static int test_spawn_common(char *my_nspace, int my_rank, int blocking)
{
int rc;
pmix_app_t *apps;
size_t napps;
char nspace[PMIX_MAX_NSLEN+1];
memset(nspace, 0, PMIX_MAX_NSLEN+1);
napps = 1;
PMIX_APP_CREATE(apps, napps);
if (blocking) {
if (PMIX_SUCCESS != (rc = PMIx_Spawn(NULL, 0, apps, napps, nspace))) {
PMIX_APP_FREE(apps, napps);
return rc;
}
} else {
spawn_cbdata cbdata;
cbdata.in_progress = 1;
memset(cbdata.nspace, 0, PMIX_MAX_NSLEN);
rc = PMIx_Spawn_nb(NULL, 0, apps, napps, spawn_cb, (void*)&cbdata);
if (PMIX_SUCCESS != rc) {
PMIX_APP_FREE(apps, napps);
return rc;
}
PMIX_WAIT_FOR_COMPLETION(cbdata.in_progress);
strncpy(nspace, cbdata.nspace, strlen(cbdata.nspace)+1);
}
PMIX_APP_FREE(apps, napps);
if (strncmp(nspace, "foobar", strlen(nspace)+1)) {
return PMIX_ERROR;
}
return rc;
}
int PMI2_Job_Spawn(int count, const char * cmds[],
int argcs[], const char ** argvs[],
const int maxprocs[],
const int info_keyval_sizes[],
const PMI_keyval_t *info_keyval_vectors[],
int preput_keyval_size,
const PMI_keyval_t *preput_keyval_vector[],
char jobId[], int jobIdSize,
int errors[])
{
pmix_status_t rc = PMIX_SUCCESS;
pmix_app_t *apps;
int i, k;
size_t j;
char *evar;
PMI2_CHECK();
if (NULL == cmds) {
return PMI2_ERR_INVALID_ARGS;
}
/* setup the apps */
PMIX_APP_CREATE(apps, count);
for (i=0; i < count; i++) {
apps[i].cmd = strdup(cmds[i]);
apps[i].maxprocs = maxprocs[i];
apps[i].argv = pmix_argv_copy((char**)argvs[i]);
apps[i].argc = pmix_argv_count(apps[i].argv);
apps[i].ninfo = info_keyval_sizes[i];
apps[i].info = (pmix_info_t*)malloc(apps[i].ninfo * sizeof(pmix_info_t));
/* copy the info objects */
for (j=0; j < apps[i].ninfo; j++) {
(void)strncpy(apps[i].info[j].key, info_keyval_vectors[i][j].key, PMIX_MAX_KEYLEN);
apps[i].info[j].value.type = PMIX_STRING;
apps[i].info[j].value.data.string = strdup(info_keyval_vectors[i][j].val);
}
/* push the preput values into the apps environ */
for (k=0; k < preput_keyval_size; k++) {
(void)asprintf(&evar, "%s=%s", preput_keyval_vector[j]->key, preput_keyval_vector[j]->val);
pmix_argv_append_nosize(&apps[i].env, evar);
free(evar);
}
}
rc = PMIx_Spawn(NULL, 0, apps, count, NULL);
/* tear down the apps array */
for (i=0; i < count; i++) {
PMIX_APP_DESTRUCT(&apps[i]);
}
free(apps);
if (NULL != errors) {
for (i=0; i < count; i++) {
errors[i] = convert_err(rc);
}
}
return convert_err(rc);
}
static pmix_status_t spawn_debugger(char *appspace, myrel_t *myrel)
{
pmix_status_t rc;
pmix_info_t *dinfo;
pmix_app_t *debugger;
size_t dninfo;
char cwd[1024];
char dspace[PMIX_MAX_NSLEN+1];
mylock_t mylock;
pmix_status_t code = PMIX_ERR_JOB_TERMINATED;
/* setup the debugger */
PMIX_APP_CREATE(debugger, 1);
debugger[0].cmd = strdup("./debuggerd");
PMIX_ARGV_APPEND(rc, debugger[0].argv, "./debuggerd");
getcwd(cwd, 1024); // point us to our current directory
debugger[0].cwd = strdup(cwd);
/* provide directives so the daemons go where we want, and
* let the RM know these are debugger daemons */
dninfo = 6;
PMIX_INFO_CREATE(dinfo, dninfo);
PMIX_INFO_LOAD(&dinfo[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING); // instruct the RM to launch one copy of the executable on each node
PMIX_INFO_LOAD(&dinfo[1], PMIX_DEBUGGER_DAEMONS, NULL, PMIX_BOOL); // these are debugger daemons
PMIX_INFO_LOAD(&dinfo[1], PMIX_DEBUG_JOB, appspace, PMIX_STRING); // the nspace being debugged
PMIX_INFO_LOAD(&dinfo[2], PMIX_NOTIFY_COMPLETION, NULL, PMIX_BOOL); // notify us when the debugger job completes
PMIX_INFO_LOAD(&dinfo[3], PMIX_DEBUG_WAITING_FOR_NOTIFY, NULL, PMIX_BOOL); // tell the daemon that the proc is waiting to be released
PMIX_INFO_LOAD(&dinfo[4], PMIX_FWD_STDOUT, NULL, PMIX_BOOL); // forward stdout to me
PMIX_INFO_LOAD(&dinfo[5], PMIX_FWD_STDERR, NULL, PMIX_BOOL); // forward stderr to me
/* spawn the daemons */
fprintf(stderr, "Debugger: spawning %s\n", debugger[0].cmd);
if (PMIX_SUCCESS != (rc = PMIx_Spawn(dinfo, dninfo, debugger, 1, dspace))) {
fprintf(stderr, "Debugger daemons failed to launch with error: %s\n", PMIx_Error_string(rc));
PMIX_INFO_FREE(dinfo, dninfo);
PMIX_APP_FREE(debugger, 1);
return rc;
}
/* cleanup */
PMIX_INFO_FREE(dinfo, dninfo);
PMIX_APP_FREE(debugger, 1);
/* register callback for when this job terminates */
myrel->nspace = strdup(dspace);
PMIX_INFO_CREATE(dinfo, 2);
PMIX_INFO_LOAD(&dinfo[0], PMIX_EVENT_RETURN_OBJECT, myrel, PMIX_POINTER);
/* only call me back when this specific job terminates */
PMIX_INFO_LOAD(&dinfo[1], PMIX_NSPACE, dspace, PMIX_STRING);
DEBUG_CONSTRUCT_LOCK(&mylock);
PMIx_Register_event_handler(&code, 1, dinfo, 2,
release_fn, evhandler_reg_callbk, (void*)&mylock);
DEBUG_WAIT_THREAD(&mylock);
rc = mylock.status;
DEBUG_DESTRUCT_LOCK(&mylock);
PMIX_INFO_FREE(dinfo, 2);
return rc;
}
int main(int argc, char **argv)
{
int rc;
pmix_value_t value;
pmix_value_t *val = &value;
pmix_proc_t proc;
uint32_t nprocs;
char nsp2[PMIX_MAX_NSLEN+1];
pmix_app_t *app;
char hostname[PMIX_MAXHOSTNAMELEN];
pmix_proc_t *peers;
size_t npeers, ntmp=0;
char *nodelist;
gethostname(hostname, sizeof(hostname));
/* init us */
if (PMIX_SUCCESS != (rc = PMIx_Init(&myproc, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Init failed: %d", myproc.nspace, myproc.rank, rc);
exit(0);
}
pmix_output(0, "Client ns %s rank %d: Running", myproc.nspace, myproc.rank);
/* get our universe size */
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_UNIV_SIZE, NULL, 0, &val))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Get universe size failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
nprocs = val->data.uint32;
PMIX_VALUE_RELEASE(val);
pmix_output(0, "Client %s:%d universe size %d", myproc.nspace, myproc.rank, nprocs);
/* call fence to sync */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Fence failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
/* rank=0 calls spawn */
if (0 == myproc.rank) {
PMIX_APP_CREATE(app, 1);
app->cmd = strdup("gumby");
app->maxprocs = 2;
pmix_argv_append_nosize(&app->argv, "gumby");
pmix_argv_append_nosize(&app->argv, "-n");
pmix_argv_append_nosize(&app->argv, "2");
pmix_setenv("PMIX_ENV_VALUE", "3", true, &app->env);
PMIX_INFO_CREATE(app->info, 2);
(void)strncpy(app->info[0].key, "DARTH", PMIX_MAX_KEYLEN);
app->info[0].value.type = PMIX_INT8;
app->info[0].value.data.int8 = 12;
(void)strncpy(app->info[1].key, "VADER", PMIX_MAX_KEYLEN);
app->info[1].value.type = PMIX_DOUBLE;
app->info[1].value.data.dval = 12.34;
pmix_output(0, "Client ns %s rank %d: calling PMIx_Spawn", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Spawn(NULL, 0, app, 1, nsp2))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Spawn failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
PMIX_APP_FREE(app, 1);
/* check to see if we got the expected info back */
if (0 != strncmp(nsp2, "DYNSPACE", PMIX_MAX_NSLEN)) {
pmix_output(0, "Client ns %s rank %d: PMIx_Spawn returned incorrect nspace: %s", myproc.nspace, myproc.rank, nsp2);
goto done;
} else {
pmix_output(0, "Client ns %s rank %d: PMIx_Spawn succeeded returning nspace: %s", myproc.nspace, myproc.rank, nsp2);
}
/* get their universe size */
(void)strncpy(proc.nspace, nsp2, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
val = NULL;
if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_UNIV_SIZE, NULL, 0, &val)) ||
NULL == val) {
pmix_output(0, "Client ns %s rank %d: PMIx_Get universe size failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
ntmp = val->data.uint32;
PMIX_VALUE_RELEASE(val);
pmix_output(0, "Client %s:%d universe %s size %d", myproc.nspace, myproc.rank, nsp2, (int)ntmp);
}
/* just cycle the connect/disconnect functions */
if (PMIX_SUCCESS != (rc = PMIx_Connect(&proc, 1, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Connect failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Connect succeeded",
myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Disconnect(&proc, 1, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Disonnect failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Disconnect succeeded", myproc.nspace, myproc.rank);
/* finally, test the resolve functions */
if (0 == myproc.rank) {
if (PMIX_SUCCESS != (rc = PMIx_Resolve_peers(hostname, NULL, &peers, &npeers))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers failed for nspace %s: %d", myproc.nspace, myproc.rank, nsp2, rc);
goto done;
}
if ((nprocs+ntmp) != npeers) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers returned incorrect npeers: %d vs %d", myproc.nspace, myproc.rank, (int)(nprocs+ntmp), (int)npeers);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers returned %d npeers", myproc.nspace, myproc.rank, (int)npeers);
if (PMIX_SUCCESS != (rc = PMIx_Resolve_nodes(nsp2, &nodelist))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_nodes failed for nspace %s: %d", myproc.nspace, myproc.rank, nsp2, rc);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_nodes %s", myproc.nspace, myproc.rank, nodelist);
} else {
if (PMIX_SUCCESS != (rc = PMIx_Resolve_peers(hostname, myproc.nspace, &peers, &npeers))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers failed for nspace %s: %d", myproc.nspace, myproc.rank, myproc.nspace, rc);
goto done;
}
if (nprocs != npeers) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers returned incorrect npeers: %d vs %d", myproc.nspace, myproc.rank, nprocs, (int)npeers);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_peers returned %d npeers", myproc.nspace, myproc.rank, (int)npeers);
if (PMIX_SUCCESS != (rc = PMIx_Resolve_nodes(myproc.nspace, &nodelist))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_nodes failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
pmix_output(0, "Client ns %s rank %d: PMIx_Resolve_nodes %s", myproc.nspace, myproc.rank, nodelist);
}
PMIX_PROC_FREE(peers, npeers);
free(nodelist);
done:
/* call fence to sync */
PMIX_PROC_CONSTRUCT(&proc);
(void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN);
proc.rank = PMIX_RANK_WILDCARD;
if (PMIX_SUCCESS != (rc = PMIx_Fence(&proc, 1, NULL, 0))) {
pmix_output(0, "Client ns %s rank %d: PMIx_Fence failed: %d", myproc.nspace, myproc.rank, rc);
goto done;
}
/* finalize us */
pmix_output(0, "Client ns %s rank %d: Finalizing", myproc.nspace, myproc.rank);
if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc);
} else {
fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank);
}
fflush(stderr);
return(0);
}
static int attach_to_running_job(char *nspace)
{
pmix_status_t rc;
pmix_proc_t myproc;
pmix_query_t *query;
size_t nq;
myquery_data_t *q;
/* query the active nspaces so we can verify that the
* specified one exists */
nq = 1;
PMIX_QUERY_CREATE(query, nq);
PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_NAMESPACES);
q = (myquery_data_t*)malloc(sizeof(myquery_data_t));
DEBUG_CONSTRUCT_LOCK(&q->lock);
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)q))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Query_info failed: %d\n", myproc.nspace, myproc.rank, rc);
return -1;
}
DEBUG_WAIT_THREAD(&q->lock);
DEBUG_DESTRUCT_LOCK(&q->lock);
if (NULL == q->info) {
fprintf(stderr, "Query returned no info\n");
return -1;
}
/* the query should have returned a comma-delimited list of nspaces */
if (PMIX_STRING != q->info[0].value.type) {
fprintf(stderr, "Query returned incorrect data type: %d\n", q->info[0].value.type);
return -1;
}
if (NULL == q->info[0].value.data.string) {
fprintf(stderr, "Query returned no active nspaces\n");
return -1;
}
fprintf(stderr, "Query returned %s\n", q->info[0].value.data.string);
return 0;
#if 0
/* split the returned string and look for the given nspace */
/* if not found, then we have an error */
PMIX_INFO_FREE(info, ninfo);
/* get the proctable for this nspace */
ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
(void)strncpy(info[0].key, PMIX_QUERY_PROC_TABLE, PMIX_MAX_KEYLEN);
(void)strncpy(info[0].qualifier, nspace, PMIX_MAX_KEYLEN);
if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(info, ninfo, infocbfunc, (void*)&active))) {
fprintf(stderr, "Client ns %s rank %d: PMIx_Query_info_nb failed: %d\n", myproc.nspace, myproc.rank, rc);
return -1;
}
/* wait to get a response */
/* the query should have returned a data_array */
if (PMIX_DATA_ARRAY != info[0].type) {
fprintf(stderr, "Query returned incorrect data type: %d\n", info[0].type);
return -1;
}
if (NULL == info[0].data.darray.array) {
fprintf(stderr, "Query returned no proctable info\n");
return -1;
}
/* the data array consists of a struct:
* size_t size;
* void* array;
*
* In this case, the array is composed of pmix_proc_info_t structs:
* pmix_proc_t proc; // contains the nspace,rank of this proc
* char* hostname;
* char* executable_name;
* pid_t pid;
* int exit_code;
* pmix_proc_state_t state;
*/
/* this is where a debugger tool would process the proctable to
* create whatever blob it needs to provide to its daemons */
PMIX_INFO_FREE(info, ninfo);
/* setup the debugger daemon spawn request */
napps = 1;
PMIX_APP_CREATE(app, napps);
/* setup the name of the daemon executable to launch */
app[0].cmd = strdup("debuggerdaemon");
app[0].argc = 1;
app[0].argv = (char**)malloc(2*sizeof(char*));
app[0].argv[0] = strdup("debuggerdaemon");
app[0].argv[1] = NULL;
/* provide directives so the daemons go where we want, and
* let the RM know these are debugger daemons */
ninfo = 3;
PMIX_INFO_CREATE(app[0].info, ninfo);
PMIX_INFO_LOAD(&app[0].info[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING); // instruct the RM to launch one copy of the executable on each node
PMIX_INFO_LOAD(&app[0].info[1], PMIX_DEBUGGER_DAEMONS, true, PMIX_BOOL); // these are debugger daemons
PMIX_INFO_LOAD(&app[0].info[2], PMIX_DEBUG_TARGET, nspace, PMIX_STRING); // the "jobid" of the application to be debugged
/* spawn the daemons */
PMIx_Spawn(NULL, 0, app, napps, dspace);
/* cleanup */
PMIX_APP_FREE(app, napps);
/* this is where a debugger tool would wait until the debug operation is complete */
return 0;
#endif
}
int main(int argc, char **argv, const char **environ)
{
pmix_status_t rc;
pmix_info_t *info = NULL;
bool flag;
pmix_status_t retval;
pmix_app_t *spawned_app = NULL;
pmix_info_t *job_info = NULL;
pmix_info_t *proc_info = NULL;
int job_info_count = 0;
int job_info_index = 0;
int proc_info_count = 0;
int proc_info_index = 0;
char spawned_nsp[PMIX_MAX_NSLEN+1];
char *path_to_app = NULL;
char *host_to_use = NULL;
int number_of_clients = 0;
int temp_counter = 0;
done_flag = false;
gethostname(hostn, 500);
int spawned_app_argc = 0;
char **scr_environ = NULL;
int proc_count = 1;
int node_count = 0;
bool blocking_mode = true;
char *node_list = NULL;
bool forward_all_scr_envs = false;
bool pmix_mode = false;
const char *optstring = "+n:N:L:x:bB:pPvhe";
int temp_slen=0;
/* todo: add arg parsing with ompi schizo */
verbose_print = false;
int sleep_max = 30;
const int fixed_sleep = 5;
int c;
while((c = getopt(argc, argv, optstring)) != -1){
switch(c){
case 'h':
print_usage(argv[0]);
exit(0);
break;
case 'n':
proc_count = atoi(optarg);
if(proc_count <= 0 || proc_count > 100){
printf("outside the range of allowable instances to spawn [1-100]\n");
exit(1);
}
if(verbose_print) {
printf("proc_count = %d\n", proc_count);
}
break;
case 'N':
/* node_count = atoi(optarg); */
node_count = 1;
if(verbose_print) {
printf("node_count = %d\n", node_count);
}
break;
case 'B':
blocking_mode = true;
sleep_max = atoi(optarg);
if(sleep_max < 0){
printf("can't sleep for less than 0 seconds\n");
exit(1);
}
if(verbose_print){
printf("blocking mode = %x\n", blocking_mode);
}
break;
case 'b':
blocking_mode = false;
if(verbose_print){
printf("blocking mode = %x\n", blocking_mode);
}
break;
case 'L':
node_list = optarg;
host_to_use = node_list;
if(verbose_print){
printf("node_list = '%s'\n", node_list);
}
break;
case 'x':
temp_slen = strlen(optarg);
/* check if the string is the same length as 'SCR', if so compare them */
if(temp_slen == strlen(SCR_STRING)){
if(strncmp(optarg, SCR_STRING, strlen(SCR_STRING)) == 0){
/* if the string is SCR, then forward all SCR related env vars */
if(verbose_print) printf("all scr envs will be forwarded\n");
forward_all_scr_envs = true;
}
else{
/* handled like a normal env var */
handle_standard_env_var(optarg, &scr_environ);
}
}
else{
/*handled like a normal env var */
handle_standard_env_var(optarg, &scr_environ);
}
break;
case 'v':
verbose_print = true;
break;
case 'p':
pmix_mode = true;
if(verbose_print){
printf("pmix_mode = %x\n", pmix_mode);
}
break;
case 'P':
pmix_mode = false;
if(verbose_print){
printf("pmix_mode = %x\n", pmix_mode);
}
break;
case 'e':
experimental = true;
break;
case '?':
printf("missing a required argument or invalid option: %x\n", optopt);
print_usage(argv[0]);
exit(1);
break;
default:
printf("Unrecognized argument: %c\n", c);
print_usage(argv[0]);
exit(1);
break;
}
}
/* number of instances to spawn */
number_of_clients = proc_count;
/* check to make sure an application was specified to launch */
if( optind < argc ){
/* if optind is < argc, it means there is at least one more arg
* beyond the args for this program */
path_to_app = argv[optind];
spawned_app_argc = argc - optind;
if(verbose_print) {
printf("app to launch: %s @ %s:%d\n",
path_to_app, __FILE__, __LINE__);
}
}
else{
printf("program_to_spawn option was not provded\n");
print_usage(argv[0]);
exit(1);
}
if(verbose_print){
printf("master process will spawn %d instances; app to run: %s\n\n",
number_of_clients, path_to_app);
printf("pmix version: %s (host: %s)\n", PMIx_Get_version(), hostn);
}
/* init pmix */
retval = PMIx_Init(&main_proc, NULL, 0);
if(retval != PMIX_SUCCESS){
error_helper(retval, hostn, "error initializing pmix");
exit(0);
}
if(verbose_print){
printf("rank %d, host '%s', nspace: '%s' init'd pmix succesfully\n\n",
main_proc.rank, hostn, main_proc.nspace);
}
/* we need to attach to a "system" PMIx server so we
* can ask it to spawn applications for us. There can
* only be one such connection on a node, so we will
* instruct the tool library to only look for it */
int ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
flag = true;
PMIX_INFO_LOAD(&info[0], PMIX_CONNECT_TO_SYSTEM, &flag, PMIX_BOOL);
/* initialize the library and make the connection */
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&tool_proc, NULL, 0 ))) {
fprintf(stderr, "PMIx_tool_init failed: %d\n", rc );
exit(rc);
}
if (0 < ninfo) {
PMIX_INFO_FREE(info, ninfo);
}
/* first call fence to sync all processes */
retval = fence_helper();
if(retval != PMIX_SUCCESS)
{
error_helper(retval, hostn, "error fencing");
exit(retval);
}
/* Process SCR env vars if needed */
if(forward_all_scr_envs){
parse_all_scr_envs(&scr_environ, environ);
}
/* finalize the env array so a NULL is in place */
finalize_array(scr_environ);
/* Setup info structs to pass to this: */
/* pmix_info_t *error_info = NULL; */
/* PMIX_INFO_CREATE(error_info, 1); */
/*
strncpy(error_info[0].key, PMIX_ERROR_GROUP_ABORT, PMIX_MAX_KEYLEN);
error_info[0].value.type = PMIX_BOOL;
error_info[0].value.data.flag = true;
*/
/* strncpy(error_info[0].key, PMIX_ERROR_GROUP_SPAWN, PMIX_MAX_KEYLEN);
int t_val = 1;
pmix_value_load(&error_info[1].value, &t_val, PMIX_BOOL);
*/
/*error_info[1].value.type = PMIX_BOOL;
error_info[1].value.data.flag = true; */
/* strncpy(error_info[2].key, PMIX_ERROR_GROUP_GENERAL, PMIX_MAX_KEYLEN);
error_info[2].value.type = PMIX_BOOL;
error_info[2].value.data.flag = true;
*/
/* TODO: setup error handling when implemented in pmix with the
* following error codes: */
/*
pmix_status_t registered_codes[5];
registered_codes[0] = PMIX_ERR_JOB_TERMINATED;
registered_codes[1] = PMIX_ERR_PROC_ABORTED;
registered_codes[2] = PMIX_ERR_PROC_ABORTING;
*/
PMIx_Register_event_handler(NULL, 0,
NULL, 0,
errhandler_cb,
errhandler_reg_callbk,
(void *) NULL);
/* PMIX_INFO_DESTRUCT(error_info); */
/* allocate memory to hold the spawend app struct */
PMIX_APP_CREATE(spawned_app, 1);
/* maxprocs isn't documented very well, but it appears to control
* how many instances of the spanwed app are created */
spawned_app->maxprocs = number_of_clients;
/* set the app to run */
(void)asprintf(&spawned_app->cmd, "%s", path_to_app);
/* set argv for spawned app starting with remaining argv */
spawned_app->argv = &argv[optind];
/* set the environment pointer */
spawned_app->env = scr_environ;
/*--START: add all proc level infos */
/* add things to the proc level info */
if(!pmix_mode){
job_info_count++;
}
if(host_to_use != NULL){
proc_info_count++;
}
if(verbose_print){
printf("enabling debug feature for forwarding stdout/stderr\n");
proc_info_count+=2;
/* add PMIX_FWD_STDOUT and PMIX_FWD_STDERR later*/
}
if(experimental){
job_info_count++;
}
if(node_count == 1){
job_info_count++;
}
/*--END: add all proc level infos */
/*--START: append actual proc level info */
PMIX_INFO_CREATE(job_info, job_info_count);
PMIX_INFO_CREATE(proc_info, proc_info_count);
/* PMIX_VAL_set_assign(_v, _field, _val ) */
/* PMIX_VAL_set_strdup(_v, _field, _val ) */
if(host_to_use != NULL){
/* add info struct to the spawned app itself for the host */
/* old way */
strncpy(proc_info[proc_info_index].key, PMIX_HOST, PMIX_MAX_KEYLEN);
//proc_info[proc_info_index].value.type = PMIX_STRING;
/* set the data for host list to use */
//proc_info[proc_info_index].value.data.string = host_to_use;
/* end old way */
if(verbose_print) printf("about to set host val\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), string,
host_to_use );
proc_info_index++;
}
if(!pmix_mode){
strncpy(job_info[job_info_index].key, PMIX_NON_PMI,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set non pmix flag\n");
PMIX_VAL_SET(&(job_info[job_info_index].value), flag, true);
job_info_index++;
}
if(verbose_print){
strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDOUT,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set stdout flag\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
proc_info_index++;
strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDERR,
PMIX_MAX_KEYLEN);
if(verbose_print) printf("about to set stderr flag\n");
PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
proc_info_index++;
}
if(experimental){
printf("attempting to perform experiment\n");
bool local_flag = true;
PMIX_INFO_LOAD(&job_info[job_info_index], PMIX_NOTIFY_COMPLETION, &local_flag, PMIX_BOOL);
job_info_index++;
}
if(node_count == 1){
strncpy(job_info[job_info_index].key, PMIX_PPR,
PMIX_MAX_KEYLEN);
PMIX_VAL_SET(&(job_info[job_info_index].value), string,
"1:n");
job_info_index++;
}
/*--END: append actual proc level info */
/* sanity check to make sure we covered all the info structs */
if(proc_info_index != proc_info_count ){
printf("bug: mismatch with appending proc info\n");
exit(1);
}
if(job_info_index != job_info_count){
printf("bug: mismatch with appending job info\n");
exit(1);
}
/* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */
/* fill in job_info */
/*
strncpy(job_info[0].key, PMIX_TIMEOUT, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_INT;
job_info[0].value.data.integer = 10; */
/* strncpy(job_info[0].key, PMIX_NOTIFY_COMPLETION, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_BOOL;
job_info[0].value.data.flag = true; */
/*strncpy(spawned_app->info[0].key, PMIX_DISPLAY_MAP, PMIX_MAX_KEYLEN);
job_info[0].value.type = PMIX_BOOL;
job_info[0].value.data.flag = true;*/
/* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */
spawned_app->info = proc_info;
spawned_app->ninfo = proc_info_count;
if(verbose_print){
printf("proc level info count: %d\n", proc_info_count);
}
/* call spawn */
retval = PMIx_Spawn(job_info, job_info_count,
spawned_app, 1, spawned_nsp);
if(verbose_print) {
printf("rank %d (host %s) just called spawn; spawned nspace: %s, retval:%d\n",
main_proc.rank,
hostn,
spawned_nsp,
retval);
}
if(retval != PMIX_SUCCESS){
error_helper(retval, hostn, "error with spawn");
goto done;
}
/* TODO: TEMPORARY WORKAROUND TO WAIT FOR A SPAWNED PROCESS */
if(blocking_mode){
sleep(fixed_sleep);
/* wait until app completes: */
while(!done_flag){
sleep(fixed_sleep);
temp_counter++;
if(temp_counter*fixed_sleep >= sleep_max) {
if(verbose_print) printf("broke out early\n");
break;
}
}
if(verbose_print){
if(done_flag == true) {
printf("done_flag was set to true!\n");
}
}
}
done:
/* fence first */
retval = fence_helper();
if(retval != PMIX_SUCCESS){
if(verbose_print) printf("error fencing, finalize may fail ! \n");
}
/* finalize */
PMIx_Deregister_event_handler(_g_errhandler_ref, NULL, NULL);
if(verbose_print){
fprintf(stdout,
"spawn master process (rank %d) (host %s) finalizing\n",
main_proc.rank,
hostn);
}
/* clean up pmix */
retval = PMIx_tool_finalize();
if(retval == PMIX_SUCCESS)
{
if(verbose_print){
printf("spawn master process %d finalize success\n\n",
main_proc.rank);
}
}
else
{
printf("spawn master process %d pmix_finalize FAILURE: %d\n\n",
main_proc.rank,
retval);
}
retval = PMIx_Finalize(NULL, 0);
fflush(stdout);
/* cleanup before returning */
PMIX_INFO_FREE(job_info, job_info_count);
spawned_app->argv = NULL;
PMIX_APP_FREE(spawned_app, 1);
if(verbose_print) printf("%s exiting cleanly :)\n", argv[0]);
return 0;
}
int main(int argc, char **argv)
{
pmix_status_t rc;
pmix_proc_t myproc;
pmix_info_t *info;
pmix_app_t *app;
size_t ninfo, napps;
/* check for user directives - this would include:
* - a flag indicating we want to attach to a specified application
* - application info if we are launching a new app
*/
/* init us - if a PMIx server pid was provided, then pass it along */
if (0 < server_pid) {
ninfo = 1;
PMIX_INFO_CREATE(info, ninfo);
PMIX_INFO_LOAD(&info[0], PMIX_SERVER_PIDINFO, server_pid, PMIX_UINT32);
} else {
info = NULL;
ninfo = 0;
}
if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, info, ninfo))) {
fprintf(stderr, "PMIx_tool_init failed: %d\n", rc);
exit(rc);
}
if (0 < ninfo) {
PMIX_INFO_FREE(info, ninfo);
}
fprintf(stderr, "Tool ns %s rank %d: Running\n", myproc.nspace, myproc.rank);
/* if we are attaching to a running job, then attach to it */
if (attach) {
ret = attach_to_running_job(argv[1]);
} else {
/* this is an initial launch - we need to launch the application
* plus the debugger daemons, letting the RM know we are debugging
* so that it will "pause" the app procs until we are ready */
napps = 2;
PMIX_APP_CREATE(app, napps);
/* setup the executable */
app[0].cmd = strdup("app");
app[0].argc = 1;
app[0].argv = (char**)malloc(2*sizeof(char*));
app[0].argv[0] = strdup("app");
app[0].argv[1] = NULL;
/* provide directives so the apps do what the user requested */
ninfo = 2;
PMIX_INFO_CREATE(app[0].info, ninfo);
PMIX_INFO_LOAD(&app[0].info[0], PMIX_NP, 128, PMIX_UINT64);
PMIX_INFO_LOAD(&app[0].info[0], PMIX_MAPBY, "slot", PMIX_STRING);
/* setup the name of the daemon executable to launch */
app[1].cmd = strdup("debuggerdaemon");
app[1].argc = 1;
app[1].argv = (char**)malloc(2*sizeof(char*));
app[1].argv[0] = strdup("debuggerdaemon");
app[1].argv[1] = NULL;
/* provide directives so the daemons go where we want, and
* let the RM know these are debugger daemons */
ninfo = 2;
PMIX_INFO_CREATE(app[1].info, ninfo);
PMIX_INFO_LOAD(&app[1].info[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING); // instruct the RM to launch one copy of the executable on each node
PMIX_INFO_LOAD(&app[1].info[1], PMIX_DEBUGGER_DAEMONS, true, PMIX_BOOL); // these are debugger daemons
/* spawn the daemons */
PMIx_Spawn(NULL, 0, app, napps, dspace);
/* cleanup */
PMIX_APP_FREE(app, napps);
/* this is where a debugger tool would wait until the debug operation is complete */
}
done:
PMIx_tool_finalize(NULL, 0);
return(ret);
}
