HYD_status HYD_pmcd_pmip_get_params(char **t_argv)
{
char **argv = t_argv;
static char dbg_prefix[2 * MAX_HOSTNAME_LEN];
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
argv++;
do {
/* Get the proxy arguments */
status = HYDU_parse_array(&argv, HYD_pmcd_pmip_match_table);
HYDU_ERR_POP(status, "error parsing input array\n");
/* No more arguments left */
if (!(*argv))
break;
} while (1);
/* Verify the arguments we got */
if (HYD_pmcd_pmip.upstream.server_name == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "server name not available\n");
if (HYD_pmcd_pmip.upstream.server_port == -1)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "server port not available\n");
if (HYD_pmcd_pmip.user_global.demux == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "demux engine not available\n");
if (HYD_pmcd_pmip.user_global.debug == -1)
HYD_pmcd_pmip.user_global.debug = 0;
status = HYDT_bsci_init(HYD_pmcd_pmip.user_global.rmk,
HYD_pmcd_pmip.user_global.launcher,
HYD_pmcd_pmip.user_global.launcher_exec,
0 /* disable x */ , HYD_pmcd_pmip.user_global.debug);
HYDU_ERR_POP(status, "proxy unable to initialize bootstrap server\n");
if (HYD_pmcd_pmip.local.id == -1) {
/* We didn't get a proxy ID during launch; query the launcher
* for it. */
status = HYDT_bsci_query_proxy_id(&HYD_pmcd_pmip.local.id);
HYDU_ERR_POP(status, "unable to query launcher for proxy ID\n");
}
if (HYD_pmcd_pmip.local.id == -1)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "proxy ID not available\n");
if (HYD_pmcd_pmip.local.pgid == -1)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "PG ID not available\n");
if (HYD_pmcd_pmip.local.retries == -1)
HYD_pmcd_pmip.local.retries = 0;
HYDU_dbg_finalize();
MPL_snprintf(dbg_prefix, 2 * MAX_HOSTNAME_LEN, "proxy:%d:%d",
HYD_pmcd_pmip.local.pgid, HYD_pmcd_pmip.local.id);
status = HYDU_dbg_init((const char *) dbg_prefix);
HYDU_ERR_POP(status, "unable to initialization debugging\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
int main(int argc, char **argv)
{
struct HYD_proxy *proxy;
struct HYD_exec *exec;
struct HYD_node *node;
int exit_status = 0, i, timeout, reset_rmk, global_core_count;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* Initialize engines that don't require use to know anything
* about the user preferences first */
status = HYDU_dbg_init("mpiexec");
HYDU_ERR_POP(status, "unable to initialization debugging\n");
status = HYDU_set_common_signals(signal_cb);
HYDU_ERR_POP(status, "unable to set signal\n");
status = HYDT_ftb_init();
HYDU_ERR_POP(status, "unable to initialize FTB\n");
/* Get user preferences */
status = HYD_uii_mpx_get_parameters(argv);
HYDU_ERR_POP(status, "error parsing parameters\n");
/* Now we initialize engines that require us to know user
* preferences */
#if HAVE_ALARM
if (HYD_ui_mpich_info.ckpoint_int != -1)
alarm(HYD_ui_mpich_info.ckpoint_int);
#endif /* HAVE_ALARM */
/* The demux engine should be initialized before any sockets are
* created, since it checks for STDIN's validity. If STDIN was
* closed and we opened a socket that got the same fd as STDIN,
* this test will not be possible. */
status = HYDT_dmx_init(&HYD_server_info.user_global.demux);
HYDU_ERR_POP(status, "unable to initialize the demux engine\n");
status =
HYDT_bsci_init(HYD_server_info.user_global.rmk, HYD_server_info.user_global.launcher,
HYD_server_info.user_global.launcher_exec,
HYD_server_info.user_global.enablex, HYD_server_info.user_global.debug);
HYDU_ERR_POP(status, "unable to initialize the bootstrap server\n");
reset_rmk = 0;
if (HYD_server_info.node_list == NULL) {
/* Node list is not created yet. The user might not have
* provided the host file. Query the RMK. */
status = HYDT_bsci_query_node_list(&HYD_server_info.node_list);
HYDU_ERR_POP(status, "unable to query the RMK for a node list\n");
if (HYD_server_info.node_list == NULL) {
char localhost[MAX_HOSTNAME_LEN] = { 0 };
/* The RMK didn't give us anything back; use localhost */
status = HYDU_gethostname(localhost);
HYDU_ERR_POP(status, "unable to get local hostname\n");
status = HYDU_add_to_node_list(localhost, 1, &HYD_server_info.node_list);
HYDU_ERR_POP(status, "unable to add to node list\n");
reset_rmk = 1;
}
}
/*
* If this is a checkpoint-restart, if the user specified the
* number of processes, we already have a dummy executable. If the
* number of processes came from the RMK, our executable list is
* still NULL; a dummy executable needs to be created.
*/
if (HYD_uii_mpx_exec_list == NULL) {
HYDU_ASSERT(HYD_server_info.user_global.ckpoint_prefix, status);
/* create a dummy executable */
status = HYDU_alloc_exec(&HYD_uii_mpx_exec_list);
HYDU_ERR_POP(status, "unable to allocate exec\n");
HYD_uii_mpx_exec_list->appnum = 0;
}
if (HYD_server_info.user_global.debug)
for (node = HYD_server_info.node_list; node; node = node->next)
HYDU_dump_noprefix(stdout, "host: %s\n", node->hostname);
/* Reset the host list to use only the number of processes per
* node as specified by the ppn option. */
if (HYD_ui_mpich_info.ppn != -1) {
for (node = HYD_server_info.node_list; node; node = node->next)
node->core_count = HYD_ui_mpich_info.ppn;
reset_rmk = 1;
}
/* The RMK returned a node list. See if the user requested us to
* manipulate it in some way */
if (HYD_ui_mpich_info.sort_order != NONE) {
qsort_node_list();
reset_rmk = 1;
}
if (reset_rmk) {
/* Reassign node IDs to each node */
for (node = HYD_server_info.node_list, i = 0; node; node = node->next, i++)
node->node_id = i;
/* Reinitialize the bootstrap server with the "user" RMK, so
* it knows that we are not using the node list provided by
* the RMK */
status = HYDT_bsci_finalize();
HYDU_ERR_POP(status, "unable to finalize bootstrap device\n");
status = HYDT_bsci_init("user", HYDT_bsci_info.launcher, HYDT_bsci_info.launcher_exec,
HYDT_bsci_info.enablex, HYDT_bsci_info.debug);
HYDU_ERR_POP(status, "unable to reinitialize the bootstrap server\n");
}
/* If the number of processes is not given, we allocate all the
* available nodes to each executable */
HYD_server_info.pg_list.pg_process_count = 0;
for (exec = HYD_uii_mpx_exec_list; exec; exec = exec->next) {
if (exec->proc_count == -1) {
global_core_count = 0;
for (node = HYD_server_info.node_list, i = 0; node; node = node->next, i++)
global_core_count += node->core_count;
exec->proc_count = global_core_count;
}
HYD_server_info.pg_list.pg_process_count += exec->proc_count;
}
status = HYDU_list_inherited_env(&HYD_server_info.user_global.global_env.inherited);
HYDU_ERR_POP(status, "unable to get the inherited env list\n");
status = HYDU_create_proxy_list(HYD_uii_mpx_exec_list, HYD_server_info.node_list,
&HYD_server_info.pg_list);
HYDU_ERR_POP(status, "unable to create proxy list\n");
/* calculate the core count used by the PG */
HYD_server_info.pg_list.pg_core_count = 0;
for (proxy = HYD_server_info.pg_list.proxy_list; proxy; proxy = proxy->next)
HYD_server_info.pg_list.pg_core_count += proxy->node->core_count;
/* If the user didn't specify a local hostname, try to find one in
* the list of nodes passed to us */
if (HYD_server_info.localhost == NULL) {
/* See if the node list contains a remotely accessible localhost */
for (node = HYD_server_info.node_list; node; node = node->next) {
int is_local, remote_access;
status = HYDU_sock_is_local(node->hostname, &is_local);
HYDU_ERR_POP(status, "unable to check if %s is local\n", node->hostname);
if (is_local) {
status = HYDU_sock_remote_access(node->hostname, &remote_access);
HYDU_ERR_POP(status, "unable to check if %s is remotely accessible\n",
node->hostname);
if (remote_access)
break;
}
}
if (node)
HYD_server_info.localhost = HYDU_strdup(node->hostname);
else {
HYDU_MALLOC(HYD_server_info.localhost, char *, MAX_HOSTNAME_LEN, status);
status = HYDU_gethostname(HYD_server_info.localhost);
HYDU_ERR_POP(status, "unable to get local hostname\n");
}
}
if (HYD_server_info.user_global.debug)
HYD_uiu_print_params();
if (MPL_env2int("MPIEXEC_TIMEOUT", &timeout) == 0)
timeout = -1; /* Infinite timeout */
if (HYD_server_info.user_global.debug)
HYDU_dump(stdout, "Timeout set to %d (-1 means infinite)\n", timeout);
/* Check if the user wants us to use a port within a certain
* range. */
if (MPL_env2str("MPIEXEC_PORTRANGE", (const char **) &HYD_server_info.port_range) ||
MPL_env2str("MPIEXEC_PORT_RANGE", (const char **) &HYD_server_info.port_range) ||
MPL_env2str("MPICH_PORT_RANGE", (const char **) &HYD_server_info.port_range))
HYD_server_info.port_range = HYDU_strdup(HYD_server_info.port_range);
/* Add the stdout/stderr callback handlers */
HYD_server_info.stdout_cb = HYD_uiu_stdout_cb;
HYD_server_info.stderr_cb = HYD_uiu_stderr_cb;
/* Create a pipe connection to wake up the process manager */
if (pipe(HYD_server_info.cmd_pipe) < 0)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "pipe error\n");
/* Launch the processes */
status = HYD_pmci_launch_procs();
HYDU_ERR_POP(status, "process manager returned error launching processes\n");
/* Wait for their completion */
status = HYD_pmci_wait_for_completion(timeout);
HYDU_ERR_POP(status, "process manager error waiting for completion\n");
/* Check for the exit status for all the processes */
if (HYD_ui_mpich_info.print_all_exitcodes)
HYDU_dump(stdout, "Exit codes: ");
exit_status = 0;
for (proxy = HYD_server_info.pg_list.proxy_list; proxy; proxy = proxy->next) {
if (proxy->exit_status == NULL) {
/* We didn't receive the exit status for this proxy */
continue;
}
if (HYD_ui_mpich_info.print_all_exitcodes)
HYDU_dump_noprefix(stdout, "[%s] ", proxy->node->hostname);
for (i = 0; i < proxy->proxy_process_count; i++) {
if (HYD_ui_mpich_info.print_all_exitcodes) {
HYDU_dump_noprefix(stdout, "%d", proxy->exit_status[i]);
if (i < proxy->proxy_process_count - 1)
HYDU_dump_noprefix(stdout, ",");
}
exit_status |= proxy->exit_status[i];
}
if (HYD_ui_mpich_info.print_all_exitcodes)
HYDU_dump_noprefix(stdout, "\n");
}
/* Call finalize functions for lower layers to cleanup their resources */
status = HYD_pmci_finalize();
HYDU_ERR_POP(status, "process manager error on finalize\n");
status = HYDT_ftb_finalize();
HYDU_ERR_POP(status, "error finalizing FTB\n");
#if defined ENABLE_PROFILING
if (HYD_server_info.enable_profiling) {
HYDU_dump_noprefix(stdout, "\n");
HYD_DRAW_LINE(80);
HYDU_dump(stdout, "Number of PMI calls seen by the server: %d\n",
HYD_server_info.num_pmi_calls);
HYD_DRAW_LINE(80);
HYDU_dump_noprefix(stdout, "\n");
}
#endif /* ENABLE_PROFILING */
/* Free the mpiexec params */
HYD_uiu_free_params();
HYDU_free_exec_list(HYD_uii_mpx_exec_list);
fn_exit:
HYDU_dbg_finalize();
HYDU_FUNC_EXIT();
if (status == HYD_GRACEFUL_ABORT)
return 0;
else if (status != HYD_SUCCESS)
return -1;
else if (WIFSIGNALED(exit_status)) {
printf("YOUR APPLICATION TERMINATED WITH THE EXIT STRING: %s (signal %d)\n",
strsignal(WTERMSIG(exit_status)), WTERMSIG(exit_status));
printf("This typically refers to a problem with your application.\n");
printf("Please see the FAQ page for debugging suggestions\n");
return exit_status;
}
else if (WIFEXITED(exit_status)) {
return WEXITSTATUS(exit_status);
}
else if (WIFSTOPPED(exit_status)) {
return WSTOPSIG(exit_status);
}
else {
return exit_status;
}
fn_fail:
goto fn_exit;
}
