int smpd_post_abort_command(char *fmt, ...)
{
int result;
char error_str[2048] = "";
smpd_command_t *cmd_ptr;
smpd_context_t *context;
va_list list;
smpd_enter_fn(FCNAME);
va_start(list, fmt);
vsnprintf(error_str, 2048, fmt, list);
va_end(list);
result = smpd_create_command("abort", smpd_process.id, 0, SMPD_FALSE, &cmd_ptr);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create an abort command.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
result = smpd_add_command_arg(cmd_ptr, "error", error_str);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to add the error string to the abort command.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
result = smpd_command_destination(0, &context);
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to find destination for command...Aborting: %s\n", error_str);
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (context == NULL)
{
if (smpd_process.left_context == NULL)
{
printf("Aborting: %s\n", error_str);
fflush(stdout);
smpd_exit_fn(FCNAME);
smpd_exit(-1);
}
smpd_process.closing = SMPD_TRUE;
result = smpd_create_command("close", 0, 1, SMPD_FALSE, &cmd_ptr);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create the close command to tear down the job tree.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
result = smpd_post_write_command(smpd_process.left_context, cmd_ptr);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to post a write of the close command to tear down the job tree as part of the abort process.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
}
else
{
smpd_dbg_printf("sending abort command to %s context: \"%s\"\n", smpd_get_context_str(context), cmd_ptr->cmd);
result = smpd_post_write_command(context, cmd_ptr);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to post a write of the abort command to the %s context.\n", smpd_get_context_str(context));
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
}
smpd_exit_fn(FCNAME);
return SMPD_SUCCESS;
}
int mpiexec_rsh()
{
int i;
smpd_launch_node_t *launch_node_ptr;
smpd_process_t *process, **processes;
int result;
char *iter1, *iter2;
char exe[SMPD_MAX_EXE_LENGTH];
char *p;
char ssh_cmd[100] = "ssh -x";
SMPDU_Sock_set_t set;
SMPD_BOOL escape_escape = SMPD_TRUE;
char *env_str;
int maxlen;
SMPDU_Sock_t abort_sock;
smpd_context_t *abort_context = NULL;
smpd_command_t *cmd_ptr;
PROCESS_HANDLE_TYPE hnd;
smpd_enter_fn("mpiexec_rsh");
#ifdef HAVE_WINDOWS_H
SetConsoleCtrlHandler(mpiexec_rsh_handler, TRUE);
#else
/* setup a signall hander? */
#endif
p = getenv("MPIEXEC_RSH");
if (p != NULL && strlen(p) > 0){
strncpy(ssh_cmd, p, 100);
}
p = getenv("MPIEXEC_RSH_NO_ESCAPE");
if (p != NULL){
if (smpd_is_affirmative(p) || strcmp(p, "1") == 0){
escape_escape = SMPD_FALSE;
}
}
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create a set for the mpiexec_rsh.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_process.nproc = smpd_process.launch_list->nproc;
if (smpd_process.use_pmi_server){
result = start_pmi_server(smpd_process.launch_list->nproc, root_host, 100, &root_port);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh is unable to start the local pmi server.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_dbg_printf("the pmi server is listening on %s:%d\n", root_host, root_port);
}
else{
/* start the root smpd */
result = start_root_smpd(root_host, SMPD_MAX_HOST_LENGTH, &root_port, &hnd);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh is unable to start the root smpd.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_dbg_printf("the root smpd is listening on %s:%d\n", root_host, root_port);
/* create a connection to the root smpd used to abort the job */
result = ConnectToHost(root_host, root_port, SMPD_CONNECTING_RPMI, set, &abort_sock, &abort_context);
if (result != SMPD_SUCCESS){
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
processes = (smpd_process_t**)MPIU_Malloc(sizeof(smpd_process_t*) * smpd_process.launch_list->nproc);
if (processes == NULL){
smpd_err_printf("unable to allocate process array.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
launch_node_ptr = smpd_process.launch_list;
for (i=0; i<smpd_process.launch_list->nproc; i++){
if (launch_node_ptr == NULL){
smpd_err_printf("Error: not enough launch nodes.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* initialize process structure */
result = smpd_create_process_struct(i, &process);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create a process structure.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* no need for a pmi context */
if (process->pmi){
smpd_free_context(process->pmi);
}
process->pmi = NULL;
/* change stdout and stderr to rsh behavior:
* write stdout/err directly to stdout/err instead of creating
* an smpd stdout/err command
*/
if (process->out != NULL){
process->out->type = SMPD_CONTEXT_STDOUT_RSH;
}
if (process->err != NULL){
process->err->type = SMPD_CONTEXT_STDERR_RSH;
}
MPIU_Strncpy(process->clique, launch_node_ptr->clique, SMPD_MAX_CLIQUE_LENGTH);
MPIU_Strncpy(process->dir, launch_node_ptr->dir, SMPD_MAX_DIR_LENGTH);
MPIU_Strncpy(process->domain_name, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
MPIU_Strncpy(process->env, launch_node_ptr->env, SMPD_MAX_ENV_LENGTH);
if (escape_escape == SMPD_TRUE && smpd_process.mpiexec_run_local != SMPD_TRUE){
/* convert \ to \\ to make cygwin ssh happy */
iter1 = launch_node_ptr->exe;
iter2 = exe;
while (*iter1){
if (*iter1 == '\\'){
*iter2 = *iter1;
iter2++;
*iter2 = *iter1;
}
else{
*iter2 = *iter1;
}
iter1++;
iter2++;
}
*iter2 = '\0';
/*printf("[%s] -> [%s]\n", launch_node_ptr->exe, exe);*/
}
else{
MPIU_Strncpy(exe, launch_node_ptr->exe, SMPD_MAX_EXE_LENGTH);
}
/* Two samples for testing on the local machine */
/* static rPMI initialization */
/*sprintf(process->exe, "env PMI_RANK=%d PMI_SIZE=%d PMI_KVS=%s PMI_ROOT_HOST=%s PMI_ROOT_PORT=8888 PMI_ROOT_LOCAL=1 PMI_APPNUM=%d %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, launch_node_ptr->appnum, exe);*/
/* dynamic rPMI initialization */
/*sprintf(process->exe, "env PMI_RANK=%d PMI_SIZE=%d PMI_KVS=%s PMI_ROOT_HOST=%s PMI_ROOT_PORT=%d PMI_ROOT_LOCAL=0 PMI_APPNUM=%d %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, root_port, launch_node_ptr->appnum, exe);*/
if (smpd_process.mpiexec_run_local == SMPD_TRUE){
/* -localonly option and dynamic rPMI initialization */
env_str = &process->env[strlen(process->env)];
maxlen = (int)(SMPD_MAX_ENV_LENGTH - strlen(process->env));
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_RANK", launch_node_ptr->iproc);
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_SIZE", launch_node_ptr->nproc);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_KVS", smpd_process.kvs_name);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_ROOT_HOST", root_host);
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_ROOT_PORT", root_port);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_ROOT_LOCAL", "0");
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_APPNUM", launch_node_ptr->appnum);
MPIU_Strncpy(process->exe, exe, SMPD_MAX_EXE_LENGTH);
}
else{
/* ssh and dynamic rPMI initialization */
char fmtEnv[SMPD_MAX_ENV_LENGTH];
int fmtEnvLen = SMPD_MAX_ENV_LENGTH;
char *pExe = process->exe;
int curLen = 0;
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH, "%s %s env", ssh_cmd, launch_node_ptr->hostname);
curLen = strlen(process->exe);
pExe = process->exe + curLen;
if(FmtEnvVarsForSSH(launch_node_ptr->env, fmtEnv, fmtEnvLen)){
/* Add user specified env vars */
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH - curLen, "%s", fmtEnv);
curLen = strlen(process->exe);
pExe = process->exe + curLen;
}
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH - curLen, " \"PMI_RANK=%d\" \"PMI_SIZE=%d\" \"PMI_KVS=%s\" \"PMI_ROOT_HOST=%s\" \"PMI_ROOT_PORT=%d\" \"PMI_ROOT_LOCAL=0\" \"PMI_APPNUM=%d\" %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, root_port, launch_node_ptr->appnum, exe);
}
MPIU_Strncpy(process->kvs_name, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
process->nproc = launch_node_ptr->nproc;
MPIU_Strncpy(process->path, launch_node_ptr->path, SMPD_MAX_PATH_LENGTH);
/* call smpd_launch_process */
smpd_dbg_printf("launching: %s\n", process->exe);
result = smpd_launch_process(process, SMPD_DEFAULT_PRIORITY_CLASS, SMPD_DEFAULT_PRIORITY, SMPD_FALSE, set);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to launch process %d <%s>.\n", i, process->exe);
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* save the new process in the list */
process->next = smpd_process.process_list;
smpd_process.process_list = process;
if (i == 0){
/* start the stdin redirection thread to the first process */
setup_stdin_redirection(process, set);
}
smpd_process.nproc_launched++;
processes[i] = process;
launch_node_ptr = launch_node_ptr->next;
} /* for (i=0; i<smpd_process.launch_list->nproc; i++) */
if (launch_node_ptr != NULL){
smpd_err_printf("Error: too many launch nodes.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* Start the timeout mechanism if specified */
if (smpd_process.timeout > 0){
smpd_context_t *reader_context;
SMPDU_Sock_t sock_reader;
SMPDU_SOCK_NATIVE_FD reader, writer;
#ifdef HAVE_WINDOWS_H
/*SOCKET reader, writer;*/
smpd_make_socket_loop((SOCKET*)&reader, (SOCKET*)&writer);
#else
/*int reader, writer;*/
int pair[2];
socketpair(AF_UNIX, SOCK_STREAM, 0, pair);
reader = pair[0];
writer = pair[1];
#endif
result = SMPDU_Sock_native_to_sock(set, reader, NULL, &sock_reader);
result = SMPDU_Sock_native_to_sock(set, writer, NULL, &smpd_process.timeout_sock);
result = smpd_create_context(SMPD_CONTEXT_TIMEOUT, set, sock_reader, -1, &reader_context);
reader_context->read_state = SMPD_READING_TIMEOUT;
result = SMPDU_Sock_post_read(sock_reader, &reader_context->read_cmd.cmd, 1, 1, NULL);
#ifdef HAVE_WINDOWS_H
/* create a Windows thread to sleep until the timeout expires */
smpd_process.timeout_thread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)timeout_thread, NULL, 0, NULL);
if (smpd_process.timeout_thread == NULL){
printf("Error: unable to create a timeout thread, errno %d.\n", GetLastError());
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else /* HAVE_WINDOWS_H */
#ifdef SIGALRM
/* create an alarm to signal mpiexec when the timeout expires */
smpd_signal(SIGALRM, timeout_function);
alarm(smpd_process.timeout);
#else /* SIGALARM */
#ifdef HAVE_PTHREAD_H
/* create a pthread to sleep until the timeout expires */
result = pthread_create(&smpd_process.timeout_thread, NULL, timeout_thread, NULL);
if (result != 0){
printf("Error: unable to create a timeout thread, errno %d.\n", result);
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else /* HAVE_PTHREAD_H */
/* no timeout mechanism available */
#endif /* HAVE_PTHREAD_H */
#endif /* SIGALARM */
#endif /* HAVE_WINDOWS_H */
} /* if (smpd_process.timeout > 0) */
result = smpd_enter_at_state(set, SMPD_IDLE);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh state machine failed.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
if (smpd_process.use_pmi_server){
result = stop_pmi_server();
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh unable to stop the pmi server.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
else{
/* Send an abort command to the root_smpd thread/process to insure that it exits.
* This only needs to be sent when there is an error or failed process of some sort
* but it is safe to send it in all cases.
*/
result = smpd_create_command("abort", 0, 0, SMPD_FALSE, &cmd_ptr);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create an abort command.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
result = smpd_post_write_command(abort_context, cmd_ptr);
if (result != SMPD_SUCCESS){
/* Only print this as a debug message instead of an error because the root_smpd thread/process may have already exited. */
smpd_dbg_printf("unable to post a write of the abort command to the %s context.\n", smpd_get_context_str(abort_context));
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
result = stop_root_smpd(hnd);
if (result != PMI_SUCCESS){
smpd_err_printf("mpiexec_rsh unable to stop the root smpd.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
smpd_exit_fn("mpiexec_rsh");
return 0;
}
int main(int argc, char* argv[])
{
int result = SMPD_SUCCESS;
smpd_host_node_t *host_node_ptr;
smpd_launch_node_t *launch_node_ptr;
smpd_context_t *context;
SMPDU_Sock_set_t set;
SMPDU_Sock_t sock = SMPDU_SOCK_INVALID_SOCK;
smpd_state_t state;
smpd_enter_fn("main");
/* catch an empty command line */
if (argc < 2)
{
mp_print_options();
exit(0);
}
smpd_process.mpiexec_argv0 = argv[0];
/* initialize */
/* FIXME: Get rid of this hack - we already create
* local KVS for all singleton clients by default
*/
putenv("PMI_SMPD_FD=0");
result = PMPI_Init(&argc, &argv);
/* SMPD_CS_ENTER(); */
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPD_Init failed,\nerror: %d\n", result);
smpd_exit_fn("main");
return result;
}
result = SMPDU_Sock_init();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_init failed,\nsock error: %s\n",
get_sock_error_string(result));
smpd_exit_fn("main");
return result;
}
result = smpd_init_process();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("smpd_init_process failed.\n");
goto quit_job;
}
smpd_process.dbg_state = SMPD_DBG_STATE_ERROUT;
/* parse the command line */
smpd_dbg_printf("parsing the command line.\n");
result = mp_parse_command_args(&argc, &argv);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to parse the mpiexec command arguments.\n");
goto quit_job;
}
/* If we are using MS HPC job scheduler we only connect
* to the local SMPD
*/
if(smpd_process.use_ms_hpc){
char host[100];
int id;
/* Free the current host list */
result = smpd_free_host_list();
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to free the global host list\n");
goto quit_job;
}
/* Add local host to the host list */
result = smpd_get_hostname(host, 100);
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to get the local hostname\n");
goto quit_job;
}
result = smpd_get_host_id(host, &id);
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to get host id for local host\n");
goto quit_job;
}
/* Set the number of PMI procs since they are not launched by mpiexec */
smpd_process.nproc = smpd_process.launch_list->nproc;
smpd_dbg_printf("Adding (%s:%d) == (localhost) to the host list\n", host, id);
}
/* print and see what we've got */
/* debugging output *************/
smpd_dbg_printf("host tree:\n");
host_node_ptr = smpd_process.host_list;
if (!host_node_ptr)
smpd_dbg_printf("<none>\n");
while (host_node_ptr)
{
smpd_dbg_printf(" host: %s, parent: %d, id: %d\n",
host_node_ptr->host,
host_node_ptr->parent, host_node_ptr->id);
host_node_ptr = host_node_ptr->next;
}
smpd_dbg_printf("launch nodes:\n");
launch_node_ptr = smpd_process.launch_list;
if (!launch_node_ptr)
smpd_dbg_printf("<none>\n");
while (launch_node_ptr)
{
smpd_dbg_printf(" iproc: %d, id: %d, exe: %s\n",
launch_node_ptr->iproc, launch_node_ptr->host_id,
launch_node_ptr->exe);
launch_node_ptr = launch_node_ptr->next;
}
/* end debug output *************/
/* set the id of the mpiexec node to zero */
smpd_process.id = 0;
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_create_set failed,\nsock error: %s\n", get_sock_error_string(result));
goto quit_job;
}
smpd_process.set = set;
/* Check to see if the user wants to use a remote shell mechanism for launching the processes
* instead of using the smpd process managers.
*/
if (smpd_process.rsh_mpiexec == SMPD_TRUE)
{
/* Do rsh or localonly stuff */
result = mpiexec_rsh();
/* skip over the non-rsh code and go to the cleanup section */
goto quit_job;
}
/* Start the timeout mechanism if specified */
/* This code occurs after the rsh_mpiexec option check because the rsh code implementes timeouts differently */
if (smpd_process.timeout > 0)
{
#ifdef HAVE_WINDOWS_H
/* create a Windows thread to sleep until the timeout expires */
if (smpd_process.timeout_thread == NULL)
{
smpd_process.timeout_thread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)timeout_thread, NULL, 0, NULL);
if (smpd_process.timeout_thread == NULL)
{
printf("Error: unable to create a timeout thread, errno %d.\n", GetLastError());
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
}
#elif defined(SIGALRM)
/* create an alarm to signal mpiexec when the timeout expires */
smpd_signal(SIGALRM, timeout_function);
alarm(smpd_process.timeout);
#elif defined(HAVE_PTHREAD_H)
/* create a pthread to sleep until the timeout expires */
result = pthread_create(&smpd_process.timeout_thread, NULL, timeout_thread, NULL);
if (result != 0)
{
printf("Error: unable to create a timeout thread, errno %d.\n", result);
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else
/* no timeout mechanism available */
#endif
}
/* make sure we have a passphrase to authenticate connections to the smpds */
if (smpd_process.passphrase[0] == '\0')
smpd_get_smpd_data("phrase", smpd_process.passphrase, SMPD_PASSPHRASE_MAX_LENGTH);
if (smpd_process.passphrase[0] == '\0')
{
if (smpd_process.noprompt)
{
printf("Error: No smpd passphrase specified through the registry or .smpd file, exiting.\n");
result = SMPD_FAIL;
goto quit_job;
}
printf("Please specify an authentication passphrase for smpd: ");
fflush(stdout);
smpd_get_password(smpd_process.passphrase);
}
/* set the state to create a console session or a job session */
state = smpd_process.do_console ? SMPD_MPIEXEC_CONNECTING_SMPD : SMPD_MPIEXEC_CONNECTING_TREE;
result = smpd_create_context(SMPD_CONTEXT_LEFT_CHILD, set, sock, 1, &context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create a context for the first host in the tree.\n");
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
if (!smpd_process.local_root)
{
#endif
/* start connecting the tree by posting a connect to the first host */
result = SMPDU_Sock_post_connect(set, context, smpd_process.host_list->host, smpd_process.port, &sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to connect to '%s:%d',\nsock error: %s\n",
smpd_process.host_list->host, smpd_process.port, get_sock_error_string(result));
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
}
#endif
context->sock = sock;
context->state = state;
context->connect_to = smpd_process.host_list;
#ifdef HAVE_WINDOWS_H
if (smpd_process.local_root)
{
int port;
smpd_context_t *rc_context;
/* The local_root option is implemented by having mpiexec act as the smpd
* and launch the smpd manager. Then mpiexec connects to this manager just
* as if it had been created by a real smpd. This causes all the processes
* destined for the first smpd host to be launched by this child process of
* mpiexec and not the smpd service. This allows for these processes to
* create windows that are visible to the interactive user. It also means
* that the job cannot be run in the context of a user other than the user
* running mpiexec. */
/* get the path to smpd.exe because pszExe is currently mpiexec.exe */
smpd_get_smpd_data("binary", smpd_process.pszExe, SMPD_MAX_EXE_LENGTH);
/* launch the manager process */
result = smpd_start_win_mgr(context, SMPD_FALSE);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to start the local smpd manager.\n");
goto quit_job;
}
/* connect to the manager */
smpd_dbg_printf("connecting a new socket.\n");
port = atol(context->port_str);
if (port < 1)
{
smpd_err_printf("Invalid reconnect port read: %d\n", port);
goto quit_job;
}
result = smpd_create_context(context->type, context->set, SMPDU_SOCK_INVALID_SOCK, context->id, &rc_context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create a new context for the reconnection.\n");
goto quit_job;
}
rc_context->state = context->state;
rc_context->write_state = SMPD_RECONNECTING;
context->state = SMPD_CLOSING;
rc_context->connect_to = context->connect_to;
rc_context->connect_return_id = context->connect_return_id;
rc_context->connect_return_tag = context->connect_return_tag;
strcpy(rc_context->host, context->host);
smpd_process.left_context = rc_context;
smpd_dbg_printf("posting a re-connect to %s:%d in %s context.\n", rc_context->connect_to->host, port, smpd_get_context_str(rc_context));
result = SMPDU_Sock_post_connect(rc_context->set, rc_context, rc_context->connect_to->host, port, &rc_context->sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to post a connect to '%s:%d',\nsock error: %s\n",
rc_context->connect_to->host, port, get_sock_error_string(result));
if (smpd_post_abort_command("Unable to connect to '%s:%d',\nsock error: %s\n",
rc_context->connect_to->host, port, get_sock_error_string(result)) != SMPD_SUCCESS)
{
goto quit_job;
}
}
}
else
{
#endif
smpd_process.left_context = context;
result = SMPDU_Sock_set_user_ptr(sock, context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to set the smpd sock user pointer,\nsock error: %s\n",
get_sock_error_string(result));
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
}
#endif
#ifdef HAVE_WINDOWS_H
{
/* Create a break handler and a socket to handle aborting the job when mpiexec receives break signals */
smpd_context_t *reader_context;
SMPDU_Sock_t sock_reader;
SMPDU_SOCK_NATIVE_FD reader, writer;
smpd_make_socket_loop((SOCKET*)&reader, (SOCKET*)&writer);
result = SMPDU_Sock_native_to_sock(set, reader, NULL, &sock_reader);
result = SMPDU_Sock_native_to_sock(set, writer, NULL, &smpd_process.mpiexec_abort_sock);
result = smpd_create_context(SMPD_CONTEXT_MPIEXEC_ABORT, set, sock_reader, -1, &reader_context);
reader_context->read_state = SMPD_READING_MPIEXEC_ABORT;
result = SMPDU_Sock_post_read(sock_reader, &reader_context->read_cmd.cmd, 1, 1, NULL);
if (!SetConsoleCtrlHandler(mpiexec_ctrl_handler, TRUE))
{
/* Don't error out; allow the job to run without a ctrl handler? */
result = GetLastError();
smpd_dbg_printf("unable to set a ctrl handler for mpiexec, error %d\n", result);
}
}
#endif
result = smpd_enter_at_state(set, state);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("state machine failed.\n");
goto quit_job;
}
quit_job:
if ((result != SMPD_SUCCESS) && (smpd_process.mpiexec_exit_code == 0))
{
smpd_process.mpiexec_exit_code = -1;
}
/* finalize */
smpd_dbg_printf("calling SMPDU_Sock_finalize\n");
result = SMPDU_Sock_finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_finalize failed,\nsock error: %s\n", get_sock_error_string(result));
}
/* SMPD_Finalize called in smpd_exit()
smpd_dbg_printf("calling SMPD_Finalize\n");
result = PMPI_Finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPD_Finalize failed,\nerror: %d\n", result);
}
*/
#ifdef HAVE_WINDOWS_H
if (smpd_process.hCloseStdinThreadEvent)
SetEvent(smpd_process.hCloseStdinThreadEvent);
if (smpd_process.hStdinThread != NULL)
{
/* close stdin so the input thread will exit */
CloseHandle(GetStdHandle(STD_INPUT_HANDLE));
if (WaitForSingleObject(smpd_process.hStdinThread, 3000) != WAIT_OBJECT_0)
{
TerminateThread(smpd_process.hStdinThread, 321);
}
CloseHandle(smpd_process.hStdinThread);
}
if (smpd_process.hCloseStdinThreadEvent)
{
CloseHandle(smpd_process.hCloseStdinThreadEvent);
smpd_process.hCloseStdinThreadEvent = NULL;
}
#elif defined(USE_PTHREAD_STDIN_REDIRECTION)
smpd_cancel_stdin_thread();
#endif
smpd_exit_fn("main");
/* SMPD_CS_EXIT(); */
return smpd_exit(smpd_process.mpiexec_exit_code);
}
