static int init(void)
{
char cwd[OPAL_PATH_MAX];
int rc;
OPAL_OUTPUT_VERBOSE((5, orcm_pvsn_base_framework.framework_output,
"%s pvsn:wwulf:init",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if (OPAL_SUCCESS != (rc = opal_getcwd(cwd, OPAL_PATH_MAX))) {
return rc;
}
/* check to see if we can execute wwsh */
cmd = opal_path_findv("wwsh", X_OK, environ, cwd);
if (NULL == cmd) {
return ORTE_ERR_EXE_NOT_FOUND;
}
OPAL_OUTPUT_VERBOSE((5, orcm_pvsn_base_framework.framework_output,
"%s pvsn:wwulf:init path to wwsh %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), cmd));
return ORCM_SUCCESS;
}
static void check_debugger(int fd, short event, void *arg)
{
struct timeval now;
opal_event_t *tmp = (opal_event_t*)arg;
orte_job_t *jdata;
orte_app_context_t *app;
char cwd[OPAL_PATH_MAX];
int rc;
int32_t ljob;
if (MPIR_being_debugged) {
if (orte_debug_flag) {
opal_output(0, "%s Launching debugger %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
MPIR_executable_path);
}
/* a debugger has attached! All the MPIR_Proctable
* data is already available, so we only need to
* check to see if we should spawn any daemons
*/
if ('\0' != MPIR_executable_path[0]) {
/* this will be launched just like a regular job,
* so we do not use the global orte_debugger_daemon
* as this is reserved for co-location upon startup
*/
jdata = OBJ_NEW(orte_job_t);
/* create a jobid for these daemons - this is done solely
* to avoid confusing the rest of the system's bookkeeping
*/
orte_plm_base_create_jobid(jdata);
/* flag the job as being debugger daemons */
jdata->controls |= ORTE_JOB_CONTROL_DEBUGGER_DAEMON;
/* unless directed, we do not forward output */
if (!MPIR_forward_output) {
jdata->controls &= ~ORTE_JOB_CONTROL_FORWARD_OUTPUT;
}
/* set the mapping policy to "pernode" so we only get
* one debugger daemon on each node
*/
jdata->map = OBJ_NEW(orte_job_map_t);
jdata->map->npernode = 1;
/* add it to the global job pool */
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* create an app_context for the debugger daemon */
app = OBJ_NEW(orte_app_context_t);
app->app = strdup((char*)MPIR_executable_path);
if (OPAL_SUCCESS != (rc = opal_getcwd(cwd, sizeof(cwd)))) {
orte_show_help("help-orterun.txt", "orterun:init-failure",
true, "get the cwd", rc);
OBJ_RELEASE(jdata);
goto RELEASE;
}
app->cwd = strdup(cwd);
app->user_specified_cwd = false;
opal_argv_append_nosize(&app->argv, app->app);
build_debugger_args(app);
opal_pointer_array_add(jdata->apps, &app->super);
jdata->num_apps = 1;
/* now go ahead and spawn this job */
if (ORTE_SUCCESS != (rc = orte_plm.spawn(jdata))) {
ORTE_ERROR_LOG(rc);
}
}
RELEASE:
/* notify the debugger that all is ready */
MPIR_Breakpoint();
} else {
/* reissue the timer to wake us up again */
now.tv_sec = orte_debugger_check_rate;
now.tv_usec = 0;
opal_evtimer_add(tmp, &now);
}
}
int main(int argc, char *argv[])
{
int32_t ret, i;
opal_cmd_line_t cmd_line;
char **inpt;
opal_buffer_t *buf;
int count;
char cwd[OPAL_PATH_MAX];
orcm_tool_cmd_t flag = ORCM_TOOL_STOP_CMD;
int32_t master=0;
uint16_t jfam=0;
/***************
* Initialize
***************/
/*
* Make sure to init util before parse_args
* to ensure installdirs is setup properly
* before calling mca_base_open();
*/
if( ORTE_SUCCESS != (ret = orcm_init_util()) ) {
return ret;
}
/* initialize the globals */
my_globals.help = false;
my_globals.replicas = NULL;
my_globals.sched = NULL;
my_globals.hnp_uri = NULL;
/* Parse the command line options */
opal_cmd_line_create(&cmd_line, cmd_line_opts);
mca_base_open();
mca_base_cmd_line_setup(&cmd_line);
ret = opal_cmd_line_parse(&cmd_line, true, argc, argv);
/* extract the MCA/GMCA params */
mca_base_cmd_line_process_args(&cmd_line, &environ, &environ);
/**
* Now start parsing our specific arguments
*/
if (OPAL_SUCCESS != ret || my_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(&cmd_line);
orte_show_help("help-orcm-stop.txt", "usage", true, args);
free(args);
return ORTE_ERROR;
}
if (NULL != my_globals.sched) {
if (0 == strncmp(my_globals.sched, "file", strlen("file")) ||
0 == strncmp(my_globals.sched, "FILE", strlen("FILE"))) {
char input[1024], *filename;
FILE *fp;
/* it is a file - get the filename */
filename = strchr(my_globals.sched, ':');
if (NULL == filename) {
/* filename is not correctly formatted */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-bad", true, "scheduler", my_globals.sched);
return ORTE_ERROR;
}
++filename; /* space past the : */
if (0 >= strlen(filename)) {
/* they forgot to give us the name! */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-bad", true, "scheduler", my_globals.sched);
return ORTE_ERROR;
}
/* open the file and extract the pid */
fp = fopen(filename, "r");
if (NULL == fp) { /* can't find or read file! */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-access", true, "scheduler", filename);
return ORTE_ERROR;
}
if (NULL == fgets(input, 1024, fp)) {
/* something malformed about file */
fclose(fp);
orte_show_help("help-openrcm-runtime.txt", "hnp-file-bad", "scheduler", true, filename);
return ORTE_ERROR;
}
fclose(fp);
input[strlen(input)-1] = '\0'; /* remove newline */
/* convert the pid */
master = strtoul(input, NULL, 10);
} else {
/* should just be the master itself */
master = strtoul(my_globals.sched, NULL, 10);
}
}
/* if we were given HNP contact info, parse it and
* setup the process_info struct with that info
*/
if (NULL != my_globals.hnp_uri) {
if (0 == strncmp(my_globals.hnp_uri, "file", strlen("file")) ||
0 == strncmp(my_globals.hnp_uri, "FILE", strlen("FILE"))) {
char input[1024], *filename;
FILE *fp;
/* it is a file - get the filename */
filename = strchr(my_globals.hnp_uri, ':');
if (NULL == filename) {
/* filename is not correctly formatted */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-bad", true, "uri", my_globals.hnp_uri);
goto cleanup;
}
++filename; /* space past the : */
if (0 >= strlen(filename)) {
/* they forgot to give us the name! */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-bad", true, "uri", my_globals.hnp_uri);
goto cleanup;
}
/* open the file and extract the uri */
fp = fopen(filename, "r");
if (NULL == fp) { /* can't find or read file! */
orte_show_help("help-openrcm-runtime.txt", "hnp-filename-access", true, filename);
goto cleanup;
}
if (NULL == fgets(input, 1024, fp)) {
/* something malformed about file */
fclose(fp);
orte_show_help("help-openrcm-runtime.txt", "hnp-file-bad", true, filename);
goto cleanup;
}
fclose(fp);
input[strlen(input)-1] = '\0'; /* remove newline */
/* put into the process info struct */
orte_process_info.my_hnp_uri = strdup(input);
} else {
/* should just be the uri itself */
orte_process_info.my_hnp_uri = strdup(my_globals.hnp_uri);
}
}
if (OPAL_SUCCESS != opal_getcwd(cwd, sizeof(cwd))) {
opal_output(orte_clean_output, "failed to get cwd\n");
return ORTE_ERR_NOT_FOUND;
}
/***************************
* We need all of OPAL and ORTE - this will
* automatically connect us to the CM
***************************/
if (ORTE_SUCCESS != orcm_init(ORCM_TOOL)) {
orcm_finalize();
return 1;
}
/* if we were given the hnp uri, extract the job family for the
* master id
*/
if (NULL != my_globals.hnp_uri) {
master = ORTE_JOB_FAMILY(ORTE_PROC_MY_HNP->jobid);
}
/* register to receive responses */
if (ORCM_SUCCESS != (ret = orcm_pnp.register_receive("orcm-stop", "0.1", "alpha",
ORCM_PNP_GROUP_INPUT_CHANNEL,
ORCM_PNP_TAG_TOOL,
ack_recv, NULL))) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
/* announce my existence */
if (ORCM_SUCCESS != (ret = orcm_pnp.announce("orcm-stop", "0.1", "alpha", NULL))) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
/* setup the buffer to send our cmd */
buf = OBJ_NEW(opal_buffer_t);
/* indicate the scheduler to be used */
jfam = master & 0x0000ffff;
opal_dss.pack(buf, &jfam, 1, OPAL_UINT16);
/* get the apps to stop */
inpt = NULL;
opal_cmd_line_get_tail(&cmd_line, &count, &inpt);
if (0 == count) {
/* if no apps were given, then we stop the entire
* DVM itself by telling the daemon's to terminate
*/
if (ORCM_SUCCESS != (ret = orcm_pnp.output_nb(ORCM_PNP_SYS_CHANNEL,
NULL, ORCM_PNP_TAG_TERMINATE,
NULL, 0, buf, cbfunc, NULL))) {
ORTE_ERROR_LOG(ret);
}
goto cleanup;
} else {
/* load the stop cmd */
opal_dss.pack(buf, &flag, 1, ORCM_TOOL_CMD_T);
/* for each app */
for (i=0; NULL != inpt[i]; i++) {
opal_dss.pack(buf, &inpt[i], 1, OPAL_STRING);
/* pack the replicas to be stopped */
opal_dss.pack(buf, &my_globals.replicas, 1, OPAL_STRING);
}
opal_argv_free(inpt);
if (ORCM_SUCCESS != (ret = orcm_pnp.output_nb(ORCM_PNP_SYS_CHANNEL,
NULL, ORCM_PNP_TAG_TOOL,
NULL, 0, buf, cbfunc, NULL))) {
ORTE_ERROR_LOG(ret);
}
}
/* now wait for ack */
opal_event_dispatch(opal_event_base);
/***************
* Cleanup
***************/
cleanup:
orcm_finalize();
return ret;
}
