ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Remove stale code

Просмотр исходного кода 
This commit was SVN r26252.
Этот коммит содержится в:
Ralph Castain 2012-04-07 13:33:40 +00:00
родитель 93bbeabc55
Коммит 19630ca28d
7 изменённых файлов: 0 добавлений и 1486 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

0
orte/mca/plm/submit/.ompi_ignore
Просмотреть файл

46
orte/mca/plm/submit/Makefile.am
Просмотреть файл

@ -1,46 +0,0 @@
#
# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2005 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) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
dist_pkgdata_DATA = help-plm-rsh.txt
sources = \
plm_rsh.h \
plm_rsh_component.c \
plm_rsh_module.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if MCA_BUILD_orte_plm_rsh_DSO
component_noinst =
component_install = mca_plm_rsh.la
else
component_noinst = libmca_plm_rsh.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_plm_rsh_la_SOURCES = $(sources)
mca_plm_rsh_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_plm_rsh_la_SOURCES =$(sources)
libmca_plm_rsh_la_LDFLAGS = -module -avoid-version

24
orte/mca/plm/submit/configure.m4
Просмотреть файл

@ -1,24 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2004-2006 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_plm_submit_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_plm_submit_CONFIG],[
AC_CONFIG_FILES([orte/mca/plm/submit/Makefile])
AC_CHECK_FUNC([fork], [plm_submit_happy="yes"], [plm_submit_happy="no"])
AS_IF([test "$plm_submit_happy" = "yes" -a "$orte_without_full_support" = 0], [$1], [$2])
])dnl

66
orte/mca/plm/submit/help-pls-submit.txt
Просмотреть файл

@ -1,66 +0,0 @@
# -*- text -*-
#
# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2005 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.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# This is the US/English general help file for Open RTE's orterun.
#
[no-local-orted]
The rsh PLS component was not able to find the executable "orted" in
your PATH or in the directory where Open MPI/OpenRTE was initially installed,
and therefore cannot continue.
For reference, your current PATH is:
%s
We also looked for orted in the following directory:
%s
[multiple-prefixes]
Specified multiple application contexts using different
settings for --prefix. Care should be taken, that corresponding
processes are mapped to different nodes. Having multiple prefixes
per node is not allowed.
The previously set prefix was
%s
the prefix to be set overriding:
%s
[concurrency-less-than-zero]
The value of the MCA parameter "pls_rsh_num_concurrent" is less than
or equal to zero (%d). This parameter is used to determine how many
remote agents (typically rsh or ssh) to invoke concurrently while
launching parallel jobs.
This value has automatically be reset to 1; processing will continue.
[deadlock-params]
The rsh launcher has been given a number of %d concurrent daemons to
launch and is in a debug-daemons option. However, the total number of
daemons to launch (%d) is greater than this value. This is a scenario that
will cause the system to deadlock.
To avoid deadlock, either increase the number of concurrent daemons, or
remove the debug-daemons flag.
[unknown-user]
The user (%d) is unknown to the system (i.e. there is no corresponding
entry in the password file). Please contact your system administrator
for a fix.

75
orte/mca/plm/submit/pls_submit.h
Просмотреть файл

@ -1,75 +0,0 @@
/*
* Copyright (c) 2004-2007 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2008 The Trustees of Indiana University.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file:
* Part of the submit launcher. See plm_submit.h for an overview of how it works.
*/
#ifndef ORTE_PLM_SUBMIT_EXPORT_H
#define ORTE_PLM_SUBMIT_EXPORT_H
#include "orte_config.h"
#include "opal/mca/mca.h"
#include "orte/mca/plm/plm.h"
#include "opal/threads/condition.h"
BEGIN_C_DECLS
/*
* Module open / close
*/
int orte_plm_submit_component_open(void);
int orte_plm_submit_component_close(void);
int orte_plm_submit_component_query(mca_base_module_t **module, int *priority);
/*
* Startup / Shutdown
*/
int orte_plm_submit_finalize(void);
/*
* Interface
*/
int orte_plm_submit_launch(orte_job_t*);
int orte_plm_submit_terminate_orteds(void);
int orte_plm_submit_signal_job(orte_jobid_t, int32_t);
/**
* PLM Component
*/
struct orte_plm_submit_component_t {
orte_plm_base_component_t super;
bool debug;
bool debug_daemons;
bool timing;
int delay;
int priority;
char *agent_param;
char** agent_argv;
int agent_argc;
char* agent_path;
char* orted;
orte_std_cntr_t num_children;
orte_std_cntr_t num_concurrent;
opal_mutex_t lock;
opal_condition_t cond;
};
typedef struct orte_plm_submit_component_t orte_plm_submit_component_t;
ORTE_MODULE_DECLSPEC extern orte_plm_submit_component_t mca_plm_submit_component;
extern orte_plm_base_module_t orte_plm_submit_module;
END_C_DECLS
#endif /* ORTE_PLM_SUBMIT_EXPORT_H */

282
orte/mca/plm/submit/pls_submit_component.c
Просмотреть файл

@ -1,282 +0,0 @@
/*
* Copyright (c) 2004-2007 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2008 The Trustees of Indiana University.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
* These symbols are in a file by themselves to provide nice linker
* semantics. Since linkers generally pull in symbols by object
* files, keeping these symbols as the only symbols in this file
* prevents utility programs such as "ompi_info" from having to import
* entire components just to query their version and parameters.
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <stdlib.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <ctype.h>
#include "opal/util/opal_environ.h"
#include "opal/util/argv.h"
#include "opal/util/path.h"
#include "opal/util/basename.h"
#include "opal/mca/base/mca_base_param.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/plm/submit/plm_submit.h"
/*
* Local function
*/
static char **search(const char* agent_list);
/*
* Public string showing the plm ompi_submit component version number
*/
const char *mca_plm_submit_component_version_string =
"Open MPI submit plm MCA component version " ORTE_VERSION;
/*
* Instantiate the public struct with all of our public information
* and pointers to our public functions in it
*/
orte_plm_submit_component_t mca_plm_submit_component = {
{
/* First, the mca_component_t struct containing meta information
about the component itself */
{
ORTE_PLM_BASE_VERSION_2_0_0,
/* Component name and version */
"submit",
ORTE_MAJOR_VERSION,
ORTE_MINOR_VERSION,
ORTE_RELEASE_VERSION,
/* Component open and close functions */
orte_plm_submit_component_open,
orte_plm_submit_component_close,
orte_plm_submit_component_query
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
}
};
int orte_plm_submit_component_open(void)
{
int tmp, value;
mca_base_component_t *c = &mca_plm_submit_component.super.base_version;
/* initialize globals */
OBJ_CONSTRUCT(&mca_plm_submit_component.lock, opal_mutex_t);
OBJ_CONSTRUCT(&mca_plm_submit_component.cond, opal_condition_t);
mca_plm_submit_component.num_children = 0;
mca_plm_submit_component.agent_argv = NULL;
mca_plm_submit_component.agent_argc = 0;
mca_plm_submit_component.agent_path = NULL;
/* lookup parameters */
mca_base_param_reg_int(c, "debug",
"Whether or not to enable debugging output for the submit plm component (0 or 1)",
false, false, false, &tmp);
mca_plm_submit_component.debug = OPAL_INT_TO_BOOL(tmp);
mca_base_param_reg_int(c, "num_concurrent",
"How many plm_submit_agent instances to invoke concurrently (must be > 0)",
false, false, 128, &tmp);
if (tmp <= 0) {
orte_show_help("help-plm-submit.txt", "concurrency-less-than-zero",
true, tmp);
tmp = 1;
}
mca_plm_submit_component.num_concurrent = tmp;
if (mca_plm_submit_component.debug == 0) {
mca_base_param_reg_int_name("orte", "debug",
"Whether or not to enable debugging output for all ORTE components (0 or 1)",
false, false, false, &tmp);
mca_plm_submit_component.debug = OPAL_INT_TO_BOOL(tmp);
}
mca_base_param_reg_int_name("orte", "debug_daemons",
"Whether or not to enable debugging of daemons (0 or 1)",
false, false, false, &tmp);
mca_plm_submit_component.debug_daemons = OPAL_INT_TO_BOOL(tmp);
tmp = mca_base_param_reg_int_name("orte", "timing",
"Request that critical timing loops be measured",
false, false, 0, &value);
if (value != 0) {
mca_plm_submit_component.timing = true;
} else {
mca_plm_submit_component.timing = false;
}
mca_base_param_reg_string(c, "orted",
"The command name that the submit plm component will invoke for the ORTE daemon",
false, false, "orted",
&mca_plm_submit_component.orted);
mca_base_param_reg_int(c, "priority",
"Priority of the submit plm component",
false, false, 10,
&mca_plm_submit_component.priority);
mca_base_param_reg_int(c, "delay",
"Delay (in seconds) between invocations of the remote agent, but only used when the \"debug\" MCA parameter is true, or the top-level MCA debugging is enabled (otherwise this value is ignored)",
false, false, 1,
&mca_plm_submit_component.delay);
mca_base_param_reg_int(c, "assume_same_shell",
"If set to 1, assume that the shell on the remote node is the same as the shell on the local node. Otherwise, probe for what the remote shell.",
false, false, 1, &tmp);
mca_plm_submit_component.assume_same_shell = OPAL_INT_TO_BOOL(tmp);
mca_base_param_reg_string(c, "agent",
"The command used to launch executables on remote nodes (typically either \"ssh\" or \"submit\")",
false, false, "ssh : submit",
&mca_plm_submit_component.agent_param);
return ORTE_SUCCESS;
}
int orte_plm_submit_component_query(mca_base_module_t **module, int *priority)
{
char *bname;
size_t i;
/* Take the string that was given to us by the pla_submit_agent MCA
param and search for it */
mca_plm_submit_component.agent_argv =
search(mca_plm_submit_component.agent_param);
mca_plm_submit_component.agent_argc =
opal_argv_count(mca_plm_submit_component.agent_argv);
mca_plm_submit_component.agent_path = NULL;
if (mca_plm_submit_component.agent_argc > 0) {
/* If the agent is ssh, and debug was not selected, then
automatically add "-x" */
bname = opal_basename(mca_plm_submit_component.agent_argv[0]);
if (NULL != bname && 0 == strcmp(bname, "ssh") &&
mca_plm_submit_component.debug == 0) {
for (i = 1; NULL != mca_plm_submit_component.agent_argv[i]; ++i) {
if (0 == strcasecmp("-x",
mca_plm_submit_component.agent_argv[i])) {
break;
}
}
if (NULL == mca_plm_submit_component.agent_argv[i]) {
opal_argv_append(&mca_plm_submit_component.agent_argc,
&mca_plm_submit_component.agent_argv, "-x");
}
}
if (NULL != bname) {
free(bname);
}
}
/* If we didn't find the agent in the path, then don't use this
component */
if (NULL == mca_plm_submit_component.agent_argv ||
NULL == mca_plm_submit_component.agent_argv[0]) {
*module = NULL;
return ORTE_ERROR:
}
mca_plm_submit_component.agent_path =
opal_path_findv(mca_plm_submit_component.agent_argv[0], X_OK,
environ, NULL);
if (NULL == mca_plm_submit_component.agent_path) {
*module = NULL;
return ORTE_ERROR:
}
*priority = mca_plm_submit_component.priority;
*module = (mca_base_module_t *) &orte_plm_submit_module;
return ORTE_SUCCESS;
}
int orte_plm_submit_component_close(void)
{
/* cleanup state */
OBJ_DESTRUCT(&mca_plm_submit_component.lock);
OBJ_DESTRUCT(&mca_plm_submit_component.cond);
if (NULL != mca_plm_submit_component.orted) {
free(mca_plm_submit_component.orted);
}
if (NULL != mca_plm_submit_component.agent_param) {
free(mca_plm_submit_component.agent_param);
}
if (NULL != mca_plm_submit_component.agent_argv) {
opal_argv_free(mca_plm_submit_component.agent_argv);
}
if (NULL != mca_plm_submit_component.agent_path) {
free(mca_plm_submit_component.agent_path);
}
return ORTE_SUCCESS;
}
/*
* Take a colon-delimited list of agents and locate the first one that
* we are able to find in the PATH. Split that one into argv and
* return it. If nothing found, then return NULL.
*/
static char **search(const char* agent_list)
{
int i, j;
char *line, **lines = opal_argv_split(agent_list, ':');
char **tokens, *tmp;
char cwd[PATH_MAX];
getcwd(cwd, PATH_MAX);
for (i = 0; NULL != lines[i]; ++i) {
line = lines[i];
/* Trim whitespace at the beginning and end of the line */
for (j = 0; '\0' != line[j] && isspace(line[j]); ++line) {
continue;
}
for (j = strlen(line) - 2; j > 0 && isspace(line[j]); ++j) {
line[j] = '\0';
}
if (strlen(line) <= 0) {
continue;
}
/* Split it */
tokens = opal_argv_split(line, ' ');
/* Look for the first token in the PATH */
tmp = opal_path_findv(tokens[0], X_OK, environ, cwd);
if (NULL != tmp) {
free(tokens[0]);
tokens[0] = tmp;
opal_argv_free(lines);
return tokens;
}
/* Didn't find it */
opal_argv_free(tokens);
}
/* Doh -- didn't find anything */
opal_argv_free(lines);
return NULL;
}

993
orte/mca/plm/submit/pls_submit_module.c
Просмотреть файл

@ -1,993 +0,0 @@
/*
* Copyright (c) 2004-2008 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
* These symbols are in a file by themselves to provide nice linker
* semantics. Since linkers generally pull in symbols by object
* files, keeping these symbols as the only symbols in this file
* prevents utility programs such as "ompi_info" from having to import
* entire components just to query their version and parameters.
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <stdlib.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <errno.h>
#include <string.h>
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#include <fcntl.h>
#include <signal.h>
#ifdef HAVE_PWD_H
#include <pwd.h>
#endif
#include "opal/mca/installdirs/installdirs.h"
#include "opal/mca/base/mca_base_param.h"
#include "opal/util/if.h"
#include "opal/util/output.h"
#include "opal/util/os_path.h"
#include "opal/util/path.h"
#include "opal/mca/event/event.h"
#include "opal/util/argv.h"
#include "opal/util/opal_environ.h"
#include "orte/util/show_help.h"
#include "opal/util/basename.h"
#include "opal/util/opal_environ.h"
#include "orte/util/proc_info.h"
#include "orte/util/univ_info.h"
#include "orte/runtime/orte_wait.h"
#include "orte/runtime/orte_globals.h"
#include "orte/runtime/params.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ras/ras_types.h"
#include "orte/mca/rmaps/rmaps.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/base.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/plm/submit/plm_submit.h"
static int orte_plm_submit_init(void);
orte_plm_base_module_t orte_plm_submit_module = {
orte_plm_submit_init,
orte_plm_base_set_hnp_name,
orte_plm_submit_launch,
NULL,
orte_plm_base_orted_terminate_job,
orte_plm_submit_terminate_orteds,
orte_plm_base_orted_kill_local_procs,
orte_plm_submit_signal_job,
orte_plm_submit_finalize
};
static void set_handler_default(int sig);
enum {
ORTE_PLM_submit_SHELL_BASH = 0,
ORTE_PLM_submit_SHELL_ZSH,
ORTE_PLM_submit_SHELL_TCSH,
ORTE_PLM_submit_SHELL_CSH,
ORTE_PLM_submit_SHELL_KSH,
ORTE_PLM_submit_SHELL_SH,
ORTE_PLM_submit_SHELL_UNKNOWN
};
typedef int orte_plm_submit_shell;
static const char * orte_plm_submit_shell_name[] = {
"bash",
"zsh",
"tcsh", /* tcsh has to be first otherwise strstr finds csh */
"csh",
"ksh",
"sh",
"unknown"
};
/* local global storage of timing variables */
static struct timeval joblaunchstart, joblaunchstop;
/* global storage of active jobid being launched */
static orte_jobid_t active_job=ORTE_JOBID_INVALID;
/*
* Init module
*/
static int orte_plm_submit_init(void)
{
int rc;
if (ORTE_SUCCESS != (rc = orte_plm_base_comm_start())) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
/**
* Check the Shell variable on the specified node
*/
static int orte_plm_submit_probe(orte_node_t *node, orte_plm_submit_shell * shell)
{
char ** argv;
int argc, rc = ORTE_SUCCESS, i;
int fd[2];
pid_t pid;
char outbuf[4096];
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: going to check SHELL variable on node %s\n",
node->name);
}
*shell = ORTE_PLM_submit_SHELL_UNKNOWN;
if (pipe(fd)) {
opal_output(0, "plm:submit: pipe failed with errno=%d\n", errno);
return ORTE_ERR_IN_ERRNO;
}
if ((pid = fork()) < 0) {
opal_output(0, "plm:submit: fork failed with errno=%d\n", errno);
return ORTE_ERR_IN_ERRNO;
}
else if (pid == 0) { /* child */
if (dup2(fd[1], 1) < 0) {
opal_output(0, "plm:submit: dup2 failed with errno=%d\n", errno);
exit(01);
}
/* Build argv array */
argv = opal_argv_copy(mca_plm_submit_component.agent_argv);
argc = mca_plm_submit_component.agent_argc;
opal_argv_append(&argc, &argv, node->name);
opal_argv_append(&argc, &argv, "echo $SHELL");
execvp(argv[0], argv);
exit(errno);
}
if (close(fd[1])) {
opal_output(0, "plm:submit: close failed with errno=%d\n", errno);
return ORTE_ERR_IN_ERRNO;
}
{
ssize_t ret = 1;
char* ptr = outbuf;
size_t outbufsize = sizeof(outbuf);
do {
ret = read (fd[0], ptr, outbufsize-1);
if (ret < 0) {
if (errno == EINTR)
continue;
opal_output( 0, "Unable to detect the remote shell (error %s)\n",
strerror(errno) );
rc = ORTE_ERR_IN_ERRNO;
break;
}
if( outbufsize > 1 ) {
outbufsize -= ret;
ptr += ret;
}
} while( 0 != ret );
*ptr = '\0';
}
close(fd[0]);
if( outbuf[0] != '\0' ) {
char *sh_name = rindex(outbuf, '/');
if( NULL != sh_name ) {
sh_name++; /* skip '/' */
/* We cannot use "echo -n $SHELL" because -n is not portable. Therefore
* we have to remove the "\n" */
if ( sh_name[strlen(sh_name)-1] == '\n' ) {
sh_name[strlen(sh_name)-1] = '\0';
}
/* Search for the substring of known shell-names */
for (i = 0; i < (int)(sizeof (orte_plm_submit_shell_name)/
sizeof(orte_plm_submit_shell_name[0])); i++) {
if ( 0 == strcmp(sh_name, orte_plm_submit_shell_name[i]) ) {
*shell = i;
break;
}
}
}
}
if (mca_plm_submit_component.debug) {
if( ORTE_PLM_submit_SHELL_UNKNOWN == *shell ) {
opal_output(0, "plm:submit: node:%s has unhandled SHELL\n",
node->name);
} else {
opal_output(0, "plm:submit: node:%s has SHELL: %s\n",
node->name, orte_plm_submit_shell_name[*shell]);
}
}
return rc;
}
/**
* Fill the exec_path variable with the directory to the orted
*/
static int orte_plm_submit_fill_exec_path ( char ** exec_path)
{
struct stat buf;
asprintf(exec_path, "%s/orted", opal_install_dirs.bindir);
if (0 != stat(*exec_path, &buf)) {
char *path = getenv("PATH");
if (NULL == path) {
path = ("PATH is empty!");
}
orte_show_help("help-plm-submit.txt", "no-local-orted",
true, path, opal_install_dirs.bindir);
return ORTE_ERR_NOT_FOUND;
}
return ORTE_SUCCESS;
}
/**
* Callback on daemon exit.
*/
static void orte_plm_submit_wait_daemon(pid_t pid, int status, void* cbdata)
{
unsigned long deltat;
if (! WIFEXITED(status) || ! WEXITSTATUS(status) == 0) {
/* tell the user something went wrong */
opal_output(0, "ERROR: A daemon failed to start as expected.");
opal_output(0, "ERROR: There may be more information available from");
opal_output(0, "ERROR: the remote shell (see above).");
if (WIFEXITED(status)) {
opal_output(0, "ERROR: The daemon exited unexpectedly with status %d.",
WEXITSTATUS(status));
} else if (WIFSIGNALED(status)) {
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
opal_output(0, "The daemon received a signal %d (with core).",
WTERMSIG(status));
} else {
opal_output(0, "The daemon received a signal %d.", WTERMSIG(status));
}
#else
opal_output(0, "The daemon received a signal %d.", WTERMSIG(status));
#endif /* WCOREDUMP */
} else {
opal_output(0, "No extra status information is available: %d.", status);
}
/* The usual reasons for ssh to exit abnormally all are a pretty good
indication that the child processes aren't going to start up properly.
Set the job state to indicate we failed to launch so orterun's exit status
will be non-zero and forcibly terminate the job so orterun can exit
*/
orte_errmgr.update_state(active_job, ORTE_JOB_STATE_FAILED_TO_START,
NULL, ORTE_PROC_STATE_UNDEF, 0, status);
} /* if abnormal exit */
/* release any waiting threads */
OPAL_THREAD_LOCK(&mca_plm_submit_component.lock);
if (mca_plm_submit_component.num_children-- >=
mca_plm_submit_component.num_concurrent ||
mca_plm_submit_component.num_children == 0) {
opal_condition_signal(&mca_plm_submit_component.cond);
}
if (mca_plm_submit_component.timing && mca_plm_submit_component.num_children == 0) {
if (0 != gettimeofday(&joblaunchstop, NULL)) {
opal_output(0, "plm_submit: could not obtain job launch stop time");
} else {
deltat = (joblaunchstop.tv_sec - joblaunchstart.tv_sec)*1000000 +
(joblaunchstop.tv_usec - joblaunchstart.tv_usec);
opal_output(0, "plm_submit: total time to launch job is %lu usec", deltat);
}
}
OPAL_THREAD_UNLOCK(&mca_plm_submit_component.lock);
}
/**
* Launch a daemon (bootproxy) on each node. The daemon will be responsible
* for launching the application.
*/
/* When working in this function, ALWAYS jump to "cleanup" if
* you encounter an error so that orterun will be woken up and
* the job can cleanly terminate
*/
int orte_plm_submit_launch(orte_job_t *jdata)
{
orte_job_map_t *map;
orte_std_cntr_t num_nodes;
int node_name_index1;
int proc_vpid_index;
int local_exec_index, local_exec_index_end;
char *vpid_string = NULL;
char *param;
char **argv = NULL;
char *prefix_dir;
int argc;
int rc;
sigset_t sigs;
struct passwd *p;
bool remote_sh = false, remote_csh = false;
bool local_sh = false, local_csh = false;
char *lib_base = NULL, *bin_base = NULL;
bool failed_launch = true;
orte_app_context_t **apps;
orte_node_t **nodes;
orte_std_cntr_t nnode;
if (mca_plm_submit_component.timing) {
if (0 != gettimeofday(&joblaunchstart, NULL)) {
opal_output(0, "plm_submit: could not obtain start time");
joblaunchstart.tv_sec = 0;
joblaunchstart.tv_usec = 0;
}
}
/* setup the job */
if (ORTE_SUCCESS != (rc = orte_plm_base_setup_job(jdata))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
OPAL_OUTPUT_VERBOSE((1, orte_plm_globals.output,
"%s plm:submit: launching job %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(jdata->jobid)));
/* set the active jobid */
active_job = jobid;
/* Get the map for this job */
if (NULL == (map = orte_rmaps.get_job_map(active_job))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
goto cleanup;
}
apps = (orte_app_context_t**)jdata->apps->addr;
nodes = (orte_node_t**)map->nodes->addr;
/* account for any reuse of daemons */
if (ORTE_SUCCESS != (rc = orte_plm_base_launch_on_existing_daemons(map))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
num_nodes = map->num_new_daemons;
if (0 == num_nodes) {
/* have all the daemons we need - launch app */
OPAL_OUTPUT_VERBOSE((1, orte_plm_globals.output,
"%s plm:submit: no new daemons to launch",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
goto launch_apps;
}
if (mca_plm_submit_component.debug_daemons &&
mca_plm_submit_component.num_concurrent < num_nodes) {
/**
* If we are in '--debug-daemons' we keep the ssh connection
* alive for the span of the run. If we use this option
* AND we launch on more than "num_concurrent" machines
* then we will deadlock. No connections are terminated
* until the job is complete, no job is started
* since all the orteds are waiting for all the others
* to come online, and the others ore not launched because
* we are waiting on those that have started to terminate
* their ssh tunnels. :(
* As we cannot run in this situation, pretty print the error
* and return an error code.
*/
orte_show_help("help-plm-submit.txt", "deadlock-params",
true, mca_plm_submit_component.num_concurrent, num_nodes);
rc = ORTE_ERR_FATAL;
goto cleanup;
}
/*
* After a discussion between Ralph & Jeff, we concluded that we
* really are handling the prefix dir option incorrectly. It currently
* is associated with an app_context, yet it really refers to the
* location where OpenRTE/Open MPI is installed on a NODE. Fixing
* this right now would involve significant change to orterun as well
* as elsewhere, so we will intentionally leave this incorrect at this
* point. The error, however, is identical to that seen in all prior
* releases of OpenRTE/Open MPI, so our behavior is no worse than before.
*
* A note to fix this, along with ideas on how to do so, has been filed
* on the project's Trac system under "feature enhancement".
*
* For now, default to the prefix_dir provided in the first app_context.
* Since there always MUST be at least one app_context, we are safe in
* doing this.
*/
prefix_dir = apps[0]->prefix_dir;
/* What is our local shell? */
p = getpwuid(getuid());
if( NULL == p ) {
/* This user is unknown to the system. Therefore, there is no reason we
* spawn whatsoever in his name. Give up with a HUGE error message.
*/
orte_show_help( "help-plm-submit.txt", "unknown-user", true, (int)getuid() );
rc = ORTE_ERR_FATAL;
goto cleanup;
} else {
int i;
char *sh_name = NULL;
sh_name = rindex(p->pw_shell, '/');
sh_name++; /* skip the '\' */
for (i = 0; i < (int)(sizeof (orte_plm_submit_shell_name)/
sizeof(orte_plm_submit_shell_name[0])); i++) {
if ( 0 == strcmp(sh_name, orte_plm_submit_shell_name[i]) ) {
switch (i) {
case ORTE_PLM_submit_SHELL_SH: /* fall through */
case ORTE_PLM_submit_SHELL_KSH: /* fall through */
case ORTE_PLM_submit_SHELL_ZSH: /* fall through */
case ORTE_PLM_submit_SHELL_BASH: local_sh = true; break;
case ORTE_PLM_submit_SHELL_TCSH: /* fall through */
case ORTE_PLM_submit_SHELL_CSH: local_csh = true; break;
/* The match has been done, there is no need for a default case here */
}
/* I did match one of the known shells, so now we're done with the shell detection */
break;
}
}
if ( i == ORTE_PLM_submit_SHELL_UNKNOWN ) {
opal_output(0, "WARNING: local probe returned unhandled shell:%s assuming bash\n",
sh_name);
local_sh = true;
}
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: local csh: %d, local sh: %d\n",
local_csh, local_sh);
}
}
/* What is our remote shell? */
if (mca_plm_submit_component.assume_same_shell) {
remote_sh = local_sh;
remote_csh = local_csh;
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: assuming same remote shell as local shell");
}
} else {
orte_plm_submit_shell shell;
rc = orte_plm_submit_probe(nodes[0], &shell);
if (ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
switch (shell) {
case ORTE_PLM_submit_SHELL_SH: /* fall through */
case ORTE_PLM_submit_SHELL_KSH: /* fall through */
case ORTE_PLM_submit_SHELL_BASH: remote_sh = true; break;
case ORTE_PLM_submit_SHELL_TCSH: /* fall through */
case ORTE_PLM_submit_SHELL_CSH: remote_csh = true; break;
default:
opal_output(0, "WARNING: submit probe returned unhandled shell:%s assuming bash\n",
orte_plm_submit_shell_name[shell]);
remote_sh = true;
}
}
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: remote csh: %d, remote sh: %d\n",
remote_csh, remote_sh);
}
/*
* Build argv array
*/
argv = opal_argv_copy(mca_plm_submit_component.agent_argv);
argc = mca_plm_submit_component.agent_argc;
node_name_index1 = argc;
opal_argv_append(&argc, &argv, "<template>");
/* add the daemon command (as specified by user) */
local_exec_index = argc;
opal_argv_append(&argc, &argv, mca_plm_submit_component.orted);
/*
* Add the basic arguments to the orted command line
*/
orte_plm_base_orted_append_basic_args(&argc, &argv,
"env",
&proc_vpid_index, NULL);
local_exec_index_end = argc;
if (mca_plm_submit_component.debug) {
param = opal_argv_join(argv, ' ');
if (NULL != param) {
opal_output(0, "plm:submit: final template argv:");
opal_output(0, "plm:submit: %s", param);
free(param);
}
}
/* Figure out the basenames for the libdir and bindir. This
requires some explanation:
- Use opal_install_dirs.libdir and opal_install_dirs.bindir.
- After a discussion on the devel-core mailing list, the
developers decided that we should use the local directory
basenames as the basis for the prefix on the remote note.
This does not handle a few notable cases (e.g., if the
libdir/bindir is not simply a subdir under the prefix, if the
libdir/bindir basename is not the same on the remote node as
it is here on the local node, etc.), but we decided that
--prefix was meant to handle "the common case". If you need
something more complex than this, a) edit your shell startup
files to set PATH/LD_LIBRARY_PATH properly on the remove
node, or b) use some new/to-be-defined options that
explicitly allow setting the bindir/libdir on the remote
node. We decided to implement these options (e.g.,
--remote-bindir and --remote-libdir) to orterun when it
actually becomes a problem for someone (vs. a hypothetical
situation).
Hence, for now, we simply take the basename of this install's
libdir and bindir and use it to append this install's prefix
and use that on the remote node.
*/
lib_base = opal_basename(opal_install_dirs.libdir);
bin_base = opal_basename(opal_install_dirs.bindir);
/*
* Iterate through each of the nodes
*/
for(nnode=0; nnode < map->num_nodes; nnode++) {
pid_t pid;
char *exec_path;
char **exec_argv;
/* if this daemon already exists, don't launch it! */
if (nodes[nnode]->daemon_launched) {
continue;
}
/* setup node name */
free(argv[node_name_index1]);
if (NULL != nodes[nnode]->username &&
0 != strlen (nodes[nnode]->username)) {
asprintf (&argv[node_name_index1], "%s@%s",
nodes[nnode]->username, nodes[nnode]->name);
} else {
argv[node_name_index1] = strdup(nodes[nnode]->name);
}
/* fork a child to exec the submit/ssh session */
pid = fork();
if (pid < 0) {
ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_CHILDREN);
rc = ORTE_ERR_SYS_LIMITS_CHILDREN;
goto cleanup;
}
/* child */
if (pid == 0) {
char* name_string;
char** env;
char* var;
long fd, fdmax = sysconf(_SC_OPEN_MAX);
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: launching on node %s\n",
rmaps_node->nodename);
}
/* We don't need to sense an oversubscribed condition and set the sched_yield
* for the node as we are only launching the daemons at this time. The daemons
* are now smart enough to set the oversubscribed condition themselves when
* they launch the local procs.
*/
/* Is this a local launch?
*
* Not all node names may be resolvable (if we found
* localhost in the hostfile, for example). So first
* check trivial case of node_name being same as the
* current nodename, which must be local. If that doesn't
* match, check using ifislocal().
*/
if (!mca_plm_submit_component.force_submit &&
(0 == strcmp(nodes[nnode]->name, orte_process_info.nodename) ||
opal_ifislocal(nodes[nnode]->name))) {
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: %s is a LOCAL node\n",
nodes[nnode]->name);
}
exec_path = opal_path_findv(argv[local_exec_index], 0, environ, NULL);
if (NULL == exec_path && NULL == prefix_dir) {
rc = orte_plm_submit_fill_exec_path (&exec_path);
if (ORTE_SUCCESS != rc) {
/* don't normally ERROR_LOG this problem as the function has already
* printed out a nice error message for us - do the ERROR_LOG only
* when we are in debug mode so we can see where it occurred
*/
if (mca_plm_submit_component.debug) {
ORTE_ERROR_LOG(rc);
}
exit(-1); /* the forked process MUST exit */
}
} else {
if (NULL != prefix_dir) {
exec_path = opal_os_path( false, prefix_dir, bin_base, mca_plm_submit_component.orted, NULL );
}
/* If we yet did not fill up the execpath, do so now */
if (NULL == exec_path) {
rc = orte_plm_submit_fill_exec_path (&exec_path);
if (ORTE_SUCCESS != rc) {
/* don't normally ERROR_LOG this problem as the function has already
* printed out a nice error message for us - do the ERROR_LOG only
* when we are in debug mode so we can see where it occurred
*/
if (mca_plm_submit_component.debug) {
ORTE_ERROR_LOG(rc);
}
exit(-1); /* the forked process MUST exit */
}
}
}
/* If we have a prefix, then modify the PATH and
LD_LIBRARY_PATH environment variables. We're
already in the child process, so it's ok to modify
environ. */
if (NULL != prefix_dir) {
char *oldenv, *newenv;
/* Reset PATH */
newenv = opal_os_path( false, prefix_dir, bin_base, NULL );
oldenv = getenv("PATH");
if (NULL != oldenv) {
char *temp;
asprintf(&temp, "%s:%s", newenv, oldenv );
free( newenv );
newenv = temp;
}
opal_setenv("PATH", newenv, true, &environ);
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: reset PATH: %s", newenv);
}
free(newenv);
/* Reset LD_LIBRARY_PATH */
newenv = opal_os_path( false, prefix_dir, lib_base, NULL );
oldenv = getenv("LD_LIBRARY_PATH");
if (NULL != oldenv) {
char* temp;
asprintf(&temp, "%s:%s", newenv, oldenv);
free(newenv);
newenv = temp;
}
opal_setenv("LD_LIBRARY_PATH", newenv, true, &environ);
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: reset LD_LIBRARY_PATH: %s",
newenv);
}
free(newenv);
}
/* Since this is a local execution, we need to
potentially whack the final ")" in the argv (if
sh/csh conditionals, from above). Note that we're
modifying the argv[] in the child process, so
there's no need to save this and restore it
afterward -- the parent's argv[] is unmodified. */
if (NULL != argv[local_exec_index_end]) {
free(argv[local_exec_index_end]);
argv[local_exec_index_end] = NULL;
}
/* tell the daemon to setup its own process session/group */
opal_argv_append(&argc, &argv, "--set-sid");
exec_argv = &argv[local_exec_index];
/* Finally, chdir($HOME) because we're making the
assumption that this is what will happen on
remote nodes (via submit/ssh). This allows a user
to specify a path that is relative to $HOME for
both the cwd and argv[0] and it will work on
all nodes -- including the local nost.
Otherwise, it would work on remote nodes and
not the local node. If the user does not start
in $HOME on the remote nodes... well... let's
hope they start in $HOME. :-) */
var = opal_home_directory();
if (NULL != var) {
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: changing to directory %s", var);
}
/* Ignore errors -- what are we going to do?
(and we ignore errors on the remote nodes
in the fork plm, so this is consistent) */
chdir(var);
}
} else {
if (mca_plm_submit_component.debug) {
opal_output(0, "plm:submit: %s is a REMOTE node\n",
nodes[nnode]->name);
}
exec_argv = argv;
exec_path = strdup(mca_plm_submit_component.agent_path);
if (NULL != prefix_dir) {
char *opal_prefix = getenv("OPAL_PREFIX");
if (remote_sh) {
asprintf (&argv[local_exec_index],
"%s%s%s PATH=%s/%s:$PATH ; export PATH ; "
"LD_LIBRARY_PATH=%s/%s:$LD_LIBRARY_PATH ; export LD_LIBRARY_PATH ; "
"%s/%s/%s",
(opal_prefix != NULL ? "OPAL_PREFIX=" : ""),
(opal_prefix != NULL ? opal_prefix : ""),
(opal_prefix != NULL ? " ;" : ""),
prefix_dir, bin_base,
prefix_dir, lib_base,
prefix_dir, bin_base,
mca_plm_submit_component.orted);
} else if (remote_csh) {
/* [t]csh is a bit more challenging -- we
have to check whether LD_LIBRARY_PATH
is already set before we try to set it.
Must be very careful about obeying
[t]csh's order of evaluation and not
using a variable before it is defined.
See this thread for more details:
http://www.open-mpi.org/community/lists/users/2006/01/0517.php. */
asprintf (&argv[local_exec_index],
"%s%s%s set path = ( %s/%s $path ) ; "
"if ( $?LD_LIBRARY_PATH == 1 ) "
"set OMPI_have_llp ; "
"if ( $?LD_LIBRARY_PATH == 0 ) "
"setenv LD_LIBRARY_PATH %s/%s ; "
"if ( $?OMPI_have_llp == 1 ) "
"setenv LD_LIBRARY_PATH %s/%s:$LD_LIBRARY_PATH ; "
"%s/%s/%s",
(opal_prefix != NULL ? "setenv OPAL_PREFIX " : ""),
(opal_prefix != NULL ? opal_prefix : ""),
(opal_prefix != NULL ? " ;" : ""),
prefix_dir, bin_base,
prefix_dir, lib_base,
prefix_dir, lib_base,
prefix_dir, bin_base,
mca_plm_submit_component.orted);
}
}
}
/* pass the vpid */
rc = orte_util_convert_vpid_to_string(&vpid_string, nodes[nnode]->daemon->name.vpid);
if (ORTE_SUCCESS != rc) {
opal_output(0, "orte_plm_submit: unable to get daemon vpid as string");
exit(-1);
}
free(argv[proc_vpid_index]);
argv[proc_vpid_index] = strdup(vpid_string);
free(vpid_string);
/* Tie /dev/null to stdin */
fd = open("/dev/null", O_RDWR);
dup2(fd, 0);
close(fd);
/* close all file descriptors w/ exception of stdin/stdout/stderr */
for(fd=3; fd<fdmax; fd++)
close(fd);
/* Set signal handlers back to the default. Do this close
to the execve() because the event library may (and likely
will) reset them. If we don't do this, the event
library may have left some set that, at least on some
OS's, don't get reset via fork() or exec(). Hence, the
orted could be unkillable (for example). */
set_handler_default(SIGTERM);
set_handler_default(SIGINT);
set_handler_default(SIGHUP);
set_handler_default(SIGPIPE);
set_handler_default(SIGCHLD);
/* Unblock all signals, for many of the same reasons that
we set the default handlers, above. This is noticable
on Linux where the event library blocks SIGTERM, but we
don't want that blocked by the orted (or, more
specifically, we don't want it to be blocked by the
orted and then inherited by the ORTE processes that it
forks, making them unkillable by SIGTERM). */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* setup environment */
env = opal_argv_copy(environ);
var = mca_base_param_environ_variable("seed",NULL,NULL);
opal_setenv(var, "0", true, &env);
/* exec the daemon */
if (mca_plm_submit_component.debug) {
param = opal_argv_join(exec_argv, ' ');
if (NULL != param) {
opal_output(0, "plm:submit: executing: (%s) [%s]", exec_path, param);
free(param);
}
}
execve(exec_path, exec_argv, env);
opal_output(0, "plm:submit: execv of %s failed with errno=%s(%d)\n",
exec_path, strerror(errno), errno);
exit(-1);
} else { /* father */
/* indicate this daemon has been launched */
nodes[nnode]->daemon->state = ORTE_PROC_STATE_LAUNCHED;
OPAL_THREAD_LOCK(&mca_plm_submit_component.lock);
/* This situation can lead to a deadlock if '--debug-daemons' is set.
* However, the deadlock condition is tested at the begining of this
* function, so we're quite confident it should not happens here.
*/
if (mca_plm_submit_component.num_children++ >=
mca_plm_submit_component.num_concurrent) {
opal_condition_wait(&mca_plm_submit_component.cond, &mca_plm_submit_component.lock);
}
OPAL_THREAD_UNLOCK(&mca_plm_submit_component.lock);
/* setup callback on sigchild - wait until setup above is complete
* as the callback can occur in the call to orte_wait_cb
*/
orte_wait_cb(pid, orte_plm_submit_wait_daemon, NULL);
/* if required - add delay to avoid problems w/ X11 authentication */
if (mca_plm_submit_component.debug && mca_plm_submit_component.delay) {
sleep(mca_plm_submit_component.delay);
}
}
}
/* wait for daemons to callback */
if (ORTE_SUCCESS != (rc = orte_plm_base_daemon_callback(map->num_new_daemons))) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_globals.output,
"%s plm:submit: daemon launch failed for job %s on error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(active_job), ORTE_ERROR_NAME(rc)));
goto cleanup;
}
launch_apps:
if (ORTE_SUCCESS != (rc = orte_plm_base_launch_apps(active_job))) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_globals.output,
"%s plm:submit: launch of apps failed for job %s on error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(active_job), ORTE_ERROR_NAME(rc)));
goto cleanup;
}
/* get here if launch went okay */
failed_launch = false;
cleanup:
if (NULL != lib_base) {
free(lib_base);
}
if (NULL != bin_base) {
free(bin_base);
}
if (NULL != argv) {
opal_argv_free(argv);
}
/* check for failed launch - if so, force terminate */
if (failed_launch) {
orte_errmgr.update_state(jdata->jobid, ORTE_JOB_STATE_FAILED_TO_START,
NULL, ORTE_PROC_STATE_UNDEF,
0, ORTE_ERROR_DEFAULT_EXIT_CODE);
}
return rc;
}
/**
* Terminate the orteds for a given job
*/
int orte_plm_submit_terminate_orteds(void)
{
int rc;
/* now tell them to die */
if (orte_abnormal_term_ordered) {
/* cannot know if a daemon is able to
* tell us it died, so just ensure they
* all terminate
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_HALT_VM_CMD))) {
ORTE_ERROR_LOG(rc);
}
} else {
/* we need them to "phone home", though,
* so we can know that they have exited
*/
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
}
return rc;
}
int orte_plm_submit_signal_job(orte_jobid_t jobid, int32_t signal)
{
int rc;
/* order them to pass this signal to their local procs */
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_signal_local_procs(jobid, signal))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
int orte_plm_submit_finalize(void)
{
int rc;
/* cleanup any pending recvs */
if (ORTE_SUCCESS != (rc = orte_plm_base_comm_stop())) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
static void set_handler_default(int sig)
{
struct sigaction act;
act.sa_handler = SIG_DFL;
act.sa_flags = 0;
sigemptyset(&act.sa_mask);
sigaction(sig, &act, (struct sigaction *)0);
}