static pmix_status_t mylog(const pmix_proc_t *source,
const pmix_info_t data[], size_t ndata,
const pmix_info_t directives[], size_t ndirs,
pmix_op_cbfunc_t cbfunc, void *cbdata)
{
size_t n;
pmix_status_t rc;
pmix_byte_object_t bo;
pmix_iof_flags_t flags= {0};
/* if there is no data, then we don't handle it */
if (NULL == data || 0 == ndata) {
return PMIX_ERR_NOT_AVAILABLE;
}
/* if we are not a gateway, then we don't handle this */
if (!PMIX_PROC_IS_GATEWAY(pmix_globals.mypeer)) {
return PMIX_ERR_TAKE_NEXT_OPTION;
}
/* check to see if there are any relevant directives */
for (n=0; n < ndirs; n++) {
if (0 == strncmp(directives[n].key, PMIX_LOG_TIMESTAMP, PMIX_MAX_KEYLEN)) {
flags.timestamp = data[n].value.data.time;
} else if (0 == strncmp(directives[n].key, PMIX_LOG_XML_OUTPUT, PMIX_MAX_KEYLEN)) {
flags.xml = PMIX_INFO_TRUE(&directives[n]);
} else if (0 == strncmp(directives[n].key, PMIX_LOG_TAG_OUTPUT, PMIX_MAX_KEYLEN)) {
flags.tag = PMIX_INFO_TRUE(&directives[n]);
}
}
/* check to see if there are any stdfd entries */
rc = PMIX_ERR_TAKE_NEXT_OPTION;
for (n=0; n < ndata; n++) {
if (0 == strncmp(data[n].key, PMIX_LOG_STDERR, PMIX_MAX_KEYLEN)) {
bo.bytes = data[n].value.data.string;
bo.size = strlen(bo.bytes);
pmix_iof_write_output(source, PMIX_FWD_STDERR_CHANNEL, &bo, &flags);
/* flag that we did this one */
PMIX_INFO_OP_COMPLETED(&data[n]);
rc = PMIX_SUCCESS;
} else if (0 == strncmp(data[n].key, PMIX_LOG_STDOUT, PMIX_MAX_KEYLEN)) {
bo.bytes = data[n].value.data.string;
bo.size = strlen(bo.bytes);
pmix_iof_write_output(source, PMIX_FWD_STDOUT_CHANNEL, &bo, &flags);
/* flag that we did this one */
PMIX_INFO_OP_COMPLETED(&data[n]);
rc = PMIX_SUCCESS;
}
}
return rc;
}
static pmix_status_t connect_to_peer(struct pmix_peer_t *peer,
pmix_info_t *info, size_t ninfo)
{
char *evar, **uri, *suri;
char *filename, *nspace=NULL;
pmix_rank_t rank = PMIX_RANK_WILDCARD;
char *p, *p2;
int sd, rc;
size_t n;
char myhost[PMIX_MAXHOSTNAMELEN];
bool system_level = false;
bool system_level_only = false;
pid_t pid = 0;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp: connecting to server");
/* see if the connection info is in the info array - if
* so, then that overrides all other options */
/* if I am a client, then we need to look for the appropriate
* connection info in the environment */
if (PMIX_PROC_IS_CLIENT(pmix_globals.mypeer)) {
if (NULL != (evar = getenv("PMIX_SERVER_URI21"))) {
/* we are talking to a v2.1 server */
pmix_client_globals.myserver->proc_type = PMIX_PROC_SERVER | PMIX_PROC_V21;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"V21 SERVER DETECTED");
/* must use the v21 bfrops module */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v21");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else if (NULL != (evar = getenv("PMIX_SERVER_URI2"))) {
/* we are talking to a v2.0 server */
pmix_client_globals.myserver->proc_type = PMIX_PROC_SERVER | PMIX_PROC_V20;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"V20 SERVER DETECTED");
/* must use the v20 bfrops module */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v20");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else {
/* not us */
return PMIX_ERR_NOT_SUPPORTED;
}
/* the server will be using the same bfrops as us */
pmix_client_globals.myserver->nptr->compat.bfrops = pmix_globals.mypeer->nptr->compat.bfrops;
/* mark that we are using the V2 protocol */
pmix_globals.mypeer->protocol = PMIX_PROTOCOL_V2;
/* the URI consists of the following elements:
* - server nspace.rank
* - ptl rendezvous URI
*/
uri = pmix_argv_split(evar, ';');
if (2 != pmix_argv_count(uri)) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
/* set the server nspace */
p = uri[0];
if (NULL == (p2 = strchr(p, '.'))) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
*p2 = '\0';
++p2;
nspace = strdup(p);
rank = strtoull(p2, NULL, 10);
/* save the URI, but do not overwrite what we may have received from
* the info-key directives */
if (NULL == mca_ptl_tcp_component.super.uri) {
mca_ptl_tcp_component.super.uri = strdup(uri[1]);
}
pmix_argv_free(uri);
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:client attempt connect to %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS != (rc = try_connect(&sd))) {
free(nspace);
return rc;
}
goto complete;
}
/* get here if we are a tool - check any provided directives
* to see where they want us to connect to */
if (NULL != info) {
for (n=0; n < ninfo; n++) {
if (0 == strcmp(info[n].key, PMIX_CONNECT_TO_SYSTEM)) {
system_level_only = PMIX_INFO_TRUE(&info[n]);
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_SYSTEM_FIRST)) {
/* try the system-level */
system_level = PMIX_INFO_TRUE(&info[n]);
} else if (0 == strcmp(info[n].key, PMIX_SERVER_PIDINFO)) {
pid = info[n].value.data.pid;
} else if (0 == strcmp(info[n].key, PMIX_SERVER_URI)) {
if (NULL == mca_ptl_tcp_component.super.uri) {
free(mca_ptl_tcp_component.super.uri);
}
mca_ptl_tcp_component.super.uri = strdup(info[n].value.data.string);
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_RETRY_DELAY)) {
mca_ptl_tcp_component.wait_to_connect = info[n].value.data.uint32;
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_MAX_RETRIES)) {
mca_ptl_tcp_component.max_retries = info[n].value.data.uint32;
}
}
}
/* mark that we are using the V2 protocol */
pmix_globals.mypeer->protocol = PMIX_PROTOCOL_V2;
gethostname(myhost, sizeof(myhost));
/* if we were given a URI via MCA param, then look no further */
if (NULL != mca_ptl_tcp_component.super.uri) {
/* if the string starts with "file:", then they are pointing
* us to a file we need to read to get the URI itself */
if (0 == strncmp(mca_ptl_tcp_component.super.uri, "file:", 5)) {
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool getting connection info from %s",
mca_ptl_tcp_component.super.uri);
nspace = NULL;
rc = parse_uri_file(&mca_ptl_tcp_component.super.uri[5], &suri, &nspace, &rank);
if (PMIX_SUCCESS != rc) {
return PMIX_ERR_UNREACH;
}
free(mca_ptl_tcp_component.super.uri);
mca_ptl_tcp_component.super.uri = suri;
} else {
/* we need to extract the nspace/rank of the server from the string */
p = strchr(mca_ptl_tcp_component.super.uri, ';');
if (NULL == p) {
return PMIX_ERR_BAD_PARAM;
}
*p = '\0';
p++;
suri = strdup(p); // save the uri portion
/* the '.' in the first part of the original string separates
* nspace from rank */
p = strchr(mca_ptl_tcp_component.super.uri, '.');
if (NULL == p) {
free(suri);
return PMIX_ERR_BAD_PARAM;
}
*p = '\0';
p++;
nspace = strdup(mca_ptl_tcp_component.super.uri);
rank = strtoull(p, NULL, 10);
/* now update the URI */
free(mca_ptl_tcp_component.super.uri);
mca_ptl_tcp_component.super.uri = suri;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool attempt connect using given URI %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS != (rc = try_connect(&sd))) {
if (NULL != nspace) {
free(nspace);
}
return rc;
}
goto complete;
}
/* if they gave us a pid, then look for it */
if (0 != pid) {
if (0 > asprintf(&filename, "pmix.%s.tool.%d", myhost, pid)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool searching for given session server %s",
filename);
nspace = NULL;
rc = df_search(mca_ptl_tcp_component.system_tmpdir,
filename, &sd, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS == rc) {
goto complete;
}
if (NULL != nspace) {
free(nspace);
}
/* since they gave us a specific pid and we couldn't
* connect to it, return an error */
return PMIX_ERR_UNREACH;
}
/* if they asked for system-level, we start there */
if (system_level || system_level_only) {
if (0 > asprintf(&filename, "%s/pmix.sys.%s", mca_ptl_tcp_component.system_tmpdir, myhost)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool looking for system server at %s",
filename);
/* try to read the file */
rc = parse_uri_file(filename, &suri, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS == rc) {
mca_ptl_tcp_component.super.uri = suri;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool attempt connect to system server at %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS == try_connect(&sd)) {
goto complete;
}
free(nspace);
}
}
/* we get here if they either didn't ask for a system-level connection,
* or they asked for it and it didn't succeed. If they _only_ wanted
* a system-level connection, then we are done */
if (system_level_only) {
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp: connecting to system failed");
return PMIX_ERR_UNREACH;
}
/* they didn't give us a pid, so we will search to see what session-level
* tools are available to this user. We will take the first connection
* that succeeds - this is based on the likelihood that there is only
* one session per user on a node */
if (0 > asprintf(&filename, "pmix.%s.tool", myhost)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool searching for session server %s",
filename);
nspace = NULL;
rc = df_search(mca_ptl_tcp_component.system_tmpdir,
filename, &sd, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS != rc) {
if (NULL != nspace){
free(nspace);
}
return PMIX_ERR_UNREACH;
}
complete:
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"sock_peer_try_connect: Connection across to server succeeded");
/* do a final bozo check */
if (NULL == nspace || PMIX_RANK_WILDCARD == rank) {
if (NULL != nspace) {
free(nspace);
}
CLOSE_THE_SOCKET(sd);
return PMIX_ERR_UNREACH;
}
/* mark the connection as made */
pmix_globals.connected = true;
pmix_client_globals.myserver->sd = sd;
/* setup the server info */
if (NULL == pmix_client_globals.myserver->info) {
pmix_client_globals.myserver->info = PMIX_NEW(pmix_rank_info_t);
}
if (NULL == pmix_client_globals.myserver->nptr) {
pmix_client_globals.myserver->nptr = PMIX_NEW(pmix_nspace_t);
}
if (NULL == pmix_client_globals.myserver->nptr->nspace) {
pmix_client_globals.myserver->nptr->nspace = nspace;
} else {
free(nspace);
}
if (NULL == pmix_client_globals.myserver->info->pname.nspace) {
pmix_client_globals.myserver->info->pname.nspace = strdup(pmix_client_globals.myserver->nptr->nspace);
}
pmix_client_globals.myserver->info->pname.rank = rank;
pmix_ptl_base_set_nonblocking(sd);
/* setup recv event */
pmix_event_assign(&pmix_client_globals.myserver->recv_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_READ | EV_PERSIST,
pmix_ptl_base_recv_handler, pmix_client_globals.myserver);
pmix_client_globals.myserver->recv_ev_active = true;
PMIX_POST_OBJECT(pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver->recv_event, 0);
/* setup send event */
pmix_event_assign(&pmix_client_globals.myserver->send_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_WRITE|EV_PERSIST,
pmix_ptl_base_send_handler, pmix_client_globals.myserver);
pmix_client_globals.myserver->send_ev_active = false;
return PMIX_SUCCESS;
}
PMIX_EXPORT pmix_status_t PMIx_Log_nb(const pmix_info_t data[], size_t ndata,
const pmix_info_t directives[], size_t ndirs,
pmix_op_cbfunc_t cbfunc, void *cbdata)
{
pmix_shift_caddy_t *cd;
pmix_cmd_t cmd = PMIX_LOG_CMD;
pmix_buffer_t *msg;
pmix_status_t rc;
size_t n;
time_t timestamp = 0;
pmix_proc_t *source = NULL;
PMIX_ACQUIRE_THREAD(&pmix_global_lock);
pmix_output_verbose(2, pmix_globals.debug_output,
"pmix:log non-blocking");
if (pmix_globals.init_cntr <= 0) {
PMIX_RELEASE_THREAD(&pmix_global_lock);
return PMIX_ERR_INIT;
}
if (0 == ndata || NULL == data) {
PMIX_RELEASE_THREAD(&pmix_global_lock);
return PMIX_ERR_BAD_PARAM;
}
/* check the directives - if they requested a timestamp, then
* get the time, also look for a source */
if (NULL != directives) {
for (n=0; n < ndirs; n++) {
if (0 == strncmp(directives[n].key, PMIX_LOG_GENERATE_TIMESTAMP, PMIX_MAX_KEYLEN)) {
if (PMIX_INFO_TRUE(&directives[n])) {
/* pickup the timestamp */
timestamp = time(NULL);
}
} else if (0 == strncmp(directives[n].key, PMIX_LOG_SOURCE, PMIX_MAX_KEYLEN)) {
source = directives[n].value.data.proc;
}
}
}
/* if we are a client or tool, we never do this ourselves - we
* always pass this request to our server for execution */
if (!PMIX_PROC_IS_SERVER(pmix_globals.mypeer) &&
!PMIX_PROC_IS_LAUNCHER(pmix_globals.mypeer)) {
/* if we aren't connected, don't attempt to send */
if (!pmix_globals.connected) {
PMIX_RELEASE_THREAD(&pmix_global_lock);
return PMIX_ERR_UNREACH;
}
PMIX_RELEASE_THREAD(&pmix_global_lock);
/* if we are not a server, then relay this request to the server */
cd = PMIX_NEW(pmix_shift_caddy_t);
cd->cbfunc.opcbfn = cbfunc;
cd->cbdata = cbdata;
msg = PMIX_NEW(pmix_buffer_t);
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, &cmd, 1, PMIX_COMMAND);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
/* provide the timestamp - zero will indicate
* that it wasn't taken */
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, ×tamp, 1, PMIX_TIME);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
/* pack the number of data entries */
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, &ndata, 1, PMIX_SIZE);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
if (0 < ndata) {
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, data, ndata, PMIX_INFO);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
}
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, &ndirs, 1, PMIX_SIZE);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
if (0 < ndirs) {
PMIX_BFROPS_PACK(rc, pmix_client_globals.myserver,
msg, directives, ndirs, PMIX_INFO);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(msg);
PMIX_RELEASE(cd);
return rc;
}
}
pmix_output_verbose(2, pmix_plog_base_framework.framework_output,
"pmix:log sending to server");
PMIX_PTL_SEND_RECV(rc, pmix_client_globals.myserver,
msg, log_cbfunc, (void*)cd);
if (PMIX_SUCCESS != rc) {
PMIX_ERROR_LOG(rc);
PMIX_RELEASE(cd);
}
return rc;
}
PMIX_RELEASE_THREAD(&pmix_global_lock);
/* if no recorded source was found, then we must be it */
if (NULL == source) {
source = &pmix_globals.myid;
cd = PMIX_NEW(pmix_shift_caddy_t);
cd->cbfunc.opcbfn = cbfunc;
cd->cbdata = cbdata;
cd->ndirs = ndirs + 1;
PMIX_INFO_CREATE(cd->directives, cd->ndirs);
for (n=0; n < ndirs; n++) {
PMIX_INFO_XFER(&cd->directives[n], (pmix_info_t*)&directives[n]);
}
PMIX_INFO_LOAD(&cd->directives[ndirs], PMIX_LOG_SOURCE, &source, PMIX_PROC);
/* call down to process the request - the various components
* will thread shift as required */
rc = pmix_plog.log(source, data, ndata, cd->directives, cd->ndirs, localcbfunc, cd);
if (PMIX_SUCCESS != rc) {
PMIX_INFO_FREE(cd->directives, cd->ndirs);
PMIX_RELEASE(cd);
}
} else if (0 == strncmp(source->nspace, pmix_globals.myid.nspace, PMIX_MAX_NSLEN) &&
source->rank == pmix_globals.myid.rank) {
/* if I am the recorded source, then this is a re-submission of
* something that got "upcalled" by a prior call. In this case,
* we return a "not supported" error as clearly we couldn't
* handle it, and neither could our host */
rc = PMIX_ERR_NOT_SUPPORTED;
} else {
/* call down to process the request - the various components
* will thread shift as required */
rc = pmix_plog.log(source, data, ndata, directives, ndirs, cbfunc, cbdata);
}
return rc;
}
