static void handlePSMsg(DDTypedBufferMsg_t *msg)
{
if (msg->header.dest == PSC_getMyTID()) {
/* message for me, let's get infos and forward to all accounters */
mdbg(PSACC_LOG_ACC_MSG, "%s: got msg %s\n", __func__,
getAccountMsgType(msg->type));
switch (msg->type) {
case PSP_ACCOUNT_QUEUE:
case PSP_ACCOUNT_DELETE:
case PSP_ACCOUNT_SLOTS:
case PSP_ACCOUNT_START:
/* nothing to do here for me */
break;
case PSP_ACCOUNT_LOG:
handleAccountLog(msg);
break;
case PSP_ACCOUNT_CHILD:
handleAccountChild(msg);
break;
case PSP_ACCOUNT_END:
handleAccountEnd(msg);
break;
default:
mlog("%s: invalid msg type %i sender %s\n", __func__, msg->type,
PSC_printTID(msg->header.sender));
}
mdbg(PSACC_LOG_VERBOSE, "%s: msg type %s sender %s\n", __func__,
getAccountMsgType(msg->type), PSC_printTID(msg->header.sender));
}
/* forward msg to accounting daemons */
if (origHandler) origHandler((DDBufferMsg_t *) msg);
}
/**
* @brief Handle signals
*
* Handle the signal @a sig received. For the time being only SIGTERM
* is expected.
*
* @param sig Signal to handle
*
* @return No return value
*/
static void sighandler(int sig)
{
switch(sig) {
case SIGTERM:
if (sigVerbose) fprintf(stderr, "Got %s\n", strsignal(sig));
DDSignalMsg_t msg = {
.header = {
.type = PSP_CD_WHODIED,
.dest = 0,
.sender = PSC_getMyTID(),
.len = sizeof(msg) },
.signal = sig };
if (PSI_sendMsg(&msg) < 0) {
fprintf(stderr, "%s: PSI_sendMsg(): %m\n", __func__);
}
break;
default:
if (sigVerbose) fprintf(stderr, "Got signal %s\n", strsignal(sig));
}
* Send information on the environment variable @a key to task @a
* dest. The information (i.e. name and value) is sent as a character
* string.
*
* @param dest Task ID of the destination to send info to
*
* @param key Name of the environment variable to send
*
* @return No return value
*/
static void sendSingleEnv(PStask_ID_t dest, char *key)
{
DDTypedBufferMsg_t msg = {
.header = {
.type = PSP_CD_INFORESPONSE,
.sender = PSC_getMyTID(),
.dest = dest,
.len = sizeof(msg.header) + sizeof(msg.type) },
.type = PSP_INFO_QUEUE_ENVS,
.buf = {0}};
size_t strLen, bufLen = sizeof(msg.buf);
char *envStr = getenv(key);
if (envStr) {
strLen = snprintf(msg.buf, bufLen, "%s=%s", key, envStr);
} else {
strLen = snprintf(msg.buf, bufLen, "%s=<NULL>", key);
}
if (strLen > bufLen) {
msg.buf[bufLen-4] = '.';
msg.buf[bufLen-3] = '.';
int main(int argc, const char *argv[])
{
int np, verbose;
int rank, i, rc;
int dup_argc;
char **dup_argv;
int waittime, killtime;
unsigned int magic;
/*
* We can't use popt for argument parsing here. popt is not
* capable to stop at the first unrecogniced option, i.e. at the
* executable separation options to the mpirun command from
* options to the application.
*/
poptContext optCon; /* context for parsing command-line options */
struct poptOption optionsTable[] = {
{ "np", '\0', POPT_ARG_INT | POPT_ARGFLAG_ONEDASH,
&np, 0, "number of processes to start", "num"},
{ "wait", 'w', POPT_ARG_INT, &waittime, 0,
"Wait <n> seconds between each spawning step", "n"},
{ "kill", 'k', POPT_ARG_INT, &killtime, 0,
"Kill all processes <n> seconds after the first exits", "n"},
{ NULL, 'v', POPT_ARG_NONE,
&verbose, 0, "verbose mode", NULL},
{ NULL, '\0', 0, NULL, 0, NULL, NULL}
};
/* The duplicated argv will contain the apps commandline */
poptDupArgv(argc, argv, &dup_argc, (const char ***)&dup_argv);
optCon = poptGetContext(NULL, dup_argc, (const char **)dup_argv,
optionsTable, 0);
poptSetOtherOptionHelp(optCon, OTHER_OPTIONS_STR);
/*
* Split the argv into two parts:
* - first one containing the mpirun options
* - second one containing the apps argv
* The first one is already parsed while splitting
*/
while (1) {
const char *unknownArg;
np = -1;
verbose = 0;
waittime = 0;
killtime = -1;
rc = poptGetNextOpt(optCon);
if ((unknownArg=poptGetArg(optCon))) {
/*
* Find the first unknown argument (which is the apps
* name) within dup_argv. Start searching from dup_argv's end
* since the apps name might be used within another
* options argument.
*/
for (i=argc-1; i>0; i--) {
if (strcmp(dup_argv[i], unknownArg)==0) {
dup_argc = i;
dup_argv[dup_argc] = NULL;
poptFreeContext(optCon);
optCon = poptGetContext(NULL,
dup_argc, (const char **)dup_argv,
optionsTable, 0);
poptSetOtherOptionHelp(optCon, OTHER_OPTIONS_STR);
break;
}
}
if (i==0) {
printf("unknownArg '%s' not found !?\n", unknownArg);
exit(1);
}
} else {
/* No unknownArg left, we are finished */
break;
}
}
if (rc < -1) {
/* an error occurred during option processing */
poptPrintUsage(optCon, stderr, 0);
fprintf(stderr, "%s: %s\n",
poptBadOption(optCon, POPT_BADOPTION_NOALIAS),
poptStrerror(rc));
exit(1);
}
if (np == -1) {
poptPrintUsage(optCon, stderr, 0);
fprintf(stderr, "You have to give at least the -np argument.\n");
exit(1);
}
if (!argv[dup_argc]) {
poptPrintUsage(optCon, stderr, 0);
fprintf(stderr, "No <command> specified.\n");
exit(1);
}
free(dup_argv);
if (verbose) {
printf("The 'gmspawner' command-line is:\n");
for (i=0; i<dup_argc; i++) {
printf("%s ", argv[i]);
}
printf("\b\n\n");
printf("The applications command-line is:\n");
for (i=dup_argc; i<argc; i++) {
printf("%s ", argv[i]);
}
printf("\b\n\n");
}
/* init PSI */
if (!PSI_initClient(TG_GMSPAWNER)) {
fprintf(stderr, "Initialization of PSI failed.");
exit(10);
}
PSI_infoInt(-1, PSP_INFO_TASKRANK, NULL, &rank, 0);
if (rank != np) {
fprintf(stderr, "%s: rank(%d) != np(%d).\n", argv[dup_argc], rank, np);
exit(1);
}
/* Propagate some environment variables */
PSI_propEnv();
PSI_propEnvList("PSI_EXPORTS");
PSI_propEnvList("__PSI_EXPORTS");
srandom(time(NULL));
magic = random()%9999999;
setIntEnv("GMPI_MAGIC", magic);
setIntEnv("GMPI_NP", np);
{
char hostname[256];
gethostname(hostname, sizeof(hostname));
setPSIEnv("GMPI_MASTER", hostname, 1);
}
{
int port = createListener(8000, np, magic, verbose);
if (port>=0) setIntEnv("GMPI_PORT", port);
}
propagateEnv("GMPI_SHMEM", 1);
propagateEnv("DISPLAY", 0);
propagateEnv("GMPI_EAGER", 0);
propagateEnv("GMPI_RECV", 1);
PSC_setSigHandler(SIGALRM, sighandler);
/* spawn all processes */
PSI_RemoteArgs(argc - dup_argc, (char **)&argv[dup_argc],
&dup_argc, &dup_argv);
for (rank=0; rank<np; rank++) {
if (waittime && rank) sleep(waittime);
setIntEnv("GMPI_ID", rank);
setIntEnv("GMPI_BOARD", -1);
char slavestring[20];
PSnodes_ID_t node;
struct in_addr ip;
int err;
err = PSI_infoNodeID(-1, PSP_INFO_RANKID, &rank, &node, 1);
if (err) {
fprintf(stderr, "Could not determine rank %d's node.\n", rank);
exit(1);
}
err = PSI_infoUInt(-1, PSP_INFO_NODE, &node, &ip.s_addr, 1);
if (err) {
fprintf(stderr, "Could not determine node %d's IP\n", node);
exit(1);
}
snprintf(slavestring, sizeof(slavestring), "%s", inet_ntoa(ip));
setPSIEnv("GMPI_SLAVE", slavestring, 1);
/* spawn the process */
int error;
if (!PSI_spawnRank(rank, ".", dup_argc, dup_argv, &error)) {
if (error) {
char *errstr = strerror(error);
fprintf(stderr, "Could not spawn process %d (%s) error = %s.\n",
rank, dup_argv[0], errstr ? errstr : "UNKNOWN");
exit(1);
}
}
}
/* Wait for the spawned processes to complete */
while (np) {
static int firstClient=1;
DDErrorMsg_t msg;
int ret;
ret = PSI_recvMsg((DDMsg_t *)&msg, sizeof(msg));
if (msg.header.type != PSP_CD_SPAWNFINISH || ret != sizeof(msg)) {
fprintf(stderr, "got strange message type %s\n",
PSP_printMsg(msg.header.type));
} else {
if (firstClient && killtime!=-1) {
// printf("Alarm set to %d\n", killtime);
if (killtime) {
alarm(killtime);
firstClient=0;
} else {
/* Stop immediately */
exit(0);
}
}
np--;
// printf("%d clients left\n", np);
}
}
PSI_release(PSC_getMyTID());
PSI_exitClient();
return 0;
}
int main(int argc, char **argv)
{
int rank = 0;
command_name = argv[0];
parse_opt(argc, argv);
if (arg_version) {
printVersion();
return 0;
}
if (arg_cp) {
unsigned int slen = strlen(copy_command_src) + strlen(arg_cp) + 1000;
char *icmd = malloc(slen);
snprintf(icmd, slen,"%s %s", copy_command_src, arg_cp);
arg_icmd = icmd;
arg_ocmd = copy_command_dest;
fprintf(stderr, " input command: %s\n", arg_icmd);
}
if (arg_icmd && arg_ifile) {
fprintf(stderr, "Warning: Ignoring -icmd %s, because -i is set\n",
arg_icmd);
arg_icmd = NULL;
}
if (arg_ocmd && arg_ofile) {
fprintf(stderr, "Warning: Ignoring -ocmd %s, because -o is set\n",
arg_ocmd);
arg_ocmd = NULL;
}
if (PSP_Init() != PSP_OK) {
fprintf(stderr, "PSP_Init() failed!\n");
exit(1);
}
if (!arg_manual) {
/* HACK HACK HACK */
setenv("__PSI_MASTERNODE", "-1", 0);
setenv("__PSI_MASTERPORT", "-1", 0);
PSE_initialize();
rank = PSE_getRank();
} else if (arg_server) {
rank = 1;
}
if (rank < 0) {
/* original process, let's spawn rank 0 and become logger */
setenv("PSI_NOMSGLOGGERDONE", "", 1);
PSE_spawnAdmin(-1, 0, argc, argv, 0);
} else if (rank == 0) {
/* server */
if (arg_progress) set_timer(1, 1 ,timer);
doServer();
} else {
/* client */
if (arg_progress && arg_manual) set_timer(1, 1 ,timer);
doClient();
}
if (!arg_manual) {
PSI_release(PSC_getMyTID());
}
if ((!rank || arg_manual) && arg_progress) {
print_stat(1);
}
return 0;
}
