static HYD_status fn_name_publish(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *thrid, *val, *name = NULL, *port = NULL;
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 ((val = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "name")) == NULL)
HYDU_ERR_POP(status, "cannot find token: name\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");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, HYDU_strdup("cmd=name-publish-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=1;errmsg=duplicate_service_"), status);
HYD_STRING_STASH(stash, HYDU_strdup(name), status);
HYD_STRING_STASH(stash, HYDU_strdup(";"), status);
}
else
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:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
if (name)
HYDU_FREE(name);
if (port)
HYDU_FREE(port);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_info_putnodeattr(int fd, char *args[])
{
struct HYD_string_stash stash;
char *key, *val, *thrid, *cmd;
struct HYD_pmcd_token *tokens = NULL;
int token_count, ret;
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");
HYDU_ERR_CHKANDJUMP(status, val == NULL, HYD_INTERNAL_ERROR, "unable to find value token\n");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
status = HYD_pmcd_pmi_add_kvs(key, val, HYD_pmcd_pmip.local.kvs, &ret);
HYDU_ERR_POP(status, "unable to put data into kvs\n");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=info-putnodeattr-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("rc="), status);
HYD_STRING_STASH(stash, HYDU_int_to_str(ret), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_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:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_publish_name(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *val;
int token_count;
struct HYD_pmcd_token *tokens = NULL;
char *name = NULL, *port = NULL;
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 = MPL_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 = MPL_strdup(val);
status = HYD_pmcd_pmi_publish(name, port, &success);
HYDU_ERR_POP(status, "error publishing service\n");
HYD_STRING_STASH_INIT(stash);
if (success)
HYD_STRING_STASH(stash, MPL_strdup("cmd=publish_result info=ok rc=0 msg=success\n"),
status);
else
HYD_STRING_STASH(stash,
MPL_strdup("cmd=publish_result info=ok rc=1 msg=key_already_present\n"),
status);
HYD_STRING_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
MPL_free(cmd);
fn_exit:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
if (name)
MPL_free(name);
if (port)
MPL_free(port);
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;
}
static HYD_status send_cmd_upstream(const char *start, int fd, char *args[])
{
int i, sent, closed;
struct HYD_string_stash stash;
char *buf = NULL;
struct HYD_pmcd_hdr hdr;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup(start), status);
for (i = 0; args[i]; i++) {
HYD_STRING_STASH(stash, MPL_strdup(args[i]), status);
if (args[i + 1])
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_SPIT(stash, buf, status);
HYD_pmcd_init_header(&hdr);
hdr.cmd = PMI_CMD;
hdr.pid = fd;
hdr.buflen = strlen(buf);
hdr.pmi_version = 2;
status =
HYDU_sock_write(HYD_pmcd_pmip.upstream.control, &hdr, sizeof(hdr), &sent, &closed,
HYDU_SOCK_COMM_MSGWAIT);
HYDU_ERR_POP(status, "unable to send PMI header upstream\n");
HYDU_ASSERT(!closed, status);
if (HYD_pmcd_pmip.user_global.debug) {
HYDU_dump(stdout, "forwarding command (%s) upstream\n", buf);
}
status = HYDU_sock_write(HYD_pmcd_pmip.upstream.control, buf, hdr.buflen, &sent, &closed,
HYDU_SOCK_COMM_MSGWAIT);
HYDU_ERR_POP(status, "unable to send PMI command upstream\n");
HYDU_ASSERT(!closed, status);
fn_exit:
if (buf)
MPL_free(buf);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_info_getjobattr(int fd, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *key, *thrid;
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 token: key\n");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
if (!strcmp(key, "PMI_process_mapping")) {
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=info-getjobattr-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("found=TRUE;value="), status);
HYD_STRING_STASH(stash, MPL_strdup(HYD_pmcd_pmip.system_global.pmi_process_mapping),
status);
HYD_STRING_STASH(stash, MPL_strdup(";rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_free(cmd);
} else {
status = send_cmd_upstream("cmd=info-getjobattr;", fd, args);
HYDU_ERR_POP(status, "error sending command upstream\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_lookup_name(int fd, int pid, int pgid, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *name, *value = NULL;
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");
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");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=lookup_result"), status);
if (value) {
HYD_STRING_STASH(stash, MPL_strdup(" port="), status);
HYD_STRING_STASH(stash, MPL_strdup(value), status);
HYD_STRING_STASH(stash, MPL_strdup(" info=ok rc=0 msg=success\n"), status);
}
else {
HYD_STRING_STASH(stash, MPL_strdup(" rc=1 msg=service_not_found\n"), status);
}
HYD_STRING_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "send command failed\n");
MPL_free(cmd);
fn_exit:
if (tokens)
HYD_pmcd_pmi_free_tokens(tokens, token_count);
if (value)
MPL_free(value);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status fn_finalize(int fd, char *args[])
{
char *thrid;
struct HYD_string_stash stash;
char *cmd;
struct HYD_pmcd_token *tokens = NULL;
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");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=finalize-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_free(cmd);
status = HYDT_dmx_deregister_fd(fd);
HYDU_ERR_POP(status, "unable to deregister fd\n");
close(fd);
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 fn_job_getid(int fd, char *args[])
{
struct HYD_string_stash stash;
char *cmd, *thrid;
struct HYD_pmcd_token *tokens = NULL;
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");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=job-getid-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("jobid="), status);
HYD_STRING_STASH(stash, MPL_strdup(HYD_pmcd_pmip.local.kvs->kvsname), status);
HYD_STRING_STASH(stash, MPL_strdup(";rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_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 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;
struct HYD_string_stash proxy_stash;
char *control_port;
struct HYD_string_stash stash;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
for (i = 0; args[i]; i++)
mcmd_args[mcmd_num_args++] = MPL_strdup(args[i]);
mcmd_args[mcmd_num_args] = NULL;
/* Initialize the proxy stash, so it can be freed if we jump to
* exit */
HYD_STRING_STASH_INIT(proxy_stash);
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_OR_JUMP(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;
MPL_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;
MPL_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 = MPL_strdup(info_val);
}
else if (!strcmp(info_key, "wdir")) {
exec->wdir = MPL_strdup(info_val);
}
else if (!strcmp(info_key, "host") || !strcmp(info_key, "hosts")) {
char *saveptr;
char *host = strtok_r(info_val, ",", &saveptr);
while (host) {
status = HYDU_process_mfile_token(host, 1, &pg->user_node_list);
HYDU_ERR_POP(status, "error creating node list\n");
host = strtok_r(NULL, ",", &saveptr);
}
}
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 = MPL_strdup(val);
else {
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup(path), status);
HYD_STRING_STASH(stash, MPL_strdup("/"), status);
HYD_STRING_STASH(stash, MPL_strdup(val), status);
HYD_STRING_SPIT(stash, execname, status);
}
i = 0;
exec->exec[i++] = execname;
for (k = 0; k < argcnt; k++) {
MPL_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++] = MPL_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");
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;
MPL_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;
MPL_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);
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");
MPL_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, HYD_FALSE, NULL);
HYDU_ERR_POP(status, "launcher cannot launch processes\n");
{
char *cmd;
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=spawn_result rc=0"), status);
HYD_STRING_STASH(stash, MPL_strdup("\n"), status);
HYD_STRING_SPIT(stash, cmd, status);
status = cmd_response(fd, pid, cmd);
HYDU_ERR_POP(status, "error writing PMI line\n");
MPL_free(cmd);
}
/* Cache the pre-initialized keyvals on the new proxies */
if (preput_num)
bcast_keyvals(fd, pid);
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYD_STRING_STASH_FREE(proxy_stash);
if (segment_list)
MPL_free(segment_list);
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_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_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;
}
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_info_getnodeattr(int fd, char *args[])
{
int found;
struct HYD_pmcd_pmi_kvs_pair *run;
char *key, *waitval, *thrid;
struct HYD_string_stash stash;
char *cmd;
struct HYD_pmcd_token *tokens = NULL;
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");
waitval = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "wait");
thrid = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "thrid");
/* if a predefined value is not found, we let the code fall back
* to regular search and return an error to the client */
found = 0;
for (run = HYD_pmcd_pmip.local.kvs->key_pair; run; run = run->next) {
if (!strcmp(run->key, key)) {
found = 1;
break;
}
}
if (found) { /* We found the attribute */
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=info-getnodeattr-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("found=TRUE;value="), status);
HYD_STRING_STASH(stash, MPL_strdup(run->val), status);
HYD_STRING_STASH(stash, MPL_strdup(";rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_free(cmd);
} else if (waitval && !strcmp(waitval, "TRUE")) {
/* The client wants to wait for a response; queue up the request */
status = HYD_pmcd_pmi_v2_queue_req(fd, -1, -1, args, key, &pending_reqs);
HYDU_ERR_POP(status, "unable to queue request\n");
goto fn_exit;
} else {
/* Tell the client that we can't find the attribute */
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash, MPL_strdup("cmd=info-getnodeattr-response;"), status);
if (thrid) {
HYD_STRING_STASH(stash, MPL_strdup("thrid="), status);
HYD_STRING_STASH(stash, MPL_strdup(thrid), status);
HYD_STRING_STASH(stash, MPL_strdup(";"), status);
}
HYD_STRING_STASH(stash, MPL_strdup("found=FALSE;rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_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 fn_fullinit(int fd, char *args[])
{
int id, i;
char *rank_str;
struct HYD_string_stash stash;
char *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");
rank_str = HYD_pmcd_pmi_find_token_keyval(tokens, token_count, "pmirank");
HYDU_ERR_CHKANDJUMP(status, rank_str == NULL, HYD_INTERNAL_ERROR,
"unable to find pmirank token\n");
id = atoi(rank_str);
/* Store the PMI_RANK to fd mapping */
for (i = 0; i < HYD_pmcd_pmip.local.proxy_process_count; i++) {
if (HYD_pmcd_pmip.downstream.pmi_rank[i] == id) {
HYD_pmcd_pmip.downstream.pmi_fd[i] = fd;
HYD_pmcd_pmip.downstream.pmi_fd_active[i] = 1;
break;
}
}
HYDU_ASSERT(i < HYD_pmcd_pmip.local.proxy_process_count, status);
HYD_STRING_STASH_INIT(stash);
HYD_STRING_STASH(stash,
MPL_strdup("cmd=fullinit-response;pmi-version=2;pmi-subversion=0;rank="),
status);
HYD_STRING_STASH(stash, HYDU_int_to_str(id), status);
HYD_STRING_STASH(stash, MPL_strdup(";size="), status);
HYD_STRING_STASH(stash, HYDU_int_to_str(HYD_pmcd_pmip.system_global.global_process_count),
status);
HYD_STRING_STASH(stash, MPL_strdup(";appnum=0"), status);
if (HYD_pmcd_pmip.local.spawner_kvsname) {
HYD_STRING_STASH(stash, MPL_strdup(";spawner-jobid="), status);
HYD_STRING_STASH(stash, MPL_strdup(HYD_pmcd_pmip.local.spawner_kvsname), status);
}
if (HYD_pmcd_pmip.user_global.debug) {
HYD_STRING_STASH(stash, MPL_strdup(";debugged=TRUE;pmiverbose=TRUE"), status);
} else {
HYD_STRING_STASH(stash, MPL_strdup(";debugged=FALSE;pmiverbose=FALSE"), status);
}
HYD_STRING_STASH(stash, MPL_strdup(";rc=0;"), status);
HYD_STRING_SPIT(stash, cmd, status);
send_cmd_downstream(fd, cmd);
MPL_free(cmd);
fn_exit:
HYD_pmcd_pmi_free_tokens(tokens, token_count);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}