = NULL;
firstranks = NULL;
for (i=0; i < jdata->apps->size; i++) {
if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
continue;
}
opal_argv_append_nosize(&nps, ORTE_VPID_PRINT(app->num_procs));
opal_argv_append_nosize(&firstranks, ORTE_VPID_PRINT(app->first_rank));
}
npstring = opal_argv_join(nps, ' ');
firstrankstring = opal_argv_join(firstranks, ' ');
opal_argv_free(nps);
opal_argv_free(firstranks);
/* add the MPI-3 envars */
opal_setenv("OMPI_NUM_APP_CTX", num_app_ctx, true, &app->env);
opal_setenv("OMPI_FIRST_RANKS", firstrankstring, true, &app->env);
opal_setenv("OMPI_APP_CTX_NUM_PROCS", npstring, true, &app->env);
free(num_app_ctx);
free(firstrankstring);
free(npstring);
return ORTE_SUCCESS;
}
static int do_child(orte_app_context_t* context,
orte_proc_t *child,
char **environ_copy,
orte_job_t *jobdat, int write_fd,
orte_iof_base_io_conf_t opts)
{
int i, rc;
sigset_t sigs;
long fd, fdmax = sysconf(_SC_OPEN_MAX);
char *param, *msg;
if (orte_forward_job_control) {
/* Set a new process group for this child, so that a
SIGSTOP can be sent to it without being sent to the
orted. */
setpgid(0, 0);
}
/* Setup the pipe to be close-on-exec */
opal_fd_set_cloexec(write_fd);
if (NULL != child) {
/* setup stdout/stderr so that any error messages that we
may print out will get displayed back at orterun.
NOTE: Definitely do this AFTER we check contexts so
that any error message from those two functions doesn't
come out to the user. IF we didn't do it in this order,
THEN a user who gives us a bad executable name or
working directory would get N error messages, where
N=num_procs. This would be very annoying for large
jobs, so instead we set things up so that orterun
always outputs a nice, single message indicating what
happened
*/
if (ORTE_SUCCESS != (i = orte_iof_base_setup_child(&opts,
&environ_copy))) {
ORTE_ERROR_LOG(i);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt",
"iof setup failed",
orte_process_info.nodename, context->app);
/* Does not return */
}
/* now set any child-level controls such as binding */
orte_rtc.set(jobdat, child, &environ_copy, write_fd);
} else if (!ORTE_FLAG_TEST(jobdat, ORTE_JOB_FLAG_FORWARD_OUTPUT)) {
/* tie stdin/out/err/internal to /dev/null */
int fdnull;
for (i=0; i < 3; i++) {
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > i && i != write_fd) {
dup2(fdnull, i);
}
close(fdnull);
}
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > opts.p_internal[1]) {
dup2(fdnull, opts.p_internal[1]);
}
close(fdnull);
}
/* if the user requested it, set the system resource limits */
if (OPAL_SUCCESS != (rc = opal_util_init_sys_limits(&msg))) {
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"set limit",
orte_process_info.nodename, context->app,
__FILE__, __LINE__, msg);
}
/* ensure we only do this once */
(void) mca_base_var_env_name("opal_set_max_sys_limits", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
/* close all file descriptors w/ exception of stdin/stdout/stderr,
the pipe used for the IOF INTERNAL messages, and the pipe up to
the parent. */
for(fd=3; fd<fdmax; fd++) {
if (fd != opts.p_internal[1] && fd != write_fd) {
close(fd);
}
}
if (context->argv == NULL) {
context->argv = malloc(sizeof(char*)*2);
context->argv[0] = strdup(context->app);
context->argv[1] = NULL;
}
/* Set signal handlers back to the default. Do this close to
the exev() because the event library may (and likely will)
reset them. If we don't do this, the event library may
have left some set that, at least on some OS's, don't get
reset via fork() or exec(). Hence, the launched process
could be unkillable (for example). */
set_handler_default(SIGTERM);
set_handler_default(SIGINT);
set_handler_default(SIGHUP);
set_handler_default(SIGPIPE);
set_handler_default(SIGCHLD);
/* Unblock all signals, for many of the same reasons that we
set the default handlers, above. This is noticable on
Linux where the event library blocks SIGTERM, but we don't
want that blocked by the launched process. */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* Exec the new executable */
if (10 < opal_output_get_verbosity(orte_odls_base_framework.framework_output)) {
int jout;
opal_output(0, "%s STARTING %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), context->app);
for (jout=0; NULL != context->argv[jout]; jout++) {
opal_output(0, "%s\tARGV[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, context->argv[jout]);
}
for (jout=0; NULL != environ_copy[jout]; jout++) {
opal_output(0, "%s\tENVIRON[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, environ_copy[jout]);
}
}
execve(context->app, context->argv, environ_copy);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt", "execve error",
orte_process_info.nodename, context->app, strerror(errno));
/* Does not return */
}
int orte_daemon(int argc, char *argv[])
{
int ret = 0;
opal_cmd_line_t *cmd_line = NULL;
char *rml_uri;
int i;
opal_buffer_t *buffer;
char hostname[100];
#if OPAL_ENABLE_FT_CR == 1
char *tmp_env_var = NULL;
#endif
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
/* initialize the singleton died pipe to an illegal value so we can detect it was set */
orted_globals.singleton_died_pipe = -1;
/* init the failure orted vpid to an invalid value */
orted_globals.fail = ORTE_VPID_INVALID;
/* setup to check common command line options that just report and die */
cmd_line = OBJ_NEW(opal_cmd_line_t);
if (OPAL_SUCCESS != opal_cmd_line_create(cmd_line, orte_cmd_line_opts)) {
OBJ_RELEASE(cmd_line);
exit(1);
}
mca_base_cmd_line_setup(cmd_line);
if (ORTE_SUCCESS != (ret = opal_cmd_line_parse(cmd_line, false,
argc, argv))) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
fprintf(stderr, "Usage: %s [OPTION]...\n%s\n", argv[0], args);
free(args);
OBJ_RELEASE(cmd_line);
return ret;
}
/*
* Since this process can now handle MCA/GMCA parameters, make sure to
* process them.
*/
mca_base_cmd_line_process_args(cmd_line, &environ, &environ);
/* Ensure that enough of OPAL is setup for us to be able to run */
/*
* NOTE: (JJH)
* We need to allow 'mca_base_cmd_line_process_args()' to process command
* line arguments *before* calling opal_init_util() since the command
* line could contain MCA parameters that affect the way opal_init_util()
* functions. AMCA parameters are one such option normally received on the
* command line that affect the way opal_init_util() behaves.
* It is "safe" to call mca_base_cmd_line_process_args() before
* opal_init_util() since mca_base_cmd_line_process_args() does *not*
* depend upon opal_init_util() functionality.
*/
if (OPAL_SUCCESS != opal_init_util(&argc, &argv)) {
fprintf(stderr, "OPAL failed to initialize -- orted aborting\n");
exit(1);
}
/* save the environment for launch purposes. This MUST be
* done so that we can pass it to any local procs we
* spawn - otherwise, those local procs won't see any
* non-MCA envars that were set in the enviro when the
* orted was executed - e.g., by .csh
*/
orte_launch_environ = opal_argv_copy(environ);
/* purge any ess flag set in the environ when we were launched */
opal_unsetenv(OPAL_MCA_PREFIX"ess", &orte_launch_environ);
/* if orte_daemon_debug is set, let someone know we are alive right
* away just in case we have a problem along the way
*/
if (orted_globals.debug) {
gethostname(hostname, 100);
fprintf(stderr, "Daemon was launched on %s - beginning to initialize\n", hostname);
}
/* check for help request */
if (orted_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
orte_show_help("help-orted.txt", "orted:usage", false,
argv[0], args);
free(args);
return 1;
}
#if defined(HAVE_SETSID)
/* see if we were directed to separate from current session */
if (orted_globals.set_sid) {
setsid();
}
#endif
/* see if they want us to spin until they can connect a debugger to us */
i=0;
while (orted_spin_flag) {
i++;
if (1000 < i) i=0;
}
#if OPAL_ENABLE_FT_CR == 1
/* Mark as a tool program */
(void) mca_base_var_env_name ("opal_cr_is_tool", &tmp_env_var);
opal_setenv(tmp_env_var,
"1",
true, &environ);
free(tmp_env_var);
#endif
/* if mapreduce set, flag it */
if (orted_globals.mapreduce) {
orte_map_reduce = true;
}
/* detach from controlling terminal
* otherwise, remain attached so output can get to us
*/
if(!orte_debug_flag &&
!orte_debug_daemons_flag &&
orted_globals.daemonize) {
opal_daemon_init(NULL);
}
/* Set the flag telling OpenRTE that I am NOT a
* singleton, but am "infrastructure" - prevents setting
* up incorrect infrastructure that only a singleton would
* require.
*/
if (orted_globals.hnp) {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
ORTE_ERROR_LOG(ret);
return ret;
}
} else {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_DAEMON))) {
ORTE_ERROR_LOG(ret);
return ret;
}
}
/* finalize the OPAL utils. As they are opened again from orte_init->opal_init
* we continue to have a reference count on them. So we have to finalize them twice...
*/
opal_finalize_util();
#if OPAL_HAVE_HWLOC
/* bind ourselves if so directed */
if (NULL != orte_daemon_cores) {
char **cores=NULL, tmp[128];
hwloc_obj_t pu;
hwloc_cpuset_t ours, pucpus, res;
int core;
/* could be a collection of comma-delimited ranges, so
* use our handy utility to parse it
*/
orte_util_parse_range_options(orte_daemon_cores, &cores);
if (NULL != cores) {
ours = hwloc_bitmap_alloc();
hwloc_bitmap_zero(ours);
pucpus = hwloc_bitmap_alloc();
res = hwloc_bitmap_alloc();
for (i=0; NULL != cores[i]; i++) {
core = strtoul(cores[i], NULL, 10);
if (NULL == (pu = opal_hwloc_base_get_pu(opal_hwloc_topology, core, OPAL_HWLOC_LOGICAL))) {
/* turn off the show help forwarding as we won't
* be able to cycle the event library to send
*/
orte_show_help_finalize();
/* the message will now come out locally */
orte_show_help("help-orted.txt", "orted:cannot-bind",
true, orte_process_info.nodename,
orte_daemon_cores);
ret = ORTE_ERR_NOT_SUPPORTED;
goto DONE;
}
hwloc_bitmap_and(pucpus, pu->online_cpuset, pu->allowed_cpuset);
hwloc_bitmap_or(res, ours, pucpus);
hwloc_bitmap_copy(ours, res);
}
/* if the result is all zeros, then don't bind */
if (!hwloc_bitmap_iszero(ours)) {
(void)hwloc_set_cpubind(opal_hwloc_topology, ours, 0);
if (opal_hwloc_report_bindings) {
opal_hwloc_base_cset2mapstr(tmp, sizeof(tmp), opal_hwloc_topology, ours);
opal_output(0, "Daemon %s is bound to cores %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
}
}
/* cleanup */
hwloc_bitmap_free(ours);
hwloc_bitmap_free(pucpus);
hwloc_bitmap_free(res);
opal_argv_free(cores);
}
}
#endif
if ((int)ORTE_VPID_INVALID != orted_globals.fail) {
orted_globals.abort=false;
/* some vpid was ordered to fail. The value can be positive
* or negative, depending upon the desired method for failure,
* so need to check both here
*/
if (0 > orted_globals.fail) {
orted_globals.fail = -1*orted_globals.fail;
orted_globals.abort = true;
}
/* are we the specified vpid? */
if ((int)ORTE_PROC_MY_NAME->vpid == orted_globals.fail) {
/* if the user specified we delay, then setup a timer
* and have it kill us
*/
if (0 < orted_globals.fail_delay) {
ORTE_TIMER_EVENT(orted_globals.fail_delay, 0, shutdown_callback, ORTE_SYS_PRI);
} else {
opal_output(0, "%s is executing clean %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_globals.abort ? "abort" : "abnormal termination");
/* do -not- call finalize as this will send a message to the HNP
* indicating clean termination! Instead, just forcibly cleanup
* the local session_dir tree and exit
*/
orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
/* if we were ordered to abort, do so */
if (orted_globals.abort) {
abort();
}
/* otherwise, return with non-zero status */
ret = ORTE_ERROR_DEFAULT_EXIT_CODE;
goto DONE;
}
}
}
/* insert our contact info into our process_info struct so we
* have it for later use and set the local daemon field to our name
*/
orte_process_info.my_daemon_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_DAEMON->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_DAEMON->vpid = ORTE_PROC_MY_NAME->vpid;
/* if I am also the hnp, then update that contact info field too */
if (ORTE_PROC_IS_HNP) {
orte_process_info.my_hnp_uri = orte_rml.get_contact_info();
ORTE_PROC_MY_HNP->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_HNP->vpid = ORTE_PROC_MY_NAME->vpid;
}
/* setup the primary daemon command receive function */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);
/* output a message indicating we are alive, our name, and our pid
* for debugging purposes
*/
if (orte_debug_daemons_flag) {
fprintf(stderr, "Daemon %s checking in as pid %ld on host %s\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.pid,
orte_process_info.nodename);
}
/* We actually do *not* want the orted to voluntarily yield() the
processor more than necessary. The orted already blocks when
it is doing nothing, so it doesn't use any more CPU cycles than
it should; but when it *is* doing something, we do not want it
to be unnecessarily delayed because it voluntarily yielded the
processor in the middle of its work.
For example: when a message arrives at the orted, we want the
OS to wake up the orted in a timely fashion (which most OS's
seem good about doing) and then we want the orted to process
the message as fast as possible. If the orted yields and lets
aggressive MPI applications get the processor back, it may be a
long time before the OS schedules the orted to run again
(particularly if there is no IO event to wake it up). Hence,
routed OOB messages (for example) may be significantly delayed
before being delivered to MPI processes, which can be
problematic in some scenarios (e.g., COMM_SPAWN, BTL's that
require OOB messages for wireup, etc.). */
opal_progress_set_yield_when_idle(false);
/* Change the default behavior of libevent such that we want to
continually block rather than blocking for the default timeout
and then looping around the progress engine again. There
should be nothing in the orted that cannot block in libevent
until "something" happens (i.e., there's no need to keep
cycling through progress because the only things that should
happen will happen in libevent). This is a minor optimization,
but what the heck... :-) */
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE);
/* if requested, report my uri to the indicated pipe */
if (orted_globals.uri_pipe > 0) {
orte_job_t *jdata;
orte_proc_t *proc;
orte_node_t *node;
orte_app_context_t *app;
char *tmp, *nptr, *sysinfo;
int32_t ljob;
/* setup the singleton's job */
jdata = OBJ_NEW(orte_job_t);
orte_plm_base_create_jobid(jdata);
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* must create a map for it (even though it has no
* info in it) so that the job info will be picked
* up in subsequent pidmaps or other daemons won't
* know how to route
*/
jdata->map = OBJ_NEW(orte_job_map_t);
/* setup an app_context for the singleton */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup("singleton");
app->num_procs = 1;
opal_pointer_array_add(jdata->apps, app);
/* setup a proc object for the singleton - since we
* -must- be the HNP, and therefore we stored our
* node on the global node pool, and since the singleton
* -must- be on the same node as us, indicate that
*/
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = jdata->jobid;
proc->name.vpid = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_ALIVE);
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
/* obviously, it is on my node */
node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0);
proc->node = node;
OBJ_RETAIN(node); /* keep accounting straight */
opal_pointer_array_add(jdata->procs, proc);
jdata->num_procs = 1;
/* and it obviously is on the node */
OBJ_RETAIN(proc);
opal_pointer_array_add(node->procs, proc);
node->num_procs++;
/* and obviously it is one of my local procs */
OBJ_RETAIN(proc);
opal_pointer_array_add(orte_local_children, proc);
jdata->num_local_procs = 1;
/* set the trivial */
proc->local_rank = 0;
proc->node_rank = 0;
proc->app_rank = 0;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_LOCAL);
/* create a string that contains our uri + sysinfo + PMIx server URI */
orte_util_convert_sysinfo_to_string(&sysinfo, orte_local_cpu_type, orte_local_cpu_model);
asprintf(&tmp, "%s[%s]%s", orte_process_info.my_daemon_uri, sysinfo, pmix_server_uri);
free(sysinfo);
/* pass that info to the singleton */
write(orted_globals.uri_pipe, tmp, strlen(tmp)+1); /* need to add 1 to get the NULL */
/* cleanup */
free(tmp);
/* since a singleton spawned us, we need to harvest
* any MCA params from the local environment so
* we can pass them along to any subsequent daemons
* we may start as the result of a comm_spawn
*/
for (i=0; NULL != environ[i]; i++) {
if (0 == strncmp(environ[i], OPAL_MCA_PREFIX, 9)) {
/* make a copy to manipulate */
tmp = strdup(environ[i]);
/* find the equal sign */
nptr = strchr(tmp, '=');
*nptr = '\0';
nptr++;
/* add the mca param to the orted cmd line */
opal_argv_append_nosize(&orted_cmd_line, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&orted_cmd_line, &tmp[9]);
opal_argv_append_nosize(&orted_cmd_line, nptr);
free(tmp);
}
}
}
/* if we were given a pipe to monitor for singleton termination, set that up */
if (orted_globals.singleton_died_pipe > 0) {
/* register shutdown handler */
pipe_handler = (opal_event_t*)malloc(sizeof(opal_event_t));
opal_event_set(orte_event_base, pipe_handler,
orted_globals.singleton_died_pipe,
OPAL_EV_READ,
pipe_closed,
pipe_handler);
opal_event_add(pipe_handler, NULL);
}
/* If I have a parent, then save his contact info so
* any messages we send can flow thru him.
*/
orte_parent_uri = NULL;
(void) mca_base_var_register ("orte", "orte", NULL, "parent_uri",
"URI for the parent if tree launch is enabled.",
MCA_BASE_VAR_TYPE_STRING, NULL, 0,
MCA_BASE_VAR_FLAG_INTERNAL,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_CONSTANT,
&orte_parent_uri);
if (NULL != orte_parent_uri) {
orte_process_name_t parent;
/* set the contact info into the hash table */
orte_rml.set_contact_info(orte_parent_uri);
ret = orte_rml_base_parse_uris(orte_parent_uri, &parent, NULL);
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
free (orte_parent_uri);
orte_parent_uri = NULL;
goto DONE;
}
/* don't need this value anymore */
free(orte_parent_uri);
orte_parent_uri = NULL;
/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(ORTE_PROC_MY_HNP, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(&parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* if we are not the HNP...the only time we will be an HNP
* is if we are launched by a singleton to provide support
* for it
*/
if (!ORTE_PROC_IS_HNP) {
/* send the information to the orted report-back point - this function
* will process the data, but also counts the number of
* orteds that reported back so the launch procedure can continue.
* We need to do this at the last possible second as the HNP
* can turn right around and begin issuing orders to us
*/
buffer = OBJ_NEW(opal_buffer_t);
/* insert our name for rollup purposes */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* for now, always include our contact info, even if we are using
* static ports. Eventually, this will be removed
*/
rml_uri = orte_rml.get_contact_info();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &rml_uri, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* include our node name */
opal_dss.pack(buffer, &orte_process_info.nodename, 1, OPAL_STRING);
/* if requested, include any non-loopback aliases for this node */
if (orte_retain_aliases) {
char **aliases=NULL;
uint8_t naliases, ni;
char hostname[ORTE_MAX_HOSTNAME_SIZE];
/* if we stripped the prefix or removed the fqdn,
* include full hostname as an alias
*/
gethostname(hostname, ORTE_MAX_HOSTNAME_SIZE);
if (strlen(orte_process_info.nodename) < strlen(hostname)) {
opal_argv_append_nosize(&aliases, hostname);
}
opal_ifgetaliases(&aliases);
naliases = opal_argv_count(aliases);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &naliases, 1, OPAL_UINT8))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
for (ni=0; ni < naliases; ni++) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &aliases[ni], 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
opal_argv_free(aliases);
}
#if OPAL_HAVE_HWLOC
{
char *coprocessors;
/* add the local topology */
if (NULL != opal_hwloc_topology &&
(1 == ORTE_PROC_MY_NAME->vpid || orte_hetero_nodes)) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &opal_hwloc_topology, 1, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(ret);
}
}
/* detect and add any coprocessors */
coprocessors = opal_hwloc_base_find_coprocessors(opal_hwloc_topology);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
/* see if I am on a coprocessor */
coprocessors = opal_hwloc_base_check_on_coprocessor();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
}
#endif
/* send to the HNP's callback - will be routed if routes are available */
if (0 > (ret = orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, buffer,
ORTE_RML_TAG_ORTED_CALLBACK,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
/* if we are tree-spawning, then we need to capture the MCA params
* from our cmd line so we can pass them along to the daemons we spawn -
* otherwise, only the first layer of daemons will ever see them
*/
if (orted_globals.tree_spawn) {
int j, k;
bool ignore;
char *no_keep[] = {
"orte_hnp_uri",
"orte_ess_jobid",
"orte_ess_vpid",
"orte_ess_num_procs",
"orte_parent_uri",
"mca_base_env_list",
NULL
};
for (i=0; i < argc; i++) {
if (0 == strcmp("-"OPAL_MCA_CMD_LINE_ID, argv[i]) ||
0 == strcmp("--"OPAL_MCA_CMD_LINE_ID, argv[i]) ) {
ignore = false;
/* see if this is something we cannot pass along */
for (k=0; NULL != no_keep[k]; k++) {
if (0 == strcmp(no_keep[k], argv[i+1])) {
ignore = true;
break;
}
}
if (!ignore) {
/* see if this is already present so we at least can
* avoid growing the cmd line with duplicates
*/
if (NULL != orted_cmd_line) {
for (j=0; NULL != orted_cmd_line[j]; j++) {
if (0 == strcmp(argv[i+1], orted_cmd_line[j])) {
/* already here - ignore it */
ignore = true;
break;
}
}
}
if (!ignore) {
opal_argv_append_nosize(&orted_cmd_line, argv[i]);
opal_argv_append_nosize(&orted_cmd_line, argv[i+1]);
opal_argv_append_nosize(&orted_cmd_line, argv[i+2]);
}
}
i += 2;
}
}
}
if (orte_debug_daemons_flag) {
opal_output(0, "%s orted: up and running - waiting for commands!", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
}
ret = ORTE_SUCCESS;
/* loop the event lib until an exit event is detected */
while (orte_event_base_active) {
opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
}
/* ensure all local procs are dead */
orte_odls.kill_local_procs(NULL);
DONE:
/* update the exit status, in case it wasn't done */
ORTE_UPDATE_EXIT_STATUS(ret);
/* cleanup and leave */
orte_finalize();
if (orte_debug_flag) {
fprintf(stderr, "exiting with status %d\n", orte_exit_status);
}
exit(orte_exit_status);
}
static void set(orte_job_t *jobdat,
orte_proc_t *child,
char ***environ_copy,
int write_fd)
{
hwloc_cpuset_t cpuset;
hwloc_obj_t root;
opal_hwloc_topo_data_t *sum;
orte_app_context_t *context;
int rc=ORTE_ERROR;
char *msg, *param;
char *cpu_bitmap;
opal_output_verbose(2, orte_rtc_base_framework.framework_output,
"%s hwloc:set on child %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == child) ? "NULL" : ORTE_NAME_PRINT(&child->name));
if (NULL == jobdat || NULL == child) {
/* nothing for us to do */
opal_output_verbose(2, orte_rtc_base_framework.framework_output,
"%s hwloc:set jobdat %s child %s - nothing to do",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == jobdat) ? "NULL" : ORTE_JOBID_PRINT(jobdat->jobid),
(NULL == child) ? "NULL" : ORTE_NAME_PRINT(&child->name));
return;
}
context = (orte_app_context_t*)opal_pointer_array_get_item(jobdat->apps, child->app_idx);
/* Set process affinity, if given */
cpu_bitmap = NULL;
if (!orte_get_attribute(&child->attributes, ORTE_PROC_CPU_BITMAP, (void**)&cpu_bitmap, OPAL_STRING) ||
NULL == cpu_bitmap || 0 == strlen(cpu_bitmap)) {
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
rc = hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
char *tmp;
(void)hwloc_bitmap_list_asprintf(&tmp, sum->available);
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), tmp);
free(tmp);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
return;
}
}
}
if (0 == rc && opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, environ_copy);
free(param);
}
} else {
/* convert the list to a cpuset */
cpuset = hwloc_bitmap_alloc();
if (0 != (rc = hwloc_bitmap_list_sscanf(cpuset, cpu_bitmap))) {
/* See comment above about "This may be a small memory leak" */
asprintf(&msg, "hwloc_bitmap_sscanf returned \"%s\" for the string \"%s\"",
opal_strerror(rc), cpu_bitmap);
if (NULL == msg) {
msg = "failed to convert bitmap list to hwloc bitmap";
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding) &&
OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
/* If binding is required and a binding directive was explicitly
* given (i.e., we are not binding due to a default policy),
* send an error up the pipe (which exits -- it doesn't return).
*/
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename,
context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
free(cpu_bitmap);
return;
}
}
/* bind as specified */
rc = hwloc_set_cpubind(opal_hwloc_topology, cpuset, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
char *tmp = NULL;
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), cpu_bitmap);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
if (NULL != tmp) {
free(tmp);
free(msg);
}
return;
}
if (NULL != tmp) {
free(tmp);
free(msg);
}
}
if (0 == rc && opal_hwloc_report_bindings) {
char tmp1[1024], tmp2[1024];
hwloc_cpuset_t mycpus;
/* get the cpus we are bound to */
mycpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology,
mycpus,
HWLOC_CPUBIND_PROCESS) < 0) {
opal_output(0, "MCW rank %d is not bound",
child->name.vpid);
} else {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus);
opal_output(0, "MCW rank %d bound to %s: %s",
child->name.vpid, tmp1, tmp2);
}
}
hwloc_bitmap_free(mycpus);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, environ_copy);
free(param);
}
/* set memory affinity policy - if we get an error, don't report
* anything unless the user actually specified the binding policy
*/
rc = opal_hwloc_base_set_process_membind_policy();
if (ORTE_SUCCESS != rc && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates memory binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates memory binding cannot be enforced";
} else {
msg = "failed to bind memory";
}
if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
orte_rtc_base_send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"memory binding error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
orte_rtc_base_send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "memory not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
free(cpu_bitmap);
return;
}
}
}
if (NULL != cpu_bitmap) {
free(cpu_bitmap);
}
}
int orte_daemon(int argc, char *argv[])
{
int ret = 0;
opal_cmd_line_t *cmd_line = NULL;
int i;
opal_buffer_t *buffer;
char hostname[OPAL_MAXHOSTNAMELEN];
#if OPAL_ENABLE_FT_CR == 1
char *tmp_env_var = NULL;
#endif
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
/* initialize the singleton died pipe to an illegal value so we can detect it was set */
orted_globals.singleton_died_pipe = -1;
bucket = OBJ_NEW(opal_buffer_t);
/* setup to check common command line options that just report and die */
cmd_line = OBJ_NEW(opal_cmd_line_t);
if (OPAL_SUCCESS != opal_cmd_line_create(cmd_line, orte_cmd_line_opts)) {
OBJ_RELEASE(cmd_line);
exit(1);
}
mca_base_cmd_line_setup(cmd_line);
if (ORTE_SUCCESS != (ret = opal_cmd_line_parse(cmd_line, false, false,
argc, argv))) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
fprintf(stderr, "Usage: %s [OPTION]...\n%s\n", argv[0], args);
free(args);
OBJ_RELEASE(cmd_line);
return ret;
}
/*
* Since this process can now handle MCA/GMCA parameters, make sure to
* process them.
*/
mca_base_cmd_line_process_args(cmd_line, &environ, &environ);
/* Ensure that enough of OPAL is setup for us to be able to run */
/*
* NOTE: (JJH)
* We need to allow 'mca_base_cmd_line_process_args()' to process command
* line arguments *before* calling opal_init_util() since the command
* line could contain MCA parameters that affect the way opal_init_util()
* functions. AMCA parameters are one such option normally received on the
* command line that affect the way opal_init_util() behaves.
* It is "safe" to call mca_base_cmd_line_process_args() before
* opal_init_util() since mca_base_cmd_line_process_args() does *not*
* depend upon opal_init_util() functionality.
*/
if (OPAL_SUCCESS != opal_init_util(&argc, &argv)) {
fprintf(stderr, "OPAL failed to initialize -- orted aborting\n");
exit(1);
}
/* save the environment for launch purposes. This MUST be
* done so that we can pass it to any local procs we
* spawn - otherwise, those local procs won't see any
* non-MCA envars that were set in the enviro when the
* orted was executed - e.g., by .csh
*/
orte_launch_environ = opal_argv_copy(environ);
/* purge any ess/pmix flags set in the environ when we were launched */
opal_unsetenv(OPAL_MCA_PREFIX"ess", &orte_launch_environ);
opal_unsetenv(OPAL_MCA_PREFIX"pmix", &orte_launch_environ);
/* if orte_daemon_debug is set, let someone know we are alive right
* away just in case we have a problem along the way
*/
if (orted_globals.debug) {
gethostname(hostname, sizeof(hostname));
fprintf(stderr, "Daemon was launched on %s - beginning to initialize\n", hostname);
}
/* check for help request */
if (orted_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(cmd_line);
orte_show_help("help-orted.txt", "orted:usage", false,
argv[0], args);
free(args);
return 1;
}
#if defined(HAVE_SETSID)
/* see if we were directed to separate from current session */
if (orted_globals.set_sid) {
setsid();
}
#endif
/* see if they want us to spin until they can connect a debugger to us */
i=0;
while (orted_spin_flag) {
i++;
if (1000 < i) i=0;
}
#if OPAL_ENABLE_FT_CR == 1
/* Mark as a tool program */
(void) mca_base_var_env_name ("opal_cr_is_tool", &tmp_env_var);
opal_setenv(tmp_env_var,
"1",
true, &environ);
free(tmp_env_var);
#endif
/* detach from controlling terminal
* otherwise, remain attached so output can get to us
*/
if(!orte_debug_flag &&
!orte_debug_daemons_flag &&
orted_globals.daemonize) {
opal_daemon_init(NULL);
}
/* Set the flag telling OpenRTE that I am NOT a
* singleton, but am "infrastructure" - prevents setting
* up incorrect infrastructure that only a singleton would
* require.
*/
if (orted_globals.hnp) {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
ORTE_ERROR_LOG(ret);
return ret;
}
} else {
if (ORTE_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_DAEMON))) {
ORTE_ERROR_LOG(ret);
return ret;
}
}
/* finalize the OPAL utils. As they are opened again from orte_init->opal_init
* we continue to have a reference count on them. So we have to finalize them twice...
*/
opal_finalize_util();
/* bind ourselves if so directed */
if (NULL != orte_daemon_cores) {
char **cores=NULL, tmp[128];
hwloc_obj_t pu;
hwloc_cpuset_t ours, res;
int core;
/* could be a collection of comma-delimited ranges, so
* use our handy utility to parse it
*/
orte_util_parse_range_options(orte_daemon_cores, &cores);
if (NULL != cores) {
ours = hwloc_bitmap_alloc();
hwloc_bitmap_zero(ours);
res = hwloc_bitmap_alloc();
for (i=0; NULL != cores[i]; i++) {
core = strtoul(cores[i], NULL, 10);
if (NULL == (pu = opal_hwloc_base_get_pu(opal_hwloc_topology, core, OPAL_HWLOC_LOGICAL))) {
/* turn off the show help forwarding as we won't
* be able to cycle the event library to send
*/
orte_show_help_finalize();
/* the message will now come out locally */
orte_show_help("help-orted.txt", "orted:cannot-bind",
true, orte_process_info.nodename,
orte_daemon_cores);
ret = ORTE_ERR_NOT_SUPPORTED;
hwloc_bitmap_free(ours);
hwloc_bitmap_free(res);
goto DONE;
}
hwloc_bitmap_or(res, ours, pu->cpuset);
hwloc_bitmap_copy(ours, res);
}
/* if the result is all zeros, then don't bind */
if (!hwloc_bitmap_iszero(ours)) {
(void)hwloc_set_cpubind(opal_hwloc_topology, ours, 0);
if (opal_hwloc_report_bindings) {
opal_hwloc_base_cset2mapstr(tmp, sizeof(tmp), opal_hwloc_topology, ours);
opal_output(0, "Daemon %s is bound to cores %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
}
}
/* cleanup */
hwloc_bitmap_free(ours);
hwloc_bitmap_free(res);
opal_argv_free(cores);
}
}
if ((int)ORTE_VPID_INVALID != orted_debug_failure) {
orted_globals.abort=false;
/* some vpid was ordered to fail. The value can be positive
* or negative, depending upon the desired method for failure,
* so need to check both here
*/
if (0 > orted_debug_failure) {
orted_debug_failure = -1*orted_debug_failure;
orted_globals.abort = true;
}
/* are we the specified vpid? */
if ((int)ORTE_PROC_MY_NAME->vpid == orted_debug_failure) {
/* if the user specified we delay, then setup a timer
* and have it kill us
*/
if (0 < orted_debug_failure_delay) {
ORTE_TIMER_EVENT(orted_debug_failure_delay, 0, shutdown_callback, ORTE_SYS_PRI);
} else {
opal_output(0, "%s is executing clean %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_globals.abort ? "abort" : "abnormal termination");
/* do -not- call finalize as this will send a message to the HNP
* indicating clean termination! Instead, just forcibly cleanup
* the local session_dir tree and exit
*/
orte_session_dir_cleanup(ORTE_JOBID_WILDCARD);
/* if we were ordered to abort, do so */
if (orted_globals.abort) {
abort();
}
/* otherwise, return with non-zero status */
ret = ORTE_ERROR_DEFAULT_EXIT_CODE;
goto DONE;
}
}
}
/* insert our contact info into our process_info struct so we
* have it for later use and set the local daemon field to our name
*/
orte_oob_base_get_addr(&orte_process_info.my_daemon_uri);
if (NULL == orte_process_info.my_daemon_uri) {
/* no way to communicate */
ret = ORTE_ERROR;
goto DONE;
}
ORTE_PROC_MY_DAEMON->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_DAEMON->vpid = ORTE_PROC_MY_NAME->vpid;
/* if I am also the hnp, then update that contact info field too */
if (ORTE_PROC_IS_HNP) {
orte_process_info.my_hnp_uri = strdup(orte_process_info.my_daemon_uri);
ORTE_PROC_MY_HNP->jobid = ORTE_PROC_MY_NAME->jobid;
ORTE_PROC_MY_HNP->vpid = ORTE_PROC_MY_NAME->vpid;
}
/* setup the primary daemon command receive function */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);
/* output a message indicating we are alive, our name, and our pid
* for debugging purposes
*/
if (orte_debug_daemons_flag) {
fprintf(stderr, "Daemon %s checking in as pid %ld on host %s\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (long)orte_process_info.pid,
orte_process_info.nodename);
}
/* We actually do *not* want the orted to voluntarily yield() the
processor more than necessary. The orted already blocks when
it is doing nothing, so it doesn't use any more CPU cycles than
it should; but when it *is* doing something, we do not want it
to be unnecessarily delayed because it voluntarily yielded the
processor in the middle of its work.
For example: when a message arrives at the orted, we want the
OS to wake up the orted in a timely fashion (which most OS's
seem good about doing) and then we want the orted to process
the message as fast as possible. If the orted yields and lets
aggressive MPI applications get the processor back, it may be a
long time before the OS schedules the orted to run again
(particularly if there is no IO event to wake it up). Hence,
routed OOB messages (for example) may be significantly delayed
before being delivered to MPI processes, which can be
problematic in some scenarios (e.g., COMM_SPAWN, BTL's that
require OOB messages for wireup, etc.). */
opal_progress_set_yield_when_idle(false);
/* Change the default behavior of libevent such that we want to
continually block rather than blocking for the default timeout
and then looping around the progress engine again. There
should be nothing in the orted that cannot block in libevent
until "something" happens (i.e., there's no need to keep
cycling through progress because the only things that should
happen will happen in libevent). This is a minor optimization,
but what the heck... :-) */
opal_progress_set_event_flag(OPAL_EVLOOP_ONCE);
/* if requested, report my uri to the indicated pipe */
if (orted_globals.uri_pipe > 0) {
orte_job_t *jdata;
orte_proc_t *proc;
orte_node_t *node;
orte_app_context_t *app;
char *tmp, *nptr, *sysinfo;
char **singenv=NULL, *string_key, *env_str;
/* setup the singleton's job */
jdata = OBJ_NEW(orte_job_t);
/* default to ompi for now */
opal_argv_append_nosize(&jdata->personality, "ompi");
orte_plm_base_create_jobid(jdata);
opal_hash_table_set_value_uint32(orte_job_data, jdata->jobid, jdata);
/* must create a map for it (even though it has no
* info in it) so that the job info will be picked
* up in subsequent pidmaps or other daemons won't
* know how to route
*/
jdata->map = OBJ_NEW(orte_job_map_t);
/* setup an app_context for the singleton */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup("singleton");
app->num_procs = 1;
opal_pointer_array_add(jdata->apps, app);
jdata->num_apps = 1;
/* setup a proc object for the singleton - since we
* -must- be the HNP, and therefore we stored our
* node on the global node pool, and since the singleton
* -must- be on the same node as us, indicate that
*/
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = jdata->jobid;
proc->name.vpid = 0;
proc->parent = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_ALIVE);
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
/* obviously, it is on my node */
node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0);
proc->node = node;
OBJ_RETAIN(node); /* keep accounting straight */
opal_pointer_array_add(jdata->procs, proc);
jdata->num_procs = 1;
/* add the node to the job map */
OBJ_RETAIN(node);
opal_pointer_array_add(jdata->map->nodes, node);
jdata->map->num_nodes++;
/* and it obviously is on the node */
OBJ_RETAIN(proc);
opal_pointer_array_add(node->procs, proc);
node->num_procs++;
/* and obviously it is one of my local procs */
OBJ_RETAIN(proc);
opal_pointer_array_add(orte_local_children, proc);
jdata->num_local_procs = 1;
/* set the trivial */
proc->local_rank = 0;
proc->node_rank = 0;
proc->app_rank = 0;
proc->state = ORTE_PROC_STATE_RUNNING;
proc->app_idx = 0;
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_LOCAL);
/* set the ORTE_JOB_TRANSPORT_KEY from the environment */
orte_pre_condition_transports(jdata, NULL);
/* register the singleton's nspace with our PMIx server */
if (ORTE_SUCCESS != (ret = orte_pmix_server_register_nspace(jdata, false))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* use setup fork to create the envars needed by the singleton */
if (OPAL_SUCCESS != (ret = opal_pmix.server_setup_fork(&proc->name, &singenv))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* append the transport key to the envars needed by the singleton */
if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_TRANSPORT_KEY, (void**)&string_key, OPAL_STRING) || NULL == string_key) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto DONE;
}
asprintf(&env_str, OPAL_MCA_PREFIX"orte_precondition_transports=%s", string_key);
opal_argv_append_nosize(&singenv, env_str);
free(env_str);
nptr = opal_argv_join(singenv, '*');
opal_argv_free(singenv);
/* create a string that contains our uri + sysinfo + PMIx server URI envars */
orte_util_convert_sysinfo_to_string(&sysinfo, orte_local_cpu_type, orte_local_cpu_model);
asprintf(&tmp, "%s[%s]%s", orte_process_info.my_daemon_uri, sysinfo, nptr);
free(sysinfo);
free(nptr);
/* pass that info to the singleton */
if (OPAL_SUCCESS != (ret = opal_fd_write(orted_globals.uri_pipe, strlen(tmp)+1, tmp))) { ; /* need to add 1 to get the NULL */
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* cleanup */
free(tmp);
close(orted_globals.uri_pipe);
/* since a singleton spawned us, we need to harvest
* any MCA params from the local environment so
* we can pass them along to any subsequent daemons
* we may start as the result of a comm_spawn
*/
for (i=0; NULL != environ[i]; i++) {
if (0 == strncmp(environ[i], OPAL_MCA_PREFIX, 9)) {
/* make a copy to manipulate */
tmp = strdup(environ[i]);
/* find the equal sign */
nptr = strchr(tmp, '=');
*nptr = '\0';
nptr++;
/* add the mca param to the orted cmd line */
opal_argv_append_nosize(&orted_cmd_line, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&orted_cmd_line, &tmp[9]);
opal_argv_append_nosize(&orted_cmd_line, nptr);
free(tmp);
}
}
}
/* if we were given a pipe to monitor for singleton termination, set that up */
if (orted_globals.singleton_died_pipe > 0) {
/* register shutdown handler */
pipe_handler = (opal_event_t*)malloc(sizeof(opal_event_t));
opal_event_set(orte_event_base, pipe_handler,
orted_globals.singleton_died_pipe,
OPAL_EV_READ,
pipe_closed,
pipe_handler);
opal_event_add(pipe_handler, NULL);
}
/* If I have a parent, then save his contact info so
* any messages we send can flow thru him.
*/
orte_parent_uri = NULL;
(void) mca_base_var_register ("orte", "orte", NULL, "parent_uri",
"URI for the parent if tree launch is enabled.",
MCA_BASE_VAR_TYPE_STRING, NULL, 0,
MCA_BASE_VAR_FLAG_INTERNAL,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_CONSTANT,
&orte_parent_uri);
if (NULL != orte_parent_uri) {
orte_process_name_t parent;
opal_value_t val;
/* set the contact info into our local database */
ret = orte_rml_base_parse_uris(orte_parent_uri, &parent, NULL);
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
free (orte_parent_uri);
orte_parent_uri = NULL;
goto DONE;
}
OBJ_CONSTRUCT(&val, opal_value_t);
val.key = OPAL_PMIX_PROC_URI;
val.type = OPAL_STRING;
val.data.string = orte_parent_uri;
if (OPAL_SUCCESS != (ret = opal_pmix.store_local(&parent, &val))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&val);
goto DONE;
}
val.key = NULL;
val.data.string = NULL;
OBJ_DESTRUCT(&val);
/* don't need this value anymore */
free(orte_parent_uri);
orte_parent_uri = NULL;
/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(NULL, ORTE_PROC_MY_HNP, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(NULL, &parent))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
}
/* if we are not the HNP...the only time we will be an HNP
* is if we are launched by a singleton to provide support
* for it
*/
if (!ORTE_PROC_IS_HNP) {
orte_process_name_t target;
target.jobid = ORTE_PROC_MY_NAME->jobid;
if (orte_fwd_mpirun_port || orte_static_ports) {
/* setup the rollup callback */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK,
ORTE_RML_PERSISTENT, rollup, NULL);
target.vpid = ORTE_PROC_MY_NAME->vpid;
/* since we will be waiting for any children to send us
* their rollup info before sending to our parent, save
* a little time in the launch phase by "warming up" the
* connection to our parent while we wait for our children */
buffer = OBJ_NEW(opal_buffer_t); // zero-byte message
if (0 > (ret = orte_rml.send_buffer_nb(orte_mgmt_conduit,
ORTE_PROC_MY_PARENT, buffer,
ORTE_RML_TAG_WARMUP_CONNECTION,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
} else {
target.vpid = 0;
}
/* send the information to the orted report-back point - this function
* will process the data, but also counts the number of
* orteds that reported back so the launch procedure can continue.
* We need to do this at the last possible second as the HNP
* can turn right around and begin issuing orders to us
*/
buffer = OBJ_NEW(opal_buffer_t);
/* insert our name for rollup purposes */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
/* get any connection info we may have pushed */
{
opal_value_t *val = NULL, *kv;
opal_list_t *modex;
int32_t flag;
if (OPAL_SUCCESS != (ret = opal_pmix.get(ORTE_PROC_MY_NAME, NULL, NULL, &val)) || NULL == val) {
/* just pack a marker indicating we don't have any to share */
flag = 0;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
} else {
/* the data is returned as a list of key-value pairs in the opal_value_t */
if (OPAL_PTR == val->type) {
modex = (opal_list_t*)val->data.ptr;
flag = (int32_t)opal_list_get_size(modex);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
OPAL_LIST_FOREACH(kv, modex, opal_value_t) {
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &kv, 1, OPAL_VALUE))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
OPAL_LIST_RELEASE(modex);
} else {
opal_output(0, "VAL KEY: %s", (NULL == val->key) ? "NULL" : val->key);
/* single value */
flag = 1;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &val, 1, OPAL_VALUE))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(buffer);
goto DONE;
}
}
OBJ_RELEASE(val);
}
/*
* setup env and argv for specified process
*/
static int setup_proc_env_and_argv(orte_job_t* jdata, orte_app_context_t* app,
orte_proc_t* proc, char ***pargv, char ***penv)
{
char* param;
char* param2;
char* value;
char* vp_id_str;
char* job_id_str;
int rc;
int i, num_nodes;
/* obtain app->argv */
if (!(app->argv)) {
opal_output(0, "%s plm::yarn::setup_proc_env_and_argv: app->argv is null",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERROR;
}
*pargv = opal_argv_copy(app->argv);
if (ORTE_SUCCESS != orte_util_convert_jobid_to_string(&job_id_str, jdata->jobid)) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
if (ORTE_SUCCESS != orte_util_convert_vpid_to_string(&vp_id_str, proc->name.vpid)) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
// add stdout, stderr to app
opal_argv_append_nosize(pargv, "1><LOG_DIR>/stdout");
opal_argv_append_nosize(pargv, "2><LOG_DIR>/stderr");
// add java executor to app
opal_argv_prepend_nosize(pargv, vp_id_str);
opal_argv_prepend_nosize(pargv, job_id_str);
opal_argv_prepend_nosize(pargv, "com.pivotal.hamster.yarnexecutor.YarnExecutor");
opal_argv_prepend_nosize(pargv, "hamster-core.jar");
opal_argv_prepend_nosize(pargv, "-cp");
opal_argv_prepend_nosize(pargv, getenv("HAMSTER_JAVA_OPT")==NULL ? "-Xmx32M -Xms8M" : getenv("HAMSTER_JAVA_OPT"));
opal_argv_prepend_nosize(pargv, "$JAVA_HOME/bin/java");
/* obtain app->env */
*penv = opal_environ_merge(environ, app->env);
if (!proc->node) {
opal_output(0, "%s plm::yarn::setup_proc_env_and_argv: node of proc[%d] is NULL",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), proc->name.vpid);
return ORTE_ERROR;
}
if (!proc->node->daemon) {
opal_output(0, "%s plm::yarn::setup_proc_env_and_argv: daemon of node[%s] is NULL",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), proc->node->name);
return ORTE_ERROR;
}
/* set the app_context number into the environment */ //??
param = mca_base_param_env_var("orte_app_num");
asprintf(¶m2, "%ld", (long)app->idx);
opal_setenv(param, param2, true, penv);
free(param);
free(param2);
// pass the daemon's name
param = mca_base_param_env_var("orte_local_daemon_uri");
opal_setenv(param, proc->node->daemon->rml_uri, true, penv);
free(param);
/* pass my contact info */
param = mca_base_param_env_var("orte_hnp_uri");
opal_setenv(param, orte_process_info.my_hnp_uri, true, penv);
free(param);
/* pass the jobid */
param = mca_base_param_env_var("orte_ess_jobid");
opal_setenv(param, job_id_str, true, penv);
free(param);
free(job_id_str);
/* pass the rank */
param = mca_base_param_env_var("orte_ess_vpid");
opal_setenv(param, vp_id_str, true, penv);
free(param);
opal_setenv("OMPI_COMM_WORLD_RANK", vp_id_str, true, penv);
free(vp_id_str); /* done with this now */
/* pass local rank */
asprintf(&value, "%lu", (unsigned long) proc->local_rank);
opal_setenv("OMPI_COMM_WORLD_LOCAL_RANK", value, true, penv);
free(value);
/* pass node rank */
asprintf(&value, "%lu", (unsigned long) proc->node_rank);
opal_setenv("OMPI_COMM_WORLD_NODE_RANK", value, true, penv);
/* set an mca param for it too */
param = mca_base_param_env_var("orte_ess_node_rank");
opal_setenv(param, value, true, penv);
free(param);
free(value);
/* pass a param telling the child what model of cpu we are on,
* if we know it
*/
if (NULL != orte_local_cpu_type) {
param = mca_base_param_env_var("orte_cpu_type");
/* do not overwrite what the user may have provided */
opal_setenv(param, orte_local_cpu_type, false, penv);
free(param);
}
if (NULL != orte_local_cpu_model) {
param = mca_base_param_env_var("orte_cpu_model");
/* do not overwrite what the user may have provided */
opal_setenv(param, orte_local_cpu_model, false, penv);
free(param);
}
/* pass the number of nodes involved in this job */
param = mca_base_param_env_var("orte_num_nodes");
/* we have to count the number of nodes as the size of orte_node_pool
* is only guaranteed to be equal or larger than that number - i.e.,
* the pointer_array increases the size by a block each time, so some
* of the locations are left empty
*/
num_nodes = 0;
for (i = 0; i < orte_node_pool->size; i++) {
if (NULL != opal_pointer_array_get_item(orte_node_pool, i)) {
num_nodes++;
}
}
asprintf(&value, "%d", num_nodes);
opal_setenv(param, value, true, penv);
free(param);
free(value);
/* setup yield schedule */
param = mca_base_param_env_var("mpi_yield_when_idle");
opal_setenv(param, "0", false, penv);
free(param);
/* set MPI universe envar */
orte_ess_env_put(jdata->num_procs, proc->node->num_procs, penv);
asprintf(&value, "%ld", (long) jdata->num_procs);
opal_setenv("OMPI_UNIVERSE_SIZE", value, true, penv);
free(value);
/* pass collective ids for the std MPI operations */
param = mca_base_param_env_var("orte_peer_modex_id");
asprintf(&value, "%d", jdata->peer_modex);
opal_setenv(param, value, true, penv);
free(param);
free(value);
param = mca_base_param_env_var("orte_peer_init_barrier_id");
asprintf(&value, "%d", jdata->peer_init_barrier);
opal_setenv(param, value, true, penv);
free(param);
free(value);
param = mca_base_param_env_var("orte_peer_fini_barrier_id");
asprintf(&value, "%d", jdata->peer_fini_barrier);
opal_setenv(param, value, true, penv);
free(param);
free(value);
/* finally, we will set/unset some mca param to select modules */
opal_unsetenv("OMPI_MCA_plm", penv);
opal_unsetenv("OMPI_MCA_ras", penv);
opal_unsetenv("OMPI_MCA_ess", penv);
opal_unsetenv("OMPI_MCA_state", penv);
opal_unsetenv("OMPI_MCA_errmgr", penv);
return 0;
}
static int initialize(int argc, char *argv[]) {
int ret, exit_status = ORTE_SUCCESS;
char * tmp_env_var = NULL;
/*
* Make sure to init util before parse_args
* to ensure installdirs is setup properly
* before calling mca_base_open();
*/
if( ORTE_SUCCESS != (ret = opal_init_util(&argc, &argv)) ) {
return ret;
}
/*
* Parse command line arguments
*/
if (ORTE_SUCCESS != (ret = parse_args(argc, argv))) {
exit_status = ret;
goto cleanup;
}
/*
* Setup OPAL Output handle from the verbose argument
*/
if( orte_restart_globals.verbose ) {
orte_restart_globals.output = opal_output_open(NULL);
opal_output_set_verbosity(orte_restart_globals.output, 10);
} else {
orte_restart_globals.output = 0; /* Default=STDERR */
}
/* Disable the checkpoint notification routine for this
* tool. As we will never need to checkpoint this tool.
* Note: This must happen before opal_init().
*/
opal_cr_set_enabled(false);
/* Select the none component, since we don't actually use a checkpointer */
(void) mca_base_var_env_name("crs", &tmp_env_var);
opal_setenv(tmp_env_var,
"none",
true, &environ);
/* Don't free the environment variable name. It is used again below */
/*
* Setup any ORTE stuff we might need
*/
if (OPAL_SUCCESS != (ret = orte_init(&argc, &argv, ORTE_PROC_TOOL))) {
exit_status = ret;
goto cleanup;
}
/* Unset these now that we no longer need them */
opal_unsetenv(tmp_env_var, &environ);
free(tmp_env_var);
tmp_env_var = NULL;
(void) mca_base_var_env_name("opal_cr_is_tool", &tmp_env_var);
opal_unsetenv(tmp_env_var, &environ);
free(tmp_env_var);
tmp_env_var = NULL;
cleanup:
return exit_status;
}
int orcm_init(orcm_proc_type_t flags)
{
int ret;
char *error, *envar;
int spin;
opal_output_stream_t lds;
if (0 < orcm_initialized) {
/* track number of times we have been called */
orcm_initialized++;
return ORCM_SUCCESS;
}
orcm_initialized++;
if (NULL != getenv("ORCM_MCA_spin")) {
spin = 1;
/* spin until a debugger can attach */
while (0 != spin) {
ret = 0;
while (ret < 10000) {
ret++;
};
}
}
/* prior to initializing the OPAL layer, check to see
* if the OPAL (and friends) install location has been
* moved. In order to avoid conflicts with any other
* OPAL-using software, the relocation point will have
* been expressed as a set of "ORCM_foo" envars. We
* therefore check for the ORCM_foo values, and name-shift
* any we find to OPAL_foo so that OPAL will find them.
* Since all ORCM tools will have already copied their
* local environment, these name-shifted vars will not
* appear in the environment of any launched processes */
if (NULL != (envar = getenv("ORCM_PREFIX"))) {
opal_unsetenv("ORCM_PREFIX", &environ);
opal_setenv("OPAL_PREFIX", envar, true, &environ);
}
if (NULL != (envar = getenv("ORCM_LIBDIR"))) {
opal_unsetenv("ORCM_LIBDIR", &environ);
opal_setenv("OPAL_LIBDIR", envar, true, &environ);
}
if (NULL != (envar = getenv("ORCM_DATADIR"))) {
opal_unsetenv("ORCM_DATADIR", &environ);
opal_setenv("OPAL_DATADIR", envar, true, &environ);
}
/* initialize the opal layer */
if (ORTE_SUCCESS != (ret = opal_init(NULL, NULL))) {
error = "opal_init";
goto error;
}
orcm_debug_verbosity = -1;
(void) mca_base_var_register ("orcm", "orcm", NULL, "debug_verbose",
"Verbosity level for ORCM debug messages (default: 1)",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orcm_debug_verbosity);
if (0 <= orcm_debug_verbosity) {
/* get a debug output channel */
OBJ_CONSTRUCT(&lds, opal_output_stream_t);
lds.lds_want_stdout = true;
orcm_debug_output = opal_output_open(&lds);
OBJ_DESTRUCT(&lds);
/* set the verbosity */
opal_output_set_verbosity(orcm_debug_output, orcm_debug_verbosity);
}
/* ensure we know the type of proc for when we finalize */
orte_process_info.proc_type = flags;
/* setup the locks */
if (ORTE_SUCCESS != (ret = orte_locks_init())) {
error = "orte_locks_init";
goto error;
}
/* register handler for errnum -> string conversion */
opal_error_register("ORTE", ORTE_ERR_BASE, ORTE_ERR_MAX, orte_err2str);
/* Ensure the rest of the process info structure is initialized */
if (ORTE_SUCCESS != (ret = orte_proc_info())) {
error = "orte_proc_info";
goto error;
}
/* register handler for errnum -> string conversion */
opal_error_register("ORCM", ORCM_ERR_BASE, ORCM_ERR_MAX, orcm_err2str);
/* register handler for attr key -> string conversion */
if (ORTE_SUCCESS != (ret = orte_attr_register("orcm", ORCM_ATTR_KEY_BASE, ORCM_ATTR_KEY_MAX,
orcm_attr_key_print))) {
error = "register attr print";
goto error;
}
/* we don't need a progress thread as all our tools loop inside themselves,
* so define orte_event_base to be the base opal_event_base
*/
orte_event_base = opal_sync_event_base;
/* setup the globals */
orcm_clusters = OBJ_NEW(opal_list_t);
orcm_schedulers = OBJ_NEW(opal_list_t);
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orcm_parser_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orcm_parser_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orcm_parser_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orcm_parser_select";
goto error;
}
/* everyone must open the cfgi framework */
if (ORCM_SUCCESS != (ret = mca_base_framework_open(&orcm_cfgi_base_framework, 0))) {
error = "orcm_cfgi_base_open";
goto error;
}
if (ORCM_SUCCESS != (ret = orcm_cfgi_base_select())) {
error = "orcm_cfgi_select";
goto error;
}
envar = getenv("ORCM_MCA_logical_group_config_file");
if (ORCM_SUCCESS != (ret = orcm_logical_group_load_to_memory(envar))) {
error = "orcm_logical_group_load_to_memory";
goto error;
}
if (ORCM_SCHED == flags) {
if (NULL == (envar = getenv("ORCM_MCA_event_exec_path"))) {
asprintf(&orcm_event_exec_path, "%s/bin", opal_install_dirs.prefix);
} else {
orcm_event_exec_path = strdup(envar);
}
if (NULL == orcm_event_exec_path) {
error = "orcm_event_exec_path";
goto error;
}
}
/* everyone must open the sst framework */
if (ORCM_SUCCESS != (ret = mca_base_framework_open(&orcm_sst_base_framework, 0))) {
error = "orcm_sst_base_open";
goto error;
}
if (ORCM_SUCCESS != (ret = orcm_sst_base_select())) {
error = "orcm_sst_select";
goto error;
}
/* open the ESS and select the correct module for this environment - the
* orcm module is basically a no-op, but we need the framework defined
* as other parts of ORTE will want to call it
*/
if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_ess_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
error = "orte_ess_base_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_ess_base_select())) {
error = "orte_ess_base_select";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_ess.init())) {
error = "orte_ess_init";
goto error;
}
/* initialize us - we will register the ORTE-level MCA params in there */
if (ORTE_SUCCESS != (ret = orcm_sst.init())) {
error = "orte_init";
goto error;
}
/* setup the orte_show_help system - don't do this until the
* end as otherwise show_help messages won't appear
*/
if (ORTE_SUCCESS != (ret = orte_show_help_init())) {
error = "opal_output_init";
goto error;
}
/* initialize orcm datatype support */
if (ORCM_SUCCESS != (ret = orcm_dt_init())) {
error = "orcm_dt_init";
goto error;
}
/* flag that orte is initialized so things can work */
orte_initialized = true;
orte_help_want_aggregate = false;
return ORCM_SUCCESS;
error:
if (ORCM_ERR_SILENT != ret) {
opal_show_help("help-orcm-runtime.txt",
"orcm_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ret;
}
static int initialize(int argc, char *argv[])
{
int ret, exit_status = OPAL_SUCCESS;
char * tmp_env_var = NULL;
/*
* Make sure to init util before parse_args
* to ensure installdirs is setup properly
* before calling mca_base_open();
*/
if( OPAL_SUCCESS != (ret = opal_init_util(&argc, &argv)) ) {
return ret;
}
/*
* Parse Command line arguments
*/
if (OPAL_SUCCESS != (ret = parse_args(argc, argv))) {
exit_status = ret;
goto cleanup;
}
/*
* Setup OPAL Output handle from the verbose argument
*/
if( opal_restart_globals.verbose ) {
opal_restart_globals.output = opal_output_open(NULL);
opal_output_set_verbosity(opal_restart_globals.output, 10);
} else {
opal_restart_globals.output = 0; /* Default=STDOUT */
}
/*
* Turn off the selection of the CRS component,
* we need to do that later
*/
(void) mca_base_var_env_name("crs_base_do_not_select", &tmp_env_var);
opal_setenv(tmp_env_var,
"1", /* turn off the selection */
true, &environ);
free(tmp_env_var);
tmp_env_var = NULL;
/*
* Make sure we select the proper compress component.
*/
if( NULL != opal_restart_globals.snapshot_compress ) {
(void) mca_base_var_env_name("compress", &tmp_env_var);
opal_setenv(tmp_env_var,
opal_restart_globals.snapshot_compress,
true, &environ);
free(tmp_env_var);
tmp_env_var = NULL;
}
/*
* Initialize the OPAL layer
*/
if (OPAL_SUCCESS != (ret = opal_init(&argc, &argv))) {
exit_status = ret;
goto cleanup;
}
/*
* If the checkpoint was compressed, then decompress it before continuing
*/
if( NULL != opal_restart_globals.snapshot_compress ) {
char * zip_dir = NULL;
char * tmp_str = NULL;
/* Make sure to clear the selection for the restart,
* this way the user can swich compression mechanism
* across restart
*/
(void) mca_base_var_env_name("compress", &tmp_env_var);
opal_unsetenv(tmp_env_var, &environ);
free(tmp_env_var);
tmp_env_var = NULL;
opal_asprintf(&zip_dir, "%s/%s%s",
opal_restart_globals.snapshot_loc,
opal_restart_globals.snapshot_ref,
opal_restart_globals.snapshot_compress_postfix);
if (0 > (ret = access(zip_dir, F_OK)) ) {
opal_output(opal_restart_globals.output,
"Error: Unable to access the file [%s]!",
zip_dir);
exit_status = OPAL_ERROR;
goto cleanup;
}
opal_output_verbose(10, opal_restart_globals.output,
"Decompressing (%s)",
zip_dir);
opal_compress.decompress(zip_dir, &tmp_str);
if( NULL != zip_dir ) {
free(zip_dir);
zip_dir = NULL;
}
if( NULL != tmp_str ) {
free(tmp_str);
tmp_str = NULL;
}
}
/*
* If a cache directory has been suggested, see if it exists
*/
if( NULL != opal_restart_globals.snapshot_cache ) {
if(0 == (ret = access(opal_restart_globals.snapshot_cache, F_OK)) ) {
opal_output_verbose(10, opal_restart_globals.output,
"Using the cached snapshot (%s) instead of (%s)",
opal_restart_globals.snapshot_cache,
opal_restart_globals.snapshot_loc);
if( NULL != opal_restart_globals.snapshot_loc ) {
free(opal_restart_globals.snapshot_loc);
opal_restart_globals.snapshot_loc = NULL;
}
opal_restart_globals.snapshot_loc = opal_dirname(opal_restart_globals.snapshot_cache);
} else {
opal_show_help("help-opal-restart.txt", "cache_not_avail", true,
opal_restart_globals.snapshot_cache,
opal_restart_globals.snapshot_loc);
}
}
/*
* Mark this process as a tool
*/
opal_cr_is_tool = true;
cleanup:
return exit_status;
}
static int do_child(orte_app_context_t* context,
orte_proc_t *child,
char **environ_copy,
orte_job_t *jobdat, int write_fd,
orte_iof_base_io_conf_t opts)
{
int i, rc;
sigset_t sigs;
long fd, fdmax = sysconf(_SC_OPEN_MAX);
char *param, *msg;
if (orte_forward_job_control) {
/* Set a new process group for this child, so that a
SIGSTOP can be sent to it without being sent to the
orted. */
setpgid(0, 0);
}
/* Setup the pipe to be close-on-exec */
fcntl(write_fd, F_SETFD, FD_CLOEXEC);
if (NULL != child) {
/* setup stdout/stderr so that any error messages that we
may print out will get displayed back at orterun.
NOTE: Definitely do this AFTER we check contexts so
that any error message from those two functions doesn't
come out to the user. IF we didn't do it in this order,
THEN a user who gives us a bad executable name or
working directory would get N error messages, where
N=num_procs. This would be very annoying for large
jobs, so instead we set things up so that orterun
always outputs a nice, single message indicating what
happened
*/
if (ORTE_SUCCESS != (i = orte_iof_base_setup_child(&opts,
&environ_copy))) {
ORTE_ERROR_LOG(i);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt",
"iof setup failed",
orte_process_info.nodename, context->app);
/* Does not return */
}
#if OPAL_HAVE_HWLOC
{
hwloc_cpuset_t cpuset;
hwloc_obj_t root;
opal_hwloc_topo_data_t *sum;
/* Set process affinity, if given */
if (NULL == child->cpu_bitmap) {
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
}
if (opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
} else {
if (0 == strlen(child->cpu_bitmap)) {
/* this proc is not bound */
if (opal_hwloc_report_bindings) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
/* if the daemon is bound, then we need to "free" this proc */
if (NULL != orte_daemon_cores) {
root = hwloc_get_root_obj(opal_hwloc_topology);
if (NULL == root->userdata) {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "incorrectly bound",
orte_process_info.nodename, context->app,
__FILE__, __LINE__);
}
sum = (opal_hwloc_topo_data_t*)root->userdata;
/* bind this proc to all available processors */
hwloc_set_cpubind(opal_hwloc_topology, sum->available, 0);
}
/* provide a nice string representation of what we bound to */
(void) mca_base_var_env_name ("orte_base_applied_binding", ¶m);
opal_setenv(param, child->cpu_bitmap, true, &environ_copy);
free(param);
goto PROCEED;
}
/* convert the list to a cpu bitmap */
cpuset = hwloc_bitmap_alloc();
if (0 != (rc = hwloc_bitmap_list_sscanf(cpuset, child->cpu_bitmap))) {
/* See comment above about "This may be a small memory leak" */
asprintf(&msg, "hwloc_bitmap_sscanf returned \"%s\" for the string \"%s\"",
opal_strerror(rc), child->cpu_bitmap);
if (NULL == msg) {
msg = "failed to convert bitmap list to hwloc bitmap";
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding) &&
OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
/* If binding is required and a binding directive was explicitly
* given (i.e., we are not binding due to a default policy),
* send an error up the pipe (which exits -- it doesn't return).
*/
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename,
context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
goto PROCEED;
}
}
/* bind as specified */
rc = hwloc_set_cpubind(opal_hwloc_topology, cpuset, 0);
/* if we got an error and this wasn't a default binding policy, then report it */
if (rc < 0 && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
char *tmp = NULL;
if (errno == ENOSYS) {
msg = "hwloc indicates cpu binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates cpu binding cannot be enforced";
} else {
asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
opal_strerror(rc), child->cpu_bitmap);
}
if (OPAL_BINDING_REQUIRED(jobdat->map->binding)) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"binding generic error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
if (NULL != tmp) {
free(tmp);
free(msg);
}
goto PROCEED;
}
if (NULL != tmp) {
free(tmp);
free(msg);
}
}
if (0 == rc && opal_hwloc_report_bindings) {
char tmp1[1024], tmp2[1024];
hwloc_cpuset_t mycpus;
/* get the cpus we are bound to */
mycpus = hwloc_bitmap_alloc();
if (hwloc_get_cpubind(opal_hwloc_topology,
mycpus,
HWLOC_CPUBIND_PROCESS) < 0) {
opal_output(0, "MCW rank %d is not bound",
child->name.vpid);
} else {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) {
opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", child->name.vpid);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus);
opal_output(0, "MCW rank %d bound to %s: %s",
child->name.vpid, tmp1, tmp2);
}
}
hwloc_bitmap_free(mycpus);
/* avoid reporting it twice */
(void) mca_base_var_env_name ("hwloc_base_report_bindings", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
}
/* set memory affinity policy - if we get an error, don't report
* anything unless the user actually specified the binding policy
*/
rc = opal_hwloc_base_set_process_membind_policy();
if (ORTE_SUCCESS != rc && OPAL_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
if (errno == ENOSYS) {
msg = "hwloc indicates memory binding not supported";
} else if (errno == EXDEV) {
msg = "hwloc indicates memory binding cannot be enforced";
} else {
msg = "failed to bind memory";
}
if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
/* If binding is required, send an error up the pipe (which exits
-- it doesn't return). */
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"memory binding error",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
} else {
send_warn_show_help(write_fd,
"help-orte-odls-default.txt", "memory not bound",
orte_process_info.nodename, context->app, msg,
__FILE__, __LINE__);
goto PROCEED;
}
}
}
}
#endif
} else if (!(ORTE_JOB_CONTROL_FORWARD_OUTPUT & jobdat->controls)) {
/* tie stdin/out/err/internal to /dev/null */
int fdnull;
for (i=0; i < 3; i++) {
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > i && i != write_fd) {
dup2(fdnull, i);
}
close(fdnull);
}
fdnull = open("/dev/null", O_RDONLY, 0);
if (fdnull > opts.p_internal[1]) {
dup2(fdnull, opts.p_internal[1]);
}
close(fdnull);
}
#if OPAL_HAVE_HWLOC
PROCEED:
#endif
/* if the user requested it, set the system resource limits */
if (OPAL_SUCCESS != (rc = opal_util_init_sys_limits(&msg))) {
send_error_show_help(write_fd, 1, "help-orte-odls-default.txt",
"set limit",
orte_process_info.nodename, context->app,
__FILE__, __LINE__, msg);
}
/* ensure we only do this once */
(void) mca_base_var_env_name("opal_set_max_sys_limits", ¶m);
opal_unsetenv(param, &environ_copy);
free(param);
/* close all file descriptors w/ exception of stdin/stdout/stderr,
the pipe used for the IOF INTERNAL messages, and the pipe up to
the parent. */
for(fd=3; fd<fdmax; fd++) {
if (fd != opts.p_internal[1] && fd != write_fd) {
close(fd);
}
}
if (context->argv == NULL) {
context->argv = malloc(sizeof(char*)*2);
context->argv[0] = strdup(context->app);
context->argv[1] = NULL;
}
/* Set signal handlers back to the default. Do this close to
the exev() because the event library may (and likely will)
reset them. If we don't do this, the event library may
have left some set that, at least on some OS's, don't get
reset via fork() or exec(). Hence, the launched process
could be unkillable (for example). */
set_handler_default(SIGTERM);
set_handler_default(SIGINT);
set_handler_default(SIGHUP);
set_handler_default(SIGPIPE);
set_handler_default(SIGCHLD);
/* Unblock all signals, for many of the same reasons that we
set the default handlers, above. This is noticable on
Linux where the event library blocks SIGTERM, but we don't
want that blocked by the launched process. */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* Exec the new executable */
if (10 < opal_output_get_verbosity(orte_odls_base_framework.framework_output)) {
int jout;
opal_output(0, "%s STARTING %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), context->app);
for (jout=0; NULL != context->argv[jout]; jout++) {
opal_output(0, "%s\tARGV[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, context->argv[jout]);
}
for (jout=0; NULL != environ_copy[jout]; jout++) {
opal_output(0, "%s\tENVIRON[%d]: %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), jout, environ_copy[jout]);
}
}
execve(context->app, context->argv, environ_copy);
send_error_show_help(write_fd, 1,
"help-orte-odls-default.txt", "execve error",
orte_process_info.nodename, context->app, strerror(errno));
/* Does not return */
}