int main(int argc, char *argv[])
{
int32_t ret;
opal_cmd_line_t cmd_line;
char *args = NULL;
/***************
* 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.monitor = false;
my_globals.update_rate = 5;
my_globals.sched = 0;
/* 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) {
args = opal_cmd_line_get_usage_msg(&cmd_line);
orte_show_help("help-orcm-ps.txt", "usage", true, args);
free(args);
return ORTE_ERROR;
}
/***************************
* We need all of OPAL and ORTE - this will
* automatically connect us to the CM
***************************/
if (ORCM_SUCCESS != (ret = orcm_init(ORCM_TOOL))) {
goto cleanup;
}
/* register to receive responses */
if (ORCM_SUCCESS != (ret = orcm_pnp.register_receive("orcm-ps", "0.1", "alpha",
ORCM_PNP_GROUP_INPUT_CHANNEL,
ORCM_PNP_TAG_PS,
ps_recv, NULL))) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
/* announce my existence */
if (ORCM_SUCCESS != (ret = orcm_pnp.announce("orcm-ps", "0.1", "alpha", NULL))) {
ORTE_ERROR_LOG(ret);
goto cleanup;
}
/* define an event to signal completion */
if (pipe(rel_pipe) < 0) {
opal_output(0, "Cannot open release pipe");
goto cleanup;
}
opal_event_set(opal_event_base, &rel_ev, rel_pipe[0],
OPAL_EV_READ, process_release, NULL);
opal_event_add(&rel_ev, 0);
/* we know we need to print the data once */
update_data(0, 0, NULL);
opal_event_dispatch(opal_event_base);
/***************
* Cleanup
***************/
cleanup:
/* cleanup orcm */
orcm_finalize();
return ret;
}
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;
}
int orcm_init(orcm_proc_type_t flags)
{
int ret;
char *error;
int i, spin;
if (NULL != getenv("ORCM_MCA_spin")) {
spin = 1;
/* spin until a debugger can attach */
while (0 != spin) {
ret = 0;
while (ret < 10000) {
ret++;
};
}
}
if (!orcm_util_initialized) {
orcm_init_util();
}
/* set the default leader policy */
orcm_default_leader_policy.jobid = ORTE_JOBID_WILDCARD;
orcm_default_leader_policy.vpid = ORTE_VPID_WILDCARD;
/* get the number of max msgs */
mca_base_param_reg_int_name("orcm", "max_buffered_msgs",
"Number of recvd messages to hold in storage from each source",
false, false, ORCM_MAX_MSG_RING_SIZE, &orcm_max_msg_ring_size);
/* independent mode or not */
mca_base_param_reg_int_name("orcm", "sched_kill_dvm",
"Whether or not scheduler kills associated daemons upon termination (default: no)",
false, false, (int)false, &ret);
orcm_sched_kill_dvm = OPAL_INT_TO_BOOL(ret);
/* setup the globals that require initialization */
orcm_triplets = OBJ_NEW(orcm_triplets_array_t);
#ifdef HAVE_QSYSTEM_H
#ifdef Q_SYSTEM_INTFCS_TO_PROBE_FOR_IP_ADDRESS
{
char *eth_ifs[] = Q_SYSTEM_INTFCS_TO_PROBE_FOR_IP_ADDRESS;
char **adds=NULL, *ifs, *envar;
int i, num_ifs;
num_ifs = sizeof(eth_ifs) / sizeof(eth_ifs[0]);
for (i=0; i < num_ifs; i++) {
opal_argv_append_nosize(&adds, eth_ifs[i]);
}
ifs = opal_argv_join(adds, ',');
opal_argv_free(adds);
/* push it into the environ so that the rmcast framework can get it */
asprintf(&envar, "OMPI_MCA_rmcast_base_if_include=%s", ifs);
putenv(envar);
/* cannot release envar as the environ doesn't keep its own copy */
free(ifs);
}
#endif
#endif
/* initialize us */
if (ORTE_SUCCESS != (ret = orte_init(NULL, NULL, flags))) {
error = "orte_init";
goto error;
}
if (!ORCM_PROC_IS_TOOL) {
opal_set_using_threads(true);
}
if (!ORCM_PROC_IS_APP) {
trap_signals();
}
orcm_initialized = true;
return ORCM_SUCCESS;
error:
if (ORCM_ERR_SILENT != ret) {
orte_show_help("help-openrcm-runtime.txt",
"orcm_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ret;
}
