void HYDU_free_strlist(char **strlist)
{
int arg;
HYDU_FUNC_ENTER();
for (arg = 0; strlist[arg]; arg++)
HYDU_FREE(strlist[arg]);
HYDU_FUNC_EXIT();
}
HYD_status HYDU_list_inherited_env(struct HYD_env **env_list)
{
char *env_str = NULL, *env_name;
int i, ret;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
*env_list = NULL;
i = 0;
while (environ[i]) {
env_str = HYDU_strdup(environ[i]);
env_name = strtok(env_str, "=");
status = HYDT_bsci_query_env_inherit(env_name, &ret);
HYDU_ERR_POP(status, "error querying environment propagation\n");
HYDU_FREE(env_str);
env_str = NULL;
if (!ret) {
i++;
continue;
}
status = HYDU_append_env_str_to_list(environ[i], env_list);
HYDU_ERR_POP(status, "unable to add env to list\n");
i++;
}
fn_exit:
if (env_str)
HYDU_FREE(env_str);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_name_lookup(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *thrid, *name, *value;
int token_count;
struct HYD_pmcd_token *tokens = NULL;
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");
if ((name = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "name")) == NULL)
HYDU_ERR_POP(status, "cannot find token: name\n");
status = HYD_pmcd_pmi_lookup(name, &value);
HYDU_ERR_POP(status, "error while looking up service\n");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=name-lookup-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);
}
if (value) {
HYD_STRING_STASH(stash, HYDU_strdup("port="), status);
HYD_STRING_STASH(stash, HYDU_strdup(value), status);
HYD_STRING_STASH(stash, HYDU_strdup(";found=TRUE;rc=0;"), status);
}
else {
HYD_STRING_STASH(stash, HYDU_strdup("found=FALSE;rc=1;"), 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:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
void HYDU_free_node_list(struct HYD_node *node_list)
{
struct HYD_node *node, *tnode;
node = node_list;
while (node) {
tnode = node->next;
if (node->hostname)
HYDU_FREE(node->hostname);
if (node->user)
HYDU_FREE(node->user);
if (node->local_binding)
HYDU_FREE(node->local_binding);
HYDU_FREE(node);
node = tnode;
}
}
static HYD_status fn_name_unpublish(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *thrid, *name;
int token_count, success;
struct HYD_pmcd_token *tokens = NULL;
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");
if ((name = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "name")) == NULL)
HYDU_ERR_POP(status, "cannot find token: name\n");
status = HYD_pmcd_pmi_unpublish(name, &success);
HYDU_ERR_POP(status, "error unpublishing service\n");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=name-unpublish-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);
}
if (success)
HYD_STRING_STASH(stash, HYDU_strdup("rc=0;"), status);
else {
HYD_STRING_STASH(stash, HYDU_strdup("rc=1;errmsg=service_"), status);
HYD_STRING_STASH(stash, HYDU_strdup(name), status);
HYD_STRING_STASH(stash, HYDU_strdup("_not_found;"), 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:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status poke_progress(char *key)
{
struct HYD_pmcd_pmi_v2_reqs *req, *list_head = NULL, *list_tail = NULL;
int i, count;
HYD_status status = HYD_SUCCESS;
for (count = 0, req = pending_reqs; req; req = req->next)
count++;
for (i = 0; i < count; i++) {
/* Dequeue a request */
req = pending_reqs;
if (pending_reqs) {
pending_reqs = pending_reqs->next;
req->next = NULL;
}
if (key && strcmp(key, req->key)) {
/* If the key doesn't match the request, just queue it back */
if (list_head == NULL) {
list_head = req;
list_tail = req;
}
else {
list_tail->next = req;
req->prev = list_tail;
list_tail = req;
}
}
else {
status = fn_kvs_get(req->fd, req->pid, req->pgid, req->args);
HYDU_ERR_POP(status, "kvs_get returned error\n");
/* Free the dequeued request */
HYDU_free_strlist(req->args);
HYDU_FREE(req);
}
}
if (list_head) {
list_tail->next = pending_reqs;
pending_reqs = list_head;
}
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
void HYDU_finalize_global_env(struct HYD_env_global *global_env)
{
if (global_env->system)
HYDU_env_free_list(global_env->system);
if (global_env->user)
HYDU_env_free_list(global_env->user);
if (global_env->inherited)
HYDU_env_free_list(global_env->inherited);
if (global_env->prop)
HYDU_FREE(global_env->prop);
}
static HYD_status handle_pmi_cmd(int fd, int pgid, int pid, char *buf, int pmi_version)
{
char *args[MAX_PMI_ARGS], *cmd = NULL;
struct HYD_pmcd_pmi_handle *h;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (pmi_version == 1)
HYD_pmcd_pmi_handle = HYD_pmcd_pmi_v1;
else
HYD_pmcd_pmi_handle = HYD_pmcd_pmi_v2;
if (HYD_server_info.user_global.debug)
HYDU_dump(stdout, "[pgid: %d] got PMI command: %s\n", pgid, buf);
status = HYD_pmcd_pmi_parse_pmi_cmd(buf, pmi_version, &cmd, args);
HYDU_ERR_POP(status, "unable to parse PMI command\n");
#if defined ENABLE_PROFILING
if (HYD_server_info.enable_profiling)
HYD_server_info.num_pmi_calls++;
#endif /* ENABLE_PROFILING */
h = HYD_pmcd_pmi_handle;
while (h->handler) {
if (!strcmp(cmd, h->cmd)) {
status = h->handler(fd, pid, pgid, args);
HYDU_ERR_POP(status, "PMI handler returned error\n");
break;
}
h++;
}
if (!h->handler) {
/* We don't understand the command */
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"Unrecognized PMI command: %s | cleaning up processes\n", cmd);
}
fn_exit:
if (cmd)
HYDU_FREE(cmd);
HYDU_free_strlist(args);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_lsf_query_node_list(struct HYD_node **node_list)
{
char *hosts, *hostname, *num_procs_str, *thosts = NULL;
int num_procs;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (MPL_env2str("LSB_MCPU_HOSTS", (const char **) &hosts) == 0)
hosts = NULL;
if (hosts == NULL) {
*node_list = NULL;
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "No LSF node list found\n");
}
else {
hosts = HYDU_strdup(hosts);
thosts = hosts;
hostname = strtok(hosts, " ");
while (1) {
if (hostname == NULL)
break;
/* the even fields in the list should be the number of
* cores */
num_procs_str = strtok(NULL, " ");
HYDU_ASSERT(num_procs_str, status);
num_procs = atoi(num_procs_str);
status = HYDU_add_to_node_list(hostname, num_procs, node_list);
HYDU_ERR_POP(status, "unable to add to node list\n");
hostname = strtok(NULL, " ");
}
if (thosts)
HYDU_FREE(thosts);
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_publish_name(int fd, int pid, int pgid, char *args[])
{
char *tmp[HYD_NUM_TMP_STRINGS], *cmd, *val;
int i, token_count;
struct HYD_pmcd_token *tokens;
char *name, *port;
int success = 0;
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");
if ((val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "service")) == NULL)
HYDU_ERR_POP(status, "cannot find token: service\n");
name = HYDU_strdup(val);
if ((val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "port")) == NULL)
HYDU_ERR_POP(status, "cannot find token: port\n");
port = HYDU_strdup(val);
status = HYD_pmcd_pmi_publish(name, port, &success);
HYDU_ERR_POP(status, "error publishing service\n");
i = 0;
if (success)
tmp[i++] = HYDU_strdup("cmd=publish_result info=ok rc=0 msg=success\n");
else
tmp[i++] = HYDU_strdup("cmd=publish_result info=ok rc=1 msg=key_already_present\n");
tmp[i++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &cmd);
HYDU_ERR_POP(status, "unable to join strings\n");
HYDU_free_strlist(tmp);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
HYDU_FREE(cmd);
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_lookup_name(int fd, int pid, int pgid, char *args[])
{
char *tmp[HYD_NUM_TMP_STRINGS], *cmd, *name, *value;
int i, token_count;
struct HYD_pmcd_token *tokens;
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");
if ((name = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "service")) == NULL)
HYDU_ERR_POP(status, "cannot find token: service\n");
status = HYD_pmcd_pmi_lookup(name, &value);
HYDU_ERR_POP(status, "error while looking up service\n");
i = 0;
tmp[i++] = HYDU_strdup("cmd=lookup_result info=ok");
if (value) {
tmp[i++] = HYDU_strdup("value=");
tmp[i++] = HYDU_strdup(value);
tmp[i++] = HYDU_strdup(" rc=0 msg=success\n");
}
else {
tmp[i++] = HYDU_strdup(" rc=1 msg=service_not_found\n");
}
tmp[i++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &cmd);
HYDU_ERR_POP(status, "unable to join strings\n");
HYDU_free_strlist(tmp);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
HYDU_FREE(cmd);
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
void HYDU_free_pg_list(struct HYD_pg *pg_list)
{
struct HYD_pg *pg, *tpg;
pg = pg_list;
while (pg) {
tpg = pg->next;
if (pg->proxy_list)
HYDU_free_proxy_list(pg->proxy_list);
if (pg->user_node_list)
HYDU_free_node_list(pg->user_node_list);
HYDU_FREE(pg);
pg = tpg;
}
}
static HYD_status control_port_fn(char *arg, char ***argv)
{
char *port = NULL;
HYD_status status = HYD_SUCCESS;
HYDU_ERR_CHKANDJUMP(status, HYD_pmcd_pmip.upstream.server_name, HYD_INTERNAL_ERROR,
"duplicate control port setting\n");
port = HYDU_strdup(**argv);
HYD_pmcd_pmip.upstream.server_name = HYDU_strdup(strtok(port, ":"));
HYD_pmcd_pmip.upstream.server_port = atoi(strtok(NULL, ":"));
(*argv)++;
fn_exit:
if (port)
HYDU_FREE(port);
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYD_pmcd_pmi_add_kvs(const char *key, char *val, struct HYD_pmcd_pmi_kvs *kvs, int *ret)
{
struct HYD_pmcd_pmi_kvs_pair *key_pair, *run, *last;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
HYDU_MALLOC(key_pair, struct HYD_pmcd_pmi_kvs_pair *, sizeof(struct HYD_pmcd_pmi_kvs_pair),
status);
MPL_snprintf(key_pair->key, PMI_MAXKEYLEN, "%s", key);
MPL_snprintf(key_pair->val, PMI_MAXVALLEN, "%s", val);
key_pair->next = NULL;
*ret = 0;
if (kvs->key_pair == NULL) {
kvs->key_pair = key_pair;
}
else {
for (run = kvs->key_pair; run; run = run->next) {
if (!strcmp(run->key, key_pair->key)) {
/* duplicate key found */
*ret = -1;
goto fn_fail;
}
last = run;
}
/* Add key_pair to end of list. */
last->next = key_pair;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
HYDU_FREE(key_pair);
goto fn_exit;
}
HYD_status HYDT_topo_finalize(void)
{
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* Finalize the topology library requested by the user */
#if defined HAVE_HWLOC
if (!strcmp(HYDT_topo_info.topolib, "hwloc")) {
status = HYDT_topo_hwloc_finalize();
HYDU_ERR_POP(status, "unable to finalize hwloc\n");
}
#endif /* HAVE_HWLOC */
if (HYDT_topo_info.topolib)
HYDU_FREE(HYDT_topo_info.topolib);
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_slurm_launch_procs(char **args, struct HYD_proxy *proxy_list,
int *control_fd)
{
int num_hosts, idx, i;
int *pid, *fd_list;
char *targs[HYD_NUM_TMP_STRINGS], *node_list_str = NULL;
char *path = NULL, *extra_arg_list = NULL, *extra_arg;
struct HYD_proxy *proxy;
struct HYDT_topo_cpuset_t cpuset;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* We use the following priority order for the executable path:
* (1) user-specified; (2) search in path; (3) Hard-coded
* location */
if (HYDT_bsci_info.launcher_exec)
path = HYDU_strdup(HYDT_bsci_info.launcher_exec);
if (!path)
path = HYDU_find_full_path("srun");
if (!path)
path = HYDU_strdup("/usr/bin/srun");
idx = 0;
targs[idx++] = HYDU_strdup(path);
if (strcmp(HYDT_bsci_info.rmk, "slurm")) {
targs[idx++] = HYDU_strdup("--nodelist");
status = proxy_list_to_node_str(proxy_list, &node_list_str);
HYDU_ERR_POP(status, "unable to build a node list string\n");
targs[idx++] = HYDU_strdup(node_list_str);
}
num_hosts = 0;
for (proxy = proxy_list; proxy; proxy = proxy->next)
num_hosts++;
targs[idx++] = HYDU_strdup("-N");
targs[idx++] = HYDU_int_to_str(num_hosts);
targs[idx++] = HYDU_strdup("-n");
targs[idx++] = HYDU_int_to_str(num_hosts);
MPL_env2str("HYDRA_LAUNCHER_EXTRA_ARGS", (const char **) &extra_arg_list);
if (extra_arg_list) {
extra_arg = strtok(extra_arg_list, " ");
while (extra_arg) {
targs[idx++] = HYDU_strdup(extra_arg);
extra_arg = strtok(NULL, " ");
}
}
/* Fill in the remaining arguments */
/* We do not need to create a quoted version of the string for
* SLURM. It seems to be internally quoting it anyway. */
for (i = 0; args[i]; i++)
targs[idx++] = HYDU_strdup(args[i]);
/* Increase pid list to accommodate the new pid */
HYDU_MALLOC(pid, int *, (HYD_bscu_pid_count + 1) * sizeof(int), status);
for (i = 0; i < HYD_bscu_pid_count; i++)
pid[i] = HYD_bscu_pid_list[i];
HYDU_FREE(HYD_bscu_pid_list);
HYD_bscu_pid_list = pid;
/* Increase fd list to accommodate these new fds */
HYDU_MALLOC(fd_list, int *, (HYD_bscu_fd_count + 3) * sizeof(int), status);
for (i = 0; i < HYD_bscu_fd_count; i++)
fd_list[i] = HYD_bscu_fd_list[i];
HYDU_FREE(HYD_bscu_fd_list);
HYD_bscu_fd_list = fd_list;
/* append proxy ID as -1 */
targs[idx++] = HYDU_int_to_str(-1);
targs[idx++] = NULL;
if (HYDT_bsci_info.debug) {
HYDU_dump(stdout, "Launch arguments: ");
HYDU_print_strlist(targs);
}
HYDT_topo_cpuset_zero(&cpuset);
status = HYDU_create_process(targs, NULL, NULL, &fd_stdout, &fd_stderr,
&HYD_bscu_pid_list[HYD_bscu_pid_count++], cpuset);
HYDU_ERR_POP(status, "create process returned error\n");
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stdout;
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stderr;
status = HYDT_dmx_register_fd(1, &fd_stdout, HYD_POLLIN,
(void *) (size_t) STDOUT_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
status = HYDT_dmx_register_fd(1, &fd_stderr, HYD_POLLIN,
(void *) (size_t) STDERR_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
fn_exit:
if (node_list_str)
HYDU_FREE(node_list_str);
HYDU_free_strlist(targs);
if (path)
HYDU_FREE(path);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_info_getjobattr(int fd, int pid, int pgid, char *args[])
{
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_pmcd_pmi_kvs_pair *run;
const char *key;
char *thrid, *val, *cmd;
struct HYD_string_stash stash;
struct HYD_pmcd_token *tokens;
int token_count;
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");
key = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "key");
HYDU_ERR_CHKANDJUMP(status, key == NULL, HYD_INTERNAL_ERROR, "unable to find key token\n");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
val = NULL;
if (!strcmp(key, "PMI_dead_processes"))
val = pg_scratch->dead_processes;
/* Try to find the key */
for (run = pg_scratch->kvs->key_pair; run; run = run->next) {
if (!strcmp(run->key, key)) {
val = run->val;
break;
}
}
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=info-getjobattr-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("found="), status);
if (val) {
HYD_STRING_STASH(stash, HYDU_strdup("TRUE;value="), status);
HYD_STRING_STASH(stash, HYDU_strdup(val), status);
HYD_STRING_STASH(stash, HYDU_strdup(";rc=0;"), status);
}
else {
HYD_STRING_STASH(stash, HYDU_strdup("FALSE;rc=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);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_kvs_put(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *key, *val, *thrid, *cmd;
int ret;
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_pmcd_token *tokens;
int token_count;
struct HYD_pmcd_pmi_v2_reqs *req;
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");
key = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "key");
HYDU_ERR_CHKANDJUMP(status, key == NULL, HYD_INTERNAL_ERROR, "unable to find key token\n");
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "value");
if (val == NULL) {
/* the user sent an empty string */
val = HYDU_strdup("");
}
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
status = HYD_pmcd_pmi_add_kvs(key, val, pg_scratch->kvs, &ret);
HYDU_ERR_POP(status, "unable to put data into kvs\n");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=kvs-put-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="), status);
HYD_STRING_STASH(stash, HYDU_int_to_str(ret), status);
HYD_STRING_STASH(stash, HYDU_strdup(";"), status);
HYD_STRING_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
HYDU_FREE(cmd);
for (req = pending_reqs; req; req = req->next) {
if (!strcmp(req->key, key)) {
/* Poke the progress engine before exiting */
status = poke_progress(key);
HYDU_ERR_POP(status, "poke progress error\n");
break;
}
}
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_ll_launch_procs(char **args, struct HYD_proxy *proxy_list,
int *control_fd)
{
int idx, i, total_procs, node_count;
int *pid, *fd_list, exec_idx;
char *targs[HYD_NUM_TMP_STRINGS], *node_list_str = NULL;
char *path = NULL, *extra_arg_list = NULL, *extra_arg, quoted_exec_string[HYD_TMP_STRLEN];
struct HYD_proxy *proxy;
struct HYDT_topo_cpuset_t cpuset;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* We use the following priority order for the executable path:
* (1) user-specified; (2) search in path; (3) Hard-coded
* location */
if (HYDT_bsci_info.launcher_exec)
path = HYDU_strdup(HYDT_bsci_info.launcher_exec);
if (!path)
path = HYDU_find_full_path("poe");
if (!path)
path = HYDU_strdup("/usr/bin/poe");
idx = 0;
targs[idx++] = HYDU_strdup(path);
if (!strcmp(HYDT_bsci_info.rmk, "ll")) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"ll does not support user-defined host lists\n");
}
/* Check how many nodes are being passed for the launch */
status = HYDTI_bscd_ll_query_node_count(&total_procs);
HYDU_ERR_POP(status, "unable to query for the node count\n");
node_count = 0;
for (proxy = proxy_list; proxy; proxy = proxy->next)
node_count++;
if (total_procs != node_count)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"processes to be launched have to cover all nodes\n");
MPL_env2str("HYDRA_LAUNCHER_EXTRA_ARGS", (const char **) &extra_arg_list);
if (extra_arg_list) {
extra_arg = strtok(extra_arg_list, " ");
while (extra_arg) {
targs[idx++] = HYDU_strdup(extra_arg);
extra_arg = strtok(NULL, " ");
}
}
/* Fill in the remaining arguments */
exec_idx = idx;
for (i = 0; args[i]; i++)
targs[idx++] = HYDU_strdup(args[i]);
/* Create a quoted version of the exec string, which is only used
* when the executable is not launched directly, but through an
* actual launcher */
HYDU_snprintf(quoted_exec_string, HYD_TMP_STRLEN, "\"%s\"", targs[exec_idx]);
HYDU_FREE(targs[exec_idx]);
targs[exec_idx] = quoted_exec_string;
/* Increase pid list to accommodate the new pid */
HYDU_MALLOC(pid, int *, (HYD_bscu_pid_count + 1) * sizeof(int), status);
for (i = 0; i < HYD_bscu_pid_count; i++)
pid[i] = HYD_bscu_pid_list[i];
HYDU_FREE(HYD_bscu_pid_list);
HYD_bscu_pid_list = pid;
/* Increase fd list to accommodate these new fds */
HYDU_MALLOC(fd_list, int *, (HYD_bscu_fd_count + 3) * sizeof(int), status);
for (i = 0; i < HYD_bscu_fd_count; i++)
fd_list[i] = HYD_bscu_fd_list[i];
HYDU_FREE(HYD_bscu_fd_list);
HYD_bscu_fd_list = fd_list;
/* append proxy ID as -1 */
targs[idx++] = HYDU_int_to_str(-1);
targs[idx++] = NULL;
HYDT_topo_cpuset_zero(&cpuset);
status = HYDU_create_process(targs, NULL, NULL, &fd_stdout, &fd_stderr,
&HYD_bscu_pid_list[HYD_bscu_pid_count++], cpuset);
HYDU_ERR_POP(status, "create process returned error\n");
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stdout;
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stderr;
status = HYDT_dmx_register_fd(1, &fd_stdout, HYD_POLLIN,
(void *) (size_t) STDOUT_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
status = HYDT_dmx_register_fd(1, &fd_stderr, HYD_POLLIN,
(void *) (size_t) STDERR_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
fn_exit:
if (node_list_str)
HYDU_FREE(node_list_str);
HYDU_free_strlist(targs);
if (path)
HYDU_FREE(path);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_put(int fd, int pid, int pgid, char *args[])
{
int i, ret;
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
char *kvsname, *key, *val;
char *tmp[HYD_NUM_TMP_STRINGS], *cmd;
struct HYD_pmcd_token *tokens;
int token_count;
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");
kvsname = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "kvsname");
HYDU_ERR_CHKANDJUMP(status, kvsname == NULL, HYD_INTERNAL_ERROR,
"unable to find token: kvsname\n");
key = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "key");
HYDU_ERR_CHKANDJUMP(status, key == NULL, HYD_INTERNAL_ERROR,
"unable to find token: key\n");
val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "value");
if (val == NULL) {
/* the user sent an empty string */
val = HYDU_strdup("");
}
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
if (strcmp(pg_scratch->kvs->kvs_name, kvsname))
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"kvsname (%s) does not match this group's kvs space (%s)\n",
kvsname, pg_scratch->kvs->kvs_name);
status = HYD_pmcd_pmi_add_kvs(key, val, pg_scratch->kvs, &ret);
HYDU_ERR_POP(status, "unable to add keypair to kvs\n");
i = 0;
tmp[i++] = HYDU_strdup("cmd=put_result rc=");
tmp[i++] = HYDU_int_to_str(ret);
if (ret == 0) {
tmp[i++] = HYDU_strdup(" msg=success");
}
else {
tmp[i++] = HYDU_strdup(" msg=duplicate_key");
tmp[i++] = HYDU_strdup(key);
}
tmp[i++] = HYDU_strdup("\n");
tmp[i++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &cmd);
HYDU_ERR_POP(status, "unable to join strings\n");
HYDU_free_strlist(tmp);
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_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_kvs_get(int fd, int pid, int pgid, char *args[])
{
int i, idx, found;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_pg *pg;
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_kvs_pair *run;
char *key, *thrid, *cmd;
struct HYD_string_stash stash;
struct HYD_pmcd_token *tokens;
int token_count;
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");
key = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "key");
HYDU_ERR_CHKANDJUMP(status, key == NULL, HYD_INTERNAL_ERROR, "unable to find key token\n");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
found = 0;
for (run = pg_scratch->kvs->key_pair; run; run = run->next) {
if (!strcmp(run->key, key)) {
found = 1;
break;
}
}
if (!found) {
pg = proxy->pg;
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) pg->pg_scratch;
idx = -1;
for (i = 0; i < pg->pg_process_count; i++)
if (pg_scratch->ecount[i].fd == fd && pg_scratch->ecount[i].pid == pid) {
idx = i;
break;
}
HYDU_ASSERT(idx != -1, status);
for (i = 0; i < pg->pg_process_count; i++) {
if (pg_scratch->ecount[i].epoch < pg_scratch->ecount[idx].epoch) {
/* We haven't reached a barrier yet; queue up request */
status = HYD_pmcd_pmi_v2_queue_req(fd, pid, pgid, args, key, &pending_reqs);
HYDU_ERR_POP(status, "unable to queue request\n");
/* We are done */
goto fn_exit;
}
}
}
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=kvs-get-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);
}
if (found) {
HYD_STRING_STASH(stash, HYDU_strdup("found=TRUE;value="), status);
HYD_STRING_STASH(stash, HYDU_strdup(run->val), status);
HYD_STRING_STASH(stash, HYDU_strdup(";"), status);
}
else {
HYD_STRING_STASH(stash, HYDU_strdup("found=FALSE;"), status);
}
HYD_STRING_STASH(stash, HYDU_strdup("rc=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);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_kvs_fence(int fd, int pid, int pgid, char *args[])
{
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_string_stash stash;
char *cmd, *thrid;
struct HYD_pmcd_token *tokens;
int token_count, i;
static int fence_count = 0;
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");
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
/* Try to find the epoch point of this process */
for (i = 0; i < proxy->pg->pg_process_count; i++)
if (pg_scratch->ecount[i].fd == fd && pg_scratch->ecount[i].pid == pid)
pg_scratch->ecount[i].epoch++;
if (i == proxy->pg->pg_process_count) {
/* couldn't find the current process; find a NULL entry */
for (i = 0; i < proxy->pg->pg_process_count; i++)
if (pg_scratch->ecount[i].fd == HYD_FD_UNSET)
break;
pg_scratch->ecount[i].fd = fd;
pg_scratch->ecount[i].pid = pid;
pg_scratch->ecount[i].epoch = 1;
}
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=kvs-fence-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_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
HYDU_FREE(cmd);
fence_count++;
if (fence_count % proxy->pg->pg_process_count == 0) {
/* Poke the progress engine before exiting */
status = poke_progress(NULL);
HYDU_ERR_POP(status, "poke progress error\n");
}
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
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 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_get(int fd, int pid, int pgid, char *args[])
{
int i;
struct HYD_proxy *proxy;
struct HYD_pmcd_pmi_pg_scratch *pg_scratch;
struct HYD_pmcd_pmi_kvs_pair *run;
char *kvsname, *key, *val;
char *tmp[HYD_NUM_TMP_STRINGS], *cmd;
struct HYD_pmcd_token *tokens;
int token_count;
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");
kvsname = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "kvsname");
HYDU_ERR_CHKANDJUMP(status, kvsname == NULL, HYD_INTERNAL_ERROR,
"unable to find token: kvsname\n");
key = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "key");
HYDU_ERR_CHKANDJUMP(status, key == NULL, HYD_INTERNAL_ERROR,
"unable to find token: key\n");
proxy = HYD_pmcd_pmi_find_proxy(fd);
HYDU_ASSERT(proxy, status);
pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch;
val = NULL;
if (!strcmp(key, "PMI_dead_processes")) {
val = pg_scratch->dead_processes;
goto found_val;
}
if (strcmp(pg_scratch->kvs->kvs_name, kvsname))
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"kvsname (%s) does not match this group's kvs space (%s)\n",
kvsname, pg_scratch->kvs->kvs_name);
/* Try to find the key */
for (run = pg_scratch->kvs->key_pair; run; run = run->next) {
if (!strcmp(run->key, key)) {
val = run->val;
break;
}
}
found_val:
i = 0;
tmp[i++] = HYDU_strdup("cmd=get_result rc=");
if (val) {
tmp[i++] = HYDU_strdup("0 msg=success value=");
tmp[i++] = HYDU_strdup(val);
}
else {
tmp[i++] = HYDU_strdup("-1 msg=key_");
tmp[i++] = HYDU_strdup(key);
tmp[i++] = HYDU_strdup("_not_found value=unknown");
}
tmp[i++] = HYDU_strdup("\n");
tmp[i++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &cmd);
HYDU_ERR_POP(status, "unable to join strings\n");
HYDU_free_strlist(tmp);
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_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYD_pmcd_pmi_parse_pmi_cmd(char *obuf, int pmi_version, char **pmi_cmd,
char *args[])
{
char *tbuf = NULL, *seg, *str1 = NULL, *cmd;
char *buf;
char *tmp[HYD_NUM_TMP_STRINGS], *targs[HYD_NUM_TMP_STRINGS];
const char *delim;
int i, j, k;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* Make a copy of the original buffer */
buf = HYDU_strdup(obuf);
if (buf[strlen(obuf) - 1] == '\n')
buf[strlen(obuf) - 1] = '\0';
if (pmi_version == 1) {
if (!strncmp(buf, "cmd=", strlen("cmd=")))
delim = " ";
else
delim = "\n";
/* Here we only get PMI-1 commands or backward compatible
* PMI-2 commands, so we always explicitly use the PMI-1
* delimiter. This allows us to get backward-compatible PMI-2
* commands interleaved with regular PMI-2 commands. */
tbuf = HYDU_strdup(buf);
cmd = strtok(tbuf, delim);
for (i = 0; i < HYD_NUM_TMP_STRINGS; i++) {
targs[i] = strtok(NULL, delim);
if (targs[i] == NULL)
break;
}
/* Make a pass through targs and merge space separated
* arguments which are actually part of the same key */
k = 0;
for (i = 0; targs[i]; i++) {
if (!strrchr(targs[i], ' ')) {
/* no spaces */
args[k++] = HYDU_strdup(targs[i]);
}
else {
/* space in the argument; each segment is either a new
* key, or a space-separated part of the previous
* key */
j = 0;
seg = strtok(targs[i], " ");
while (1) {
if (!seg || strrchr(seg, '=')) {
/* segment has an '='; it's a start of a new key */
if (j) {
tmp[j++] = NULL;
status = HYDU_str_alloc_and_join(tmp, &args[k++]);
HYDU_ERR_POP(status, "error while joining strings\n");
HYDU_free_strlist(tmp);
}
j = 0;
if (!seg)
break;
}
else {
/* no '='; part of the previous key */
tmp[j++] = HYDU_strdup(" ");
}
tmp[j++] = HYDU_strdup(seg);
seg = strtok(NULL, " ");
}
}
}
args[k++] = NULL;
}
else { /* PMI-v2 */
delim = ";";
tbuf = HYDU_strdup(buf);
cmd = strtok(tbuf, delim);
for (i = 0; i < HYD_NUM_TMP_STRINGS; i++) {
args[i] = strtok(NULL, delim);
if (args[i] == NULL)
break;
args[i] = HYDU_strdup(args[i]);
}
}
/* Search for the PMI command in our table */
status = HYDU_strsplit(cmd, &str1, pmi_cmd, '=');
HYDU_ERR_POP(status, "string split returned error\n");
fn_exit:
HYDU_FREE(buf);
if (tbuf)
HYDU_FREE(tbuf);
if (str1)
HYDU_FREE(str1);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscu_wait_for_completion(int timeout)
{
int pid, ret, count, i, time_elapsed, time_left;
struct timeval start, now;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* FIXME: We rely on gettimeofday here. This needs to detect the
* timer type available and use that. Probably more of an MPL
* functionality than Hydra's. */
gettimeofday(&start, NULL);
/* Loop till all sockets have closed */
restart_wait:
while (1) {
count = 0;
for (i = 0; i < HYD_bscu_fd_count; i++) {
if (HYD_bscu_fd_list[i] == HYD_FD_CLOSED)
continue;
ret = HYDT_dmx_query_fd_registration(HYD_bscu_fd_list[i]);
if (ret) { /* still registered */
count++; /* We still need to wait */
gettimeofday(&now, NULL);
time_elapsed = (now.tv_sec - start.tv_sec); /* Ignore microsec granularity */
if (timeout > 0) {
if (time_elapsed > timeout) {
status = HYD_TIMED_OUT;
goto fn_exit;
}
else
time_left = timeout - time_elapsed;
}
else
time_left = -1;
status = HYDT_dmx_wait_for_event(time_left);
HYDU_ERR_POP(status, "error waiting for event\n");
/* Check if any processes terminated badly; if they
* did, return an error. */
pid = waitpid(-1, &ret, WNOHANG);
if (pid > 0) {
/* Find the pid and mark it as complete */
for (i = 0; i < HYD_bscu_pid_count; i++)
if (HYD_bscu_pid_list[i] == pid) {
HYD_bscu_pid_list[i] = -1;
break;
}
if (ret) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"one of the processes terminated badly; aborting\n");
}
}
goto restart_wait;
}
else
HYD_bscu_fd_list[i] = HYD_FD_CLOSED;
}
if (count == 0)
break;
}
/* Loop till all processes have completed */
while (1) {
count = 0;
for (i = 0; i < HYD_bscu_pid_count; i++)
if (HYD_bscu_pid_list[i] != -1)
count++;
/* If there are no processes to wait, we are done */
if (count == 0)
break;
pid = waitpid(-1, &ret, WNOHANG);
if (pid > 0) {
/* Find the pid and mark it as complete */
for (i = 0; i < HYD_bscu_pid_count; i++)
if (HYD_bscu_pid_list[i] == pid) {
HYD_bscu_pid_list[i] = -1;
break;
}
}
}
if (HYD_bscu_pid_list) {
HYDU_FREE(HYD_bscu_pid_list);
HYD_bscu_pid_list = NULL;
HYD_bscu_pid_count = 0;
}
if (HYD_bscu_fd_list) {
HYDU_FREE(HYD_bscu_fd_list);
HYD_bscu_fd_list = NULL;
HYD_bscu_fd_count = 0;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static void cleanup_params(void)
{
int i;
HYDU_finalize_user_global(&HYD_pmcd_pmip.user_global);
/* System global */
if (HYD_pmcd_pmip.system_global.pmi_fd)
HYDU_FREE(HYD_pmcd_pmip.system_global.pmi_fd);
if (HYD_pmcd_pmip.system_global.pmi_process_mapping)
HYDU_FREE(HYD_pmcd_pmip.system_global.pmi_process_mapping);
/* Upstream */
if (HYD_pmcd_pmip.upstream.server_name)
HYDU_FREE(HYD_pmcd_pmip.upstream.server_name);
/* Downstream */
if (HYD_pmcd_pmip.downstream.out)
HYDU_FREE(HYD_pmcd_pmip.downstream.out);
if (HYD_pmcd_pmip.downstream.err)
HYDU_FREE(HYD_pmcd_pmip.downstream.err);
if (HYD_pmcd_pmip.downstream.pid)
HYDU_FREE(HYD_pmcd_pmip.downstream.pid);
if (HYD_pmcd_pmip.downstream.exit_status)
HYDU_FREE(HYD_pmcd_pmip.downstream.exit_status);
if (HYD_pmcd_pmip.downstream.pmi_rank)
HYDU_FREE(HYD_pmcd_pmip.downstream.pmi_rank);
if (HYD_pmcd_pmip.downstream.pmi_fd)
HYDU_FREE(HYD_pmcd_pmip.downstream.pmi_fd);
if (HYD_pmcd_pmip.downstream.pmi_fd_active)
HYDU_FREE(HYD_pmcd_pmip.downstream.pmi_fd_active);
/* Local */
if (HYD_pmcd_pmip.local.iface_ip_env_name)
HYDU_FREE(HYD_pmcd_pmip.local.iface_ip_env_name);
if (HYD_pmcd_pmip.local.hostname)
HYDU_FREE(HYD_pmcd_pmip.local.hostname);
if (HYD_pmcd_pmip.local.spawner_kvs_name)
HYDU_FREE(HYD_pmcd_pmip.local.spawner_kvs_name);
if (HYD_pmcd_pmip.local.ckpoint_prefix_list) {
for (i = 0; HYD_pmcd_pmip.local.ckpoint_prefix_list[i]; i++)
HYDU_FREE(HYD_pmcd_pmip.local.ckpoint_prefix_list[i]);
HYDU_FREE(HYD_pmcd_pmip.local.ckpoint_prefix_list);
}
HYD_pmcd_free_pmi_kvs_list(HYD_pmcd_pmip.local.kvs);
/* Exec list */
HYDU_free_exec_list(HYD_pmcd_pmip.exec_list);
HYDT_topo_finalize();
}
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;
}