static HYD_status fn_spawn(int fd, int pid, int pgid, char *args[])
{
struct HYD_pg *pg;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_proxy *proxy;
struct HYD_pmcd_token *tokens;
struct HYD_exec *exec_list = NULL, *exec;
struct HYD_env *env;
struct HYD_node *node;
char key[PMI_MAXKEYLEN], *val;
int nprocs, preput_num, info_num, ret;
char *execname, *path = NULL;
struct HYD_pmcd_token_segment *segment_list = NULL;
int token_count, i, j, k, new_pgid, total_spawns;
int argcnt, num_segments;
char *control_port, *proxy_args[HYD_NUM_TMP_STRINGS] = { NULL };
char *tmp[HYD_NUM_TMP_STRINGS];
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
for (i = 0; args[i]; i++)
mcmd_args[mcmd_num_args++] = HYDU_strdup(args[i]);
mcmd_args[mcmd_num_args] = NULL;
status = HYD_pmcd_pmi_args_to_tokens(mcmd_args, &tokens, &token_count);
HYDU_ERR_POP(status, "unable to convert args to tokens\n");
/* Here's the order of things we do:
*
* 1. Break the token list into multiple segments, each segment
* corresponding to a command. Each command represents
* information for one executable.
*
* 2. Allocate a process group for the new set of spawned
* processes
*
* 3. Get all the common keys and deal with them
*
* 4. Create an executable list based on the segments.
*
* 5. Create a proxy list using the created executable list and
* spawn it.
*/
/* Break the token list into multiple segments and create an
* executable list based on the segments. */
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "totspawns");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: totspawns\n");
total_spawns = atoi(val);
HYDU_MALLOC(segment_list, struct HYD_pmcd_token_segment *,
total_spawns * sizeof(struct HYD_pmcd_token_segment), status);
segment_tokens(tokens, token_count, segment_list, &num_segments);
if (num_segments != total_spawns) {
/* We didn't read the entire PMI string; wait for the rest to
* arrive */
goto fn_exit;
}
else {
/* Got the entire PMI string; free the arguments and reset */
HYDU_free_strlist(mcmd_args);
mcmd_num_args = 0;
}
/* Allocate a new process group */
for (pg = &HYD_server_info.pg_list; pg->next; pg = pg->next);
new_pgid = pg->pgid + 1;
status = HYDU_alloc_pg(&pg->next, new_pgid);
HYDU_ERR_POP(status, "unable to allocate process group\n");
pg = pg->next;
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg->spawner_pg = proxy->pg;
for (j = 0; j < total_spawns; j++) {
/* For each segment, we create an exec structure */
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "nprocs");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: nprocs\n");
nprocs = atoi(val);
pg->pg_process_count += nprocs;
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "argcnt");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: argcnt\n");
argcnt = atoi(val);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "info_num");
if (val)
info_num = atoi(val);
else
info_num = 0;
if (exec_list == NULL) {
status = HYDU_alloc_exec(&exec_list);
HYDU_ERR_POP(status, "unable to allocate exec\n");
exec_list->appnum = 0;
exec = exec_list;
}
else {
for (exec = exec_list; exec->next; exec = exec->next);
status = HYDU_alloc_exec(&exec->next);
HYDU_ERR_POP(status, "unable to allocate exec\n");
exec->next->appnum = exec->appnum + 1;
exec = exec->next;
}
/* Info keys */
for (i = 0; i < info_num; i++) {
char *info_key, *info_val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "info_key_%d", i);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
info_key = val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "info_val_%d", i);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
info_val = val;
if (!strcmp(info_key, "path")) {
path = HYDU_strdup(info_val);
}
else if (!strcmp(info_key, "wdir")) {
exec->wdir = HYDU_strdup(info_val);
}
else if (!strcmp(info_key, "host")) {
status = HYDU_process_mfile_token(info_val, 1, &pg->user_node_list);
HYDU_ERR_POP(status, "error create node list\n");
}
else if (!strcmp(info_key, "hostfile")) {
status =
HYDU_parse_hostfile(info_val, &pg->user_node_list, HYDU_process_mfile_token);
HYDU_ERR_POP(status, "error parsing hostfile\n");
}
else {
/* Unrecognized info key; ignore */
}
}
status = HYDU_correct_wdir(&exec->wdir);
HYDU_ERR_POP(status, "unable to correct wdir\n");
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "execname");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: execname\n");
if (path == NULL)
execname = HYDU_strdup(val);
else {
i = 0;
tmp[i++] = HYDU_strdup(path);
tmp[i++] = HYDU_strdup("/");
tmp[i++] = HYDU_strdup(val);
tmp[i++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &execname);
HYDU_ERR_POP(status, "error while joining strings\n");
HYDU_free_strlist(tmp);
}
i = 0;
exec->exec[i++] = execname;
for (k = 0; k < argcnt; k++) {
HYDU_snprintf(key, PMI_MAXKEYLEN, "arg%d", k + 1);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
exec->exec[i++] = HYDU_strdup(val);
}
exec->exec[i++] = NULL;
exec->proc_count = nprocs;
/* It is not clear what kind of environment needs to get
* passed to the spawned process. Don't set anything here, and
* let the proxy do whatever it does by default. */
exec->env_prop = NULL;
status = HYDU_env_create(&env, "PMI_SPAWNED", "1");
HYDU_ERR_POP(status, "unable to create PMI_SPAWNED environment\n");
exec->user_env = env;
}
status = HYD_pmcd_pmi_alloc_pg_scratch(pg);
HYDU_ERR_POP(status, "unable to allocate pg scratch space\n");
if (pg->user_node_list) {
pg->pg_core_count = 0;
for (i = 0, node = pg->user_node_list; node; node = node->next, i++) {
pg->pg_core_count += node->core_count;
node->node_id = i;
}
}
else {
pg->pg_core_count = HYD_server_info.pg_list.pg_core_count;
}
pg->pg_process_count = 0;
for (exec = exec_list; exec; exec = exec->next)
pg->pg_process_count += exec->proc_count;
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) pg->pg_scratch;
/* Get the common keys and deal with them */
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "preput_num");
if (val)
preput_num = atoi(val);
else
preput_num = 0;
for (i = 0; i < preput_num; i++) {
char *preput_key, *preput_val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "preput_key_%d", i);
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
preput_key = val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "preput_val_%d", i);
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
preput_val = val;
status = HYD_pmcd_pmi_add_kvs(preput_key, preput_val, pg_scratch->kvs, &ret);
HYDU_ERR_POP(status, "unable to add keypair to kvs\n");
}
/* Create the proxy list */
if (pg->user_node_list) {
status = HYDU_create_proxy_list(exec_list, pg->user_node_list, pg);
HYDU_ERR_POP(status, "error creating proxy list\n");
}
else {
status = HYDU_create_proxy_list(exec_list, HYD_server_info.node_list, pg);
HYDU_ERR_POP(status, "error creating proxy list\n");
}
HYDU_free_exec_list(exec_list);
status = HYDU_sock_create_and_listen_portstr(HYD_server_info.user_global.iface,
HYD_server_info.local_hostname,
HYD_server_info.port_range, &control_port,
HYD_pmcd_pmiserv_control_listen_cb,
(void *) (size_t) new_pgid);
HYDU_ERR_POP(status, "unable to create PMI port\n");
if (HYD_server_info.user_global.debug)
HYDU_dump(stdout, "Got a control port string of %s\n", control_port);
/* Go to the last PG */
for (pg = &HYD_server_info.pg_list; pg->next; pg = pg->next);
status = HYD_pmcd_pmi_fill_in_proxy_args(proxy_args, control_port, new_pgid);
HYDU_ERR_POP(status, "unable to fill in proxy arguments\n");
HYDU_FREE(control_port);
status = HYD_pmcd_pmi_fill_in_exec_launch_info(pg);
HYDU_ERR_POP(status, "unable to fill in executable arguments\n");
status = HYDT_bsci_launch_procs(proxy_args, pg->proxy_list, NULL);
HYDU_ERR_POP(status, "launcher cannot launch processes\n");
{
char *cmd_str[HYD_NUM_TMP_STRINGS], *cmd;
i = 0;
cmd_str[i++] = HYDU_strdup("cmd=spawn_result rc=0");
cmd_str[i++] = HYDU_strdup("\n");
cmd_str[i++] = NULL;
status = HYDU_str_alloc_and_join(cmd_str, &cmd);
HYDU_ERR_POP(status, "unable to join strings\n");
HYDU_free_strlist(cmd_str);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "error writing PMI line\n");
HYDU_FREE(cmd);
}
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_free_strlist(proxy_args);
if (segment_list)
HYDU_FREE(segment_list);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_spawn(int fd, int pid, int pgid, char *args[])
{
struct HYD_pg *pg;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_proxy *proxy;
struct HYD_pmcd_token *tokens;
struct HYD_exec *exec_list = NULL, *exec;
struct HYD_env *env;
struct HYD_node *node;
char *thrid;
char key[PMI_MAXKEYLEN], *val;
int maxprocs, preputcount, infokeycount, ret;
int ncmds;
char *execname, *path = NULL;
struct HYD_pmcd_token_segment *segment_list = NULL;
int token_count, i, j, k, new_pgid;
int argcnt, num_segments;
struct HYD_string_stash proxy_stash;
char *control_port;
struct HYD_string_stash stash;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
status = HYD_pmcd_pmi_args_to_tokens(args, &tokens, &token_count);
HYDU_ERR_POP(status, "unable to convert args to tokens\n");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
/* Here's the order of things we do:
*
* 1. Break the token list into multiple segments, each segment
* corresponding to a command. Each command represents
* information for one executable.
*
* 2. Allocate a process group for the new set of spawned
* processes
*
* 3. Get all the common keys and deal with them
*
* 4. Create an executable list based on the segments.
*
* 5. Create a proxy list using the created executable list and
* spawn it.
*/
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "ncmds");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR, "unable to find token: ncmds\n");
ncmds = atoi(val);
HYDU_MALLOC(segment_list, struct HYD_pmcd_token_segment *,
(ncmds + 1) * sizeof(struct HYD_pmcd_token_segment), status);
segment_tokens(tokens, token_count, segment_list, &num_segments);
HYDU_ASSERT((ncmds + 1) == num_segments, status);
/* Allocate a new process group */
for (pg = &HYD_server_info.pg_list; pg->next; pg = pg->next);
new_pgid = pg->pgid + 1;
status = HYDU_alloc_pg(&pg->next, new_pgid);
HYDU_ERR_POP(status, "unable to allocate process group\n");
pg = pg->next;
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg->spawner_pg = proxy->pg;
for (j = 1; j <= ncmds; j++) {
/* For each segment, we create an exec structure */
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "maxprocs");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: maxprocs\n");
maxprocs = atoi(val);
pg->pg_process_count += maxprocs;
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "argc");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: argc\n");
argcnt = atoi(val);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "infokeycount");
if (val)
infokeycount = atoi(val);
else
infokeycount = 0;
if (exec_list == NULL) {
status = HYDU_alloc_exec(&exec_list);
HYDU_ERR_POP(status, "unable to allocate exec\n");
exec_list->appnum = 0;
exec = exec_list;
}
else {
for (exec = exec_list; exec->next; exec = exec->next);
status = HYDU_alloc_exec(&exec->next);
HYDU_ERR_POP(status, "unable to allocate exec\n");
exec->next->appnum = exec->appnum + 1;
exec = exec->next;
}
/* Info keys */
for (i = 0; i < infokeycount; i++) {
char *info_key, *info_val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "infokey%d", i);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
info_key = val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "infoval%d", i);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
info_val = val;
if (!strcmp(info_key, "path")) {
path = HYDU_strdup(info_val);
}
else if (!strcmp(info_key, "wdir")) {
exec->wdir = HYDU_strdup(info_val);
}
else if (!strcmp(info_key, "host") || !strcmp(info_key, "hosts")) {
char *host = strtok(info_val, ",");
while (host) {
status = HYDU_process_mfile_token(host, 1, &pg->user_node_list);
HYDU_ERR_POP(status, "error creating node list\n");
host = strtok(NULL, ",");
}
}
else if (!strcmp(info_key, "hostfile")) {
status = HYDU_parse_hostfile(info_val, &pg->user_node_list,
HYDU_process_mfile_token);
HYDU_ERR_POP(status, "error parsing hostfile\n");
}
else {
/* Unrecognized info key; ignore */
}
}
status = HYDU_correct_wdir(&exec->wdir);
HYDU_ERR_POP(status, "unable to correct wdir\n");
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, "subcmd");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: subcmd\n");
if (path == NULL)
execname = HYDU_strdup(val);
else {
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup(path), status);
HYD_STRING_STASH(stash, HYDU_strdup("/"), status);
HYD_STRING_STASH(stash, HYDU_strdup(val), status);
HYD_STRING_SPIT(stash, execname, status);
}
i = 0;
exec->exec[i++] = execname;
for (k = 0; k < argcnt; k++) {
HYDU_snprintf(key, PMI_MAXKEYLEN, "argv%d", k);
val = HYD_pmcd_pmi_find_token_keyval(&tokens[segment_list[j].start_idx],
segment_list[j].token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
exec->exec[i++] = HYDU_strdup(val);
}
exec->exec[i++] = NULL;
exec->proc_count = maxprocs;
/* It is not clear what kind of environment needs to get
* passed to the spawned process. Don't set anything here, and
* let the proxy do whatever it does by default. */
exec->env_prop = NULL;
status = HYDU_env_create(&env, "PMI_SPAWNED", "1");
HYDU_ERR_POP(status, "unable to create PMI_SPAWNED environment\n");
exec->user_env = env;
}
status = HYD_pmcd_pmi_alloc_pg_scratch(pg);
HYDU_ERR_POP(status, "unable to allocate pg scratch space\n");
pg->pg_process_count = 0;
for (exec = exec_list; exec; exec = exec->next)
pg->pg_process_count += exec->proc_count;
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) pg->pg_scratch;
/* Get the common keys and deal with them */
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "preputcount");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: preputcount\n");
preputcount = atoi(val);
for (i = 0; i < preputcount; i++) {
char *preput_key, *preput_val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "ppkey%d", i);
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
preput_key = val;
HYDU_snprintf(key, PMI_MAXKEYLEN, "ppval%d", i);
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, key);
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR,
"unable to find token: %s\n", key);
preput_val = val;
status = HYD_pmcd_pmi_add_kvs(preput_key, preput_val, pg_scratch->kvs, &ret);
HYDU_ERR_POP(status, "unable to add keypair to kvs\n");
}
/* Create the proxy list */
if (pg->user_node_list) {
status = HYDU_create_proxy_list(exec_list, pg->user_node_list, pg);
HYDU_ERR_POP(status, "error creating proxy list\n");
}
else {
status = HYDU_create_proxy_list(exec_list, HYD_server_info.node_list, pg);
HYDU_ERR_POP(status, "error creating proxy list\n");
}
HYDU_free_exec_list(exec_list);
if (pg->user_node_list) {
pg->pg_core_count = 0;
for (i = 0, node = pg->user_node_list; node; node = node->next, i++)
pg->pg_core_count += node->core_count;
}
else {
pg->pg_core_count = 0;
for (proxy = pg->proxy_list; proxy; proxy = proxy->next)
pg->pg_core_count += proxy->node->core_count;
}
status = HYDU_sock_create_and_listen_portstr(HYD_server_info.user_global.iface,
HYD_server_info.localhost,
HYD_server_info.port_range, &control_port,
HYD_pmcd_pmiserv_control_listen_cb,
(void *) (size_t) new_pgid);
HYDU_ERR_POP(status, "unable to create PMI port\n");
if (HYD_server_info.user_global.debug)
HYDU_dump(stdout, "Got a control port string of %s\n", control_port);
/* Go to the last PG */
for (pg = &HYD_server_info.pg_list; pg->next; pg = pg->next);
status = HYD_pmcd_pmi_fill_in_proxy_args(&proxy_stash, control_port, new_pgid);
HYDU_ERR_POP(status, "unable to fill in proxy arguments\n");
HYDU_FREE(control_port);
status = HYD_pmcd_pmi_fill_in_exec_launch_info(pg);
HYDU_ERR_POP(status, "unable to fill in executable arguments\n");
status = HYDT_bsci_launch_procs(proxy_stash.strlist, pg->proxy_list, NULL);
HYDU_ERR_POP(status, "launcher cannot launch processes\n");
{
char *cmd;
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=spawn-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, HYDU_strdup("thrid="), status);
HYD_STRING_STASH(stash, HYDU_strdup(thrid), status);
HYD_STRING_STASH(stash, HYDU_strdup(";"), status);
}
HYD_STRING_STASH(stash, HYDU_strdup("rc=0;"), status);
HYD_STRING_STASH(stash, HYDU_strdup("jobid="), status);
HYD_STRING_STASH(stash, HYDU_strdup(pg_scratch->kvs->kvsname), status);
HYD_STRING_STASH(stash, HYDU_strdup(";"), status);
HYD_STRING_STASH(stash, HYDU_strdup("nerrs=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
HYDU_FREE(cmd);
}
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYD_STRING_STASH_FREE(proxy_stash);
if (segment_list)
HYDU_FREE(segment_list);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
