1
1
openmpi/orte/runtime/orte_universe_exists.c
Ralph Castain 85df3bd92f Bring in the generalized xcast communication system along with the correspondingly revised orted launch. I will send a message out to developers explaining the basic changes. In brief:
1. generalize orte_rml.xcast to become a general broadcast-like messaging system. Messages can now be sent to any tag on the daemons or processes. Note that any message sent via xcast will be delivered to ALL processes in the specified job - you don't get to pick and choose. At a later date, we will introduce an augmented capability that will use the daemons as relays, but will allow you to send to a specified array of process names.

2. extended orte_rml.xcast so it supports more scalable message routing methodologies. At the moment, we support three: (a) direct, which sends the message directly to all recipients; (b) linear, which sends the message to the local daemon on each node, which then relays it to its own local procs; and (b) binomial, which sends the message via a binomial algo across all the daemons, each of which then relays to its own local procs. The crossover points between the algos are adjustable via MCA param, or you can simply demand that a specific algo be used.

3. orteds no longer exhibit two types of behavior: bootproxy or VM. Orteds now always behave like they are part of a virtual machine - they simply launch a job if mpirun tells them to do so. This is another step towards creating an "orteboot" functionality, but also provided a clean system for supporting message relaying.

Note one major impact of this commit: multiple daemons on a node cannot be supported any longer! Only a single daemon/node is now allowed.

This commit is known to break support for the following environments: POE, Xgrid, Xcpu, Windows. It has been tested on rsh, SLURM, and Bproc. Modifications for TM support have been made but could not be verified due to machine problems at LANL. Modifications for SGE have been made but could not be verified. The developers for the non-verified environments will be separately notified along with suggestions on how to fix the problems.

This commit was SVN r15007.
2007-06-12 13:28:54 +00:00

573 строки
19 KiB
C

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2006 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2007 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* Setup command line options for the Open MPI Run Time Environment
*/
#include "orte_config.h"
#include "orte/orte_constants.h"
#include <string.h>
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#include <sys/types.h>
#ifdef HAVE_DIRENT_H
#include <dirent.h>
#endif /* HAVE_DIRENT_H */
#include "opal/util/output.h"
#include "opal/util/os_path.h"
#include "opal/util/os_dirpath.h"
#include "orte/util/univ_info.h"
#include "orte/util/sys_info.h"
#include "orte/util/proc_info.h"
#include "orte/util/session_dir.h"
#include "orte/util/universe_setup_file_io.h"
#include "orte/runtime/params.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/ns/ns.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/runtime/runtime.h"
static struct timeval ompi_rte_ping_wait = {2, 0};
int orte_universe_search(opal_list_t *universe_list, bool report_broken_files, bool remove_broken_files)
{
int ret, exit_status = ORTE_SUCCESS;
#ifndef __WINDOWS__
DIR *cur_dirp = NULL;
struct dirent * dir_entry;
#else
HANDLE hFind = INVALID_HANDLE_VALUE;
WIN32_FIND_DATA file_data;
#endif /* __WINDOWS__ */
char *univ_setup_filename = NULL;
char *fulldirpath = NULL;
char *prefix = NULL;
char *frontend = NULL;
char *frontend_abs = NULL;
/*
* Get the session directory
*/
if( ORTE_SUCCESS != (ret = orte_session_dir_get_name(&fulldirpath,
&prefix,
&frontend,
orte_system_info.user,
orte_system_info.nodename,
NULL, /* batch ID -- Not used */
NULL, /* Universe Name -- NONE */
NULL, /* jobid */
NULL /* vpid */
) ) ) {
exit_status = ret;
goto cleanup;
}
#if !defined(__WINDOWS__)
frontend_abs = opal_os_path(false, prefix, frontend, NULL);
/*
* Check to make sure we have access to this directory
*/
if( ORTE_SUCCESS != (ret = opal_os_dirpath_access(frontend_abs, 0) )) {
exit_status = ret;
goto cleanup;
}
/*
* Open up the base directory so we can get a listing
*/
if( NULL == (cur_dirp = opendir(frontend_abs)) ) {
exit_status = ORTE_ERROR;
goto cleanup;
}
/*
* For each directory/universe
*/
while( NULL != (dir_entry = readdir(cur_dirp)) ) {
orte_universe_t *univ = NULL;
/*
* Skip non-universe directories
*/
if( 0 == strncmp(dir_entry->d_name, ".", strlen(".")) ||
0 == strncmp(dir_entry->d_name, ".", strlen("..")) ) {
continue;
}
/*
* Read the setup file
*/
univ_setup_filename = opal_os_path( false, frontend_abs,
dir_entry->d_name, "universe-setup.txt", NULL );
univ = OBJ_NEW(orte_universe_t);
if(ORTE_SUCCESS != (ret = orte_read_universe_setup_file(univ_setup_filename, univ) ) ){
if (report_broken_files) {
printf("universe_search: Unable to read the file (%s)\n", univ_setup_filename);
exit_status = ret;
}
/*
* See if we want to remove any cases with broken
* universe-setup.txt files. If so, print out a message and
* remove the directory. This is used by the orte-clean
* routine.
*/
if (remove_broken_files) {
char *univ_directory;
univ_directory = opal_os_path(false, frontend_abs,
dir_entry->d_name, NULL);
printf("universe_search: Removing defunct directory (%s)\n", univ_directory);
opal_os_dirpath_destroy(univ_directory, true, NULL);
free(univ_directory);
}
OBJ_RELEASE(univ);
} else {
OBJ_RETAIN(univ);
opal_list_append(universe_list, &(univ->super));
}
}
#else
/*
* Open up the base directory so we can get a listing.
*
* On Windows if we want to parse the content of a directory the filename
* should end with the "*". Otherwise we will only open the directory
* structure (and not the content).
*/
frontend_abs = opal_os_path(false, prefix, frontend, "*", NULL);
hFind = FindFirstFile( frontend_abs, &file_data );
if( INVALID_HANDLE_VALUE == hFind ) {
exit_status = GetLastError();
exit_status = ORTE_ERR_NOT_FOUND;
goto cleanup;
}
/*
* For each directory/universe
*/
do {
orte_universe_t *univ = NULL;
/*
* Skip non-universe directories
*/
if( 0 == strncmp(file_data.cFileName, ".", strlen(".")) ||
0 == strncmp(file_data.cFileName, ".", strlen("..")) ) {
continue;
}
/*
* Read the setup file
*/
univ_setup_filename = opal_os_path( false, prefix, frontend,
file_data.cFileName, "universe-setup.txt", NULL);
univ = OBJ_NEW(orte_universe_t);
if(ORTE_SUCCESS != (ret = orte_read_universe_setup_file(univ_setup_filename, univ) ) ){
if (report_broken_files) {
printf("universe_search: Unable to read the file (%s)\n", univ_setup_filename);
exit_status = ret;
}
/*
* See if we want to remove any cases with broken
* universe-setup.txt files. If so, print out a message and
* remove the directory. This is used by the orte-clean
* routine.
*/
if (remove_broken_files) {
char *univ_directory;
univ_directory = opal_os_path(false, frontend_abs,
file_data.cFileName, NULL);
printf("universe_search: Removing defunct directory (%s)\n", univ_directory);
opal_os_dirpath_destroy(univ_directory, true, NULL);
free(univ_directory);
}
OBJ_RELEASE(univ);
} else {
OBJ_RETAIN(univ);
opal_list_append(universe_list, &(univ->super));
}
} while( 0 != FindNextFile( hFind, &file_data ) );
#endif /* !defined(__WINDOWS__) */
cleanup:
#ifndef __WINDOWS__
if( NULL != cur_dirp )
closedir(cur_dirp);
#else
FindClose(hFind);
#endif /* __WINDOWS__ */
if( NULL != univ_setup_filename)
free(univ_setup_filename);
if( NULL != fulldirpath)
free(fulldirpath);
if( NULL != prefix)
free(prefix);
if( NULL != frontend)
free(frontend);
if( NULL != frontend_abs)
free(frontend_abs);
return (opal_list_is_empty(universe_list) ? exit_status : ORTE_SUCCESS);
}
static int orte_universe_check_connect(orte_universe_t *uni)
{
int rc;
if (!orte_universe_info.console) { /* if we aren't trying to connect a console */
if (!uni->persistence || /* if the target universe is not persistent... */
(0 == strncmp(uni->scope, "exclusive", strlen("exclusive")))) { /* ...or no connection allowed */
/* also need to check "local" and that we did not specify the exact
* matching universe name
*/
if (orte_debug_flag) {
opal_output(0, "connect_uni: connection not allowed");
}
/* NOTE: THIS IS NOT AN ERROR - DON'T ERROR_LOG IT */
return ORTE_ERR_NO_CONNECTION_ALLOWED;
}
}
if (orte_debug_flag) {
opal_output(0, "connect_uni: contact info to set: %s", uni->seed_uri);
}
/* insert the universe contact info into the RML hash tables */
if (ORTE_SUCCESS != (rc = orte_rml.set_uri(uni->seed_uri))) {
ORTE_ERROR_LOG(rc);
return(rc);
}
/* ping to verify it's alive */
if (ORTE_SUCCESS != (rc = orte_rml.ping(uni->seed_uri, &ompi_rte_ping_wait))) {
if (orte_debug_flag) {
ORTE_ERROR_LOG(rc);
}
return ORTE_ERR_CONNECTION_FAILED;
}
return ORTE_SUCCESS;
}
int orte_universe_exists(orte_universe_t *univ)
{
char *contact_file;
opal_list_t universes;
opal_list_item_t *item;
orte_universe_t *uniptr;
int ret;
/* if the user didn't provide a name for our universe, then we have to check
* for other universe names we could join. It is virtually impossible for
* another universe to have our exact default universe name as they would
* have to have the same PID - and that would be bad in so many ways!
*/
if (orte_universe_info.default_name) {
/* if we just have the default name - i.e., no name was specified -
* then get a list of all universes known on the local system. All
* we can do here is just loop through the session directory tree
* for universes - we have no better discovery mechanism at this time
*/
OBJ_CONSTRUCT(&universes, opal_list_t);
if (ORTE_SUCCESS != (ret = orte_universe_search(&universes, false, false))) {
/* if nothing was found, that's okay - report anything else */
if (ORTE_ERR_NOT_FOUND != ret) {
ORTE_ERROR_LOG(ret);
}
return ret;
}
/* if the list is empty, then we can just return */
if (opal_list_is_empty(&universes)) return ORTE_ERR_NOT_FOUND;
/* we have no real criteria for picking one over the other, so
* we just loop through the returned objects and pick the first
* one that will support connection
*/
while (NULL != (item = opal_list_remove_first(&universes))) {
uniptr = (orte_universe_t*)item;
if (ORTE_SUCCESS == orte_universe_check_connect(uniptr)) {
univ->name = strdup(uniptr->name);
univ->host = strdup(uniptr->host);
univ->uid = strdup(uniptr->uid);
univ->persistence = uniptr->persistence;
univ->scope = strdup(uniptr->scope);
univ->seed_uri = strdup(uniptr->seed_uri);
univ->console_connected = uniptr->console_connected;
return ORTE_SUCCESS;
}
}
/* if we get here, then we did not success in connecting to
* anyone - report that situation
*/
return ORTE_ERR_NOT_FOUND;
}
/* if the user did provide a name, then see if we can join it */
/* check to see if local universe session directory already exists */
if (ORTE_SUCCESS != orte_session_dir(false,
orte_process_info.tmpdir_base,
orte_system_info.user,
orte_system_info.nodename,
NULL,
orte_universe_info.name,
NULL,
NULL)) { /* not found */
/* NOTE: NOT FINDING THE DIRECTORY IS NOT AN ERROR - DON'T ERROR_LOG IT */
return ORTE_ERR_NOT_FOUND;
}
/* check for "contact-info" file. if present, read it in. */
if (NULL == (contact_file = opal_os_path(false, orte_process_info.universe_session_dir,
"universe-setup.txt", NULL))) {
/* NOTE: NOT FINDING THE FILE IS NOT AN ERROR - DON'T ERROR_LOG IT */
return ORTE_ERR_NOT_FOUND;
}
if (ORTE_SUCCESS != (ret = orte_read_universe_setup_file(contact_file, univ))) {
/* NOTE: THIS IS NOT AN ERROR - DON'T ERROR_LOG IT */
free(contact_file);
return ret;
}
/* don't need this string any more - free it */
free(contact_file);
if (orte_debug_flag) {
opal_output(0, "connect_uni: contact info read");
}
return orte_universe_check_connect(univ);
}
void
orte_universe_clean_directories(char *my_universe, int verbose) {
char *session_dir = NULL;
#if !defined(__WINDOWS__)
DIR *cur_dirp = NULL;
struct dirent * dir_entry;
#else
HANDLE hFind = INVALID_HANDLE_VALUE;
WIN32_FIND_DATA file_data;
#endif /* __WINDOWS__ */
char *fulldirpath = NULL;
char *prefix = NULL;
char *frontend = NULL;
/*
* Compute the full pathname to the session directory.
*/
if (ORTE_SUCCESS != orte_session_dir_get_name(&fulldirpath,
&prefix,
&frontend,
orte_system_info.user,
orte_system_info.nodename,
NULL, /* batch ID -- Not used */
NULL, /* Universe Name -- NONE */
NULL, /* jobid */
NULL /* vpid */
)) {
goto cleanup;
}
#if !defined(__WINDOWS__)
session_dir = opal_os_path(false, prefix, frontend, NULL);
/*
* Free up the various strings as these are allocated within
* the previous function.
*/
if (NULL != fulldirpath) {
free(fulldirpath);
fulldirpath = NULL;
}
if (NULL != prefix) {
free(prefix);
prefix = NULL;
}
if (NULL != frontend) {
free(frontend);
frontend = NULL;
}
/*
* Check to make sure we have access to this directory
*/
if (ORTE_SUCCESS != opal_os_dirpath_access(session_dir, 0)) {
goto cleanup;
}
/*
* Open up the base directory so we can get a listing
*/
if (NULL == (cur_dirp = opendir(session_dir))) {
goto cleanup;
}
/*
* For each directory/universe
*/
while (NULL != (dir_entry = readdir(cur_dirp))) {
/*
* Skip non-universe directories
*/
if (0 == strncmp(dir_entry->d_name, ".", strlen(".")) ||
0 == strncmp(dir_entry->d_name, ".", strlen(".."))) {
continue;
}
/*
* Skip my own universe. Let normal cleanup take care of that.
*/
if ((0 == strcmp(dir_entry->d_name, my_universe)) &&
(strlen(dir_entry->d_name) == strlen(my_universe))) {
if (verbose) {
opal_output(0, "orte-clean: skipping ourselves, name=%s\n",
orte_universe_info.name);
}
continue;
}
if (ORTE_SUCCESS != orte_session_dir_get_name(&fulldirpath,
&prefix,
&frontend,
orte_system_info.user,
orte_system_info.nodename,
NULL, /* batch ID -- Not used */
dir_entry->d_name,
NULL, /* jobid */
NULL /* vpid */
)) {
continue;
}
if (verbose) {
opal_output(0, "orte-clean: removing directory %s\n", fulldirpath);
}
opal_os_dirpath_destroy(fulldirpath, true, NULL);
/*
* The orte_session_dir_get_name handles the freeing of the
* fulldirpath each time it is called. The prefix gets reused.
* So, there is no need to free them on each call.
*/
if (NULL != frontend) {
free(frontend);
}
}
#else
/*
* Open up the base directory so we can get a listing.
*
* On Windows if we want to parse the content of a directory the filename
* should end with the "*". Otherwise we will only open the directory
* structure (and not the content).
*/
frontend = opal_os_path(false, prefix, frontend, "*", NULL);
hFind = FindFirstFile (frontend, &file_data);
if (INVALID_HANDLE_VALUE == hFind) {
goto cleanup;
}
do {
/*
* Skip non-universe directories
*/
/* Skip . and .. */
if ((0 == strcmp(file_data.cFileName, ".")) ||
(0 == strcmp(file_data.cFileName, "..")) ) {
continue;
}
if ((0 == strcmp(file_data.cFileName, my_universe)) &&
(strlen(file_data.cFileName) == strlen(my_universe))) {
if (verbose) {
opal_output(0, "orte-clean: skipping ourseleves, name=%s\n",
orte_universe_info.name);
}
continue;
}
if (ORTE_SUCCESS != orte_session_dir_get_name(&fulldirpath,
&prefix,
&frontend,
orte_system_info.user,
orte_system_info.nodename,
NULL, /* batch ID -- Not used */
file_data.cFileName,
NULL, /* jobid */
NULL /* vpid */
)) {
continue;
}
if (verbose) {
opal_output(0, "orte-clean: removing directory %s\n", fulldirpath);
}
opal_os_dirpath_destroy(fulldirpath, true, NULL);
/*
* The orte_session_dir_get_name handles the freeing of the
* fulldirpath each time it is called. The prefix gets reused.
* So, there is no need to free them on each call.
*/
if (NULL != frontend)
free(frontend);
} while (0 != FindNextFile(hFind, &file_data));
#endif /* !defined(__WINDOWS__) */
#if !defined(__WINDOWS__)
if (NULL != cur_dirp) {
closedir(cur_dirp);
}
#else
FindClose(hFind);
#endif /* __WINDOWS__ */
if(NULL != fulldirpath) {
free(fulldirpath);
}
if(NULL != prefix) {
free(prefix);
}
/*
* If the session directory is empty, then remove that too
*/
opal_os_dirpath_destroy(session_dir, false, NULL);
free(session_dir);
cleanup:
return;
}