Split ompi_rte_init into two stages to allow detection of universe existence prior to starting remainder of rte. Adjust mpirun2 and mpi_init to accommodate.
This commit was SVN r2499.
Этот коммит содержится в:
Ralph Castain
родитель
7feccee44c
Коммит
6b05166b75
@ -87,8 +87,10 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
|
||||
/* Join the run-time environment */
|
||||
allow_multi_user_threads = true;
|
||||
have_hidden_threads = false;
|
||||
if (OMPI_SUCCESS != (ret = ompi_rte_init(&allow_multi_user_threads,
|
||||
&have_hidden_threads))) {
|
||||
if ((OMPI_SUCCESS != (ret = ompi_rte_init_stage1(&allow_multi_user_threads,
|
||||
&have_hidden_threads))) ||
|
||||
(OMPI_SUCCESS != (ret = ompi_rte_init_stage2(&allow_multi_user_threads,
|
||||
&have_hidden_threads)))) {
|
||||
/* JMS show_help */
|
||||
printf("show_help: ompi_mpi_init failed in mca_rte_init\n");
|
||||
return ret;
|
||||
|
||||
@ -25,7 +25,8 @@ libruntime_la_SOURCES = \
|
||||
ompi_universe.c \
|
||||
ompi_progress.c \
|
||||
ompi_rte_finalize.c \
|
||||
ompi_rte_init.c \
|
||||
ompi_rte_init_stage1.c \
|
||||
ompi_rte_init_stage2.c \
|
||||
ompi_rte_llm.c \
|
||||
ompi_rte_monitor.c \
|
||||
ompi_rte_cmd_line_setup.c \
|
||||
|
||||
150
src/runtime/ompi_rte_init_stage1.c
Обычный файл
150
src/runtime/ompi_rte_init_stage1.c
Обычный файл
@ -0,0 +1,150 @@
|
||||
/*
|
||||
* $HEADER$
|
||||
*/
|
||||
|
||||
/** @file **/
|
||||
|
||||
/* #define _GNU_SOURCE */
|
||||
|
||||
#include "ompi_config.h"
|
||||
|
||||
#include "include/constants.h"
|
||||
#include "event/event.h"
|
||||
#include "util/output.h"
|
||||
#include "threads/mutex.h"
|
||||
#include "mca/mca.h"
|
||||
#include "mca/base/base.h"
|
||||
#include "mca/base/mca_base_param.h"
|
||||
#include "mca/pcm/base/base.h"
|
||||
#include "mca/pcmclient/base/base.h"
|
||||
#include "mca/llm/base/base.h"
|
||||
#include "mca/oob/oob.h"
|
||||
#include "mca/ns/base/base.h"
|
||||
#include "mca/gpr/base/base.h"
|
||||
#include "util/proc_info.h"
|
||||
#include "util/session_dir.h"
|
||||
#include "util/sys_info.h"
|
||||
|
||||
#include "runtime/runtime.h"
|
||||
|
||||
/**
|
||||
* Initialze and setup a process in the OMPI RTE.
|
||||
*
|
||||
* @retval OMPI_SUCCESS Upon success.
|
||||
* @retval OMPI_ERROR Upon failure.
|
||||
*
|
||||
* This function performs
|
||||
*
|
||||
* Just a note for developer:
|
||||
|
||||
* So there are 3 ways in which an application can be started
|
||||
* 1) rte_boot, followed by mpirun
|
||||
* 2) mpirun (alone)
|
||||
* 3) singleton (./a.out)
|
||||
*
|
||||
* Case 1) If the rte has already been booted, then mpirun will accept
|
||||
* an optional command line parameter --universe=[rte universe name]
|
||||
* which says which universe this application wants to be a part
|
||||
* of. mpirun will then package this universe name and send it to the
|
||||
* processes it will be starting off (fork/exec) on local or remote
|
||||
* node.The packaging mechanism can be either command line parameter
|
||||
* to the a.out it forks or make it part of environment
|
||||
* (implementation dependent).
|
||||
*
|
||||
* Case 2) When mpirun is done alone and no universe is present, then
|
||||
* the mpirun starts off the universe (using rte_boot), then
|
||||
* fork/execs the processes, passin g along the [universe_name].
|
||||
*
|
||||
* Case 3) For a singleton, if there is alrady an existing rte
|
||||
* universe which it wants to join, it can specify that using the
|
||||
* --universe command line. So it will do
|
||||
*
|
||||
* $ ./a.out --universe=[universe_name]
|
||||
*
|
||||
* In this case, MPI_Init will have to be called as MPI_Init(&argc, &argv)
|
||||
|
||||
* If it does not want to join any existing rte, then it just starts
|
||||
* off as ./a.out with no command line option. In that case, MPI_Init
|
||||
* does not necesaarily needs to passed argc and argv. Infact if argc
|
||||
* and argv are not passed or just have one entry (the command name),
|
||||
* then MPI_Init would assume that new rte universe needs to be
|
||||
* started up.
|
||||
*
|
||||
*
|
||||
* MPI_Init() will look at its argc, argv. If it find the universe
|
||||
* name there, fine. Else it looks at the environment variables for
|
||||
* universe_name. If it finds there, fine again. Under such
|
||||
* conditions, it joins the existing rte universe. If no universe
|
||||
* name is found, it calls rte_boot to start off a new rte universe.
|
||||
*
|
||||
* For singleton, MPI_Init() do:
|
||||
*
|
||||
* if (I am a singleton) and (there is no universe)
|
||||
* do rte_boot
|
||||
*
|
||||
* But if I am not a singleton, then I have been started by mpirun and
|
||||
* already provided a universe_name to join. So I wont ever start a
|
||||
* universe under such conditons. mpirun will pass me the
|
||||
* universe_name (either mpirun would have started the universe, or it
|
||||
* would have been already started by rte_boot)
|
||||
*/
|
||||
|
||||
/* globals used by RTE */
|
||||
int ompi_rte_debug_flag=0;
|
||||
ompi_universe_t ompi_universe_info = {
|
||||
/* .name = */ "default-universe",
|
||||
/* .host = */ "localhost",
|
||||
/* .uid = */ NULL,
|
||||
/* .pid = */ 0,
|
||||
/* .persistence = */ false,
|
||||
/* .scope = */ "local",
|
||||
/* .silent_mode = */ true,
|
||||
/* .web_server = */ false,
|
||||
/* .socket_contact_info = */ NULL,
|
||||
/* .oob_contact_info = */ NULL,
|
||||
/* .console_connected = */ false,
|
||||
/* .scriptfile = */ NULL,
|
||||
/* .hostfile = */ NULL
|
||||
};
|
||||
|
||||
|
||||
int ompi_rte_init_stage1(bool *allow_multi_user_threads, bool *have_hidden_threads)
|
||||
{
|
||||
int ret;
|
||||
bool user_threads, hidden_threads;
|
||||
|
||||
*allow_multi_user_threads = true;
|
||||
*have_hidden_threads = false;
|
||||
|
||||
ret = mca_base_param_register_int("ompi", "rte", "debug", NULL, 0);
|
||||
mca_base_param_lookup_int(ret, &ompi_rte_debug_flag);
|
||||
|
||||
/*
|
||||
* Initialize the event library
|
||||
*/
|
||||
if (OMPI_SUCCESS != (ret = ompi_event_init())) {
|
||||
ompi_output(0, "ompi_rte_init: ompi_event_init failed with error status: %d\n", ret);
|
||||
return ret;
|
||||
}
|
||||
|
||||
/*
|
||||
* Out of Band Messaging
|
||||
*/
|
||||
if (OMPI_SUCCESS != (ret = mca_oob_base_open())) {
|
||||
/* JMS show_help */
|
||||
printf("show_help: ompi_rte_init failed in oob_base_open\n");
|
||||
return ret;
|
||||
}
|
||||
user_threads = true;
|
||||
hidden_threads = false;
|
||||
if (OMPI_SUCCESS != (ret = mca_oob_base_init(&user_threads,
|
||||
&hidden_threads))) {
|
||||
/* JMS show_help */
|
||||
printf("show_help: ompi_rte_init failed in mca_oob_base_init()\n");
|
||||
return ret;
|
||||
}
|
||||
*allow_multi_user_threads &= user_threads;
|
||||
*have_hidden_threads |= hidden_threads;
|
||||
|
||||
return OMPI_SUCCESS;
|
||||
}
|
||||
@ -9,7 +9,6 @@
|
||||
#include "ompi_config.h"
|
||||
|
||||
#include "include/constants.h"
|
||||
#include "runtime/runtime.h"
|
||||
#include "event/event.h"
|
||||
#include "util/output.h"
|
||||
#include "threads/mutex.h"
|
||||
@ -26,127 +25,14 @@
|
||||
#include "util/session_dir.h"
|
||||
#include "util/sys_info.h"
|
||||
|
||||
/**
|
||||
* Initialze and setup a process in the OMPI RTE.
|
||||
*
|
||||
* @retval OMPI_SUCCESS Upon success.
|
||||
* @retval OMPI_ERROR Upon failure.
|
||||
*
|
||||
* This function performs
|
||||
*
|
||||
* Just a note for developer:
|
||||
#include "runtime/runtime.h"
|
||||
|
||||
* So there are 3 ways in which an application can be started
|
||||
* 1) rte_boot, followed by mpirun
|
||||
* 2) mpirun (alone)
|
||||
* 3) singleton (./a.out)
|
||||
*
|
||||
* Case 1) If the rte has already been booted, then mpirun will accept
|
||||
* an optional command line parameter --universe=[rte universe name]
|
||||
* which says which universe this application wants to be a part
|
||||
* of. mpirun will then package this universe name and send it to the
|
||||
* processes it will be starting off (fork/exec) on local or remote
|
||||
* node.The packaging mechanism can be either command line parameter
|
||||
* to the a.out it forks or make it part of environment
|
||||
* (implementation dependent).
|
||||
*
|
||||
* Case 2) When mpirun is done alone and no universe is present, then
|
||||
* the mpirun starts off the universe (using rte_boot), then
|
||||
* fork/execs the processes, passin g along the [universe_name].
|
||||
*
|
||||
* Case 3) For a singleton, if there is alrady an existing rte
|
||||
* universe which it wants to join, it can specify that using the
|
||||
* --universe command line. So it will do
|
||||
*
|
||||
* $ ./a.out --universe=[universe_name]
|
||||
*
|
||||
* In this case, MPI_Init will have to be called as MPI_Init(&argc, &argv)
|
||||
|
||||
* If it does not want to join any existing rte, then it just starts
|
||||
* off as ./a.out with no command line option. In that case, MPI_Init
|
||||
* does not necesaarily needs to passed argc and argv. Infact if argc
|
||||
* and argv are not passed or just have one entry (the command name),
|
||||
* then MPI_Init would assume that new rte universe needs to be
|
||||
* started up.
|
||||
*
|
||||
*
|
||||
* MPI_Init() will look at its argc, argv. If it find the universe
|
||||
* name there, fine. Else it looks at the environment variables for
|
||||
* universe_name. If it finds there, fine again. Under such
|
||||
* conditions, it joins the existing rte universe. If no universe
|
||||
* name is found, it calls rte_boot to start off a new rte universe.
|
||||
*
|
||||
* For singleton, MPI_Init() do:
|
||||
*
|
||||
* if (I am a singleton) and (there is no universe)
|
||||
* do rte_boot
|
||||
*
|
||||
* But if I am not a singleton, then I have been started by mpirun and
|
||||
* already provided a universe_name to join. So I wont ever start a
|
||||
* universe under such conditons. mpirun will pass me the
|
||||
* universe_name (either mpirun would have started the universe, or it
|
||||
* would have been already started by rte_boot)
|
||||
*/
|
||||
|
||||
/* globals used by RTE */
|
||||
int ompi_rte_debug_flag=0;
|
||||
ompi_universe_t ompi_universe_info = {
|
||||
/* .name = */ "default-universe",
|
||||
/* .host = */ "localhost",
|
||||
/* .uid = */ NULL,
|
||||
/* .pid = */ 0,
|
||||
/* .persistence = */ false,
|
||||
/* .scope = */ "local",
|
||||
/* .silent_mode = */ true,
|
||||
/* .script_mode = */ false,
|
||||
/* .web_server = */ false,
|
||||
/* .socket_contact_info = */ NULL,
|
||||
/* .oob_contact_info = */ NULL,
|
||||
/* .console_connected = */ false,
|
||||
/* .scriptfile = */ NULL,
|
||||
/* .hostfile = */ NULL
|
||||
};
|
||||
|
||||
|
||||
int ompi_rte_init(bool *allow_multi_user_threads, bool *have_hidden_threads)
|
||||
int ompi_rte_init_stage2(bool *allow_multi_user_threads, bool *have_hidden_threads)
|
||||
{
|
||||
int ret;
|
||||
bool user_threads, hidden_threads;
|
||||
char *jobid_str=NULL, *procid_str=NULL;
|
||||
|
||||
*allow_multi_user_threads = true;
|
||||
*have_hidden_threads = false;
|
||||
|
||||
ret = mca_base_param_register_int("ompi", "rte", "debug", NULL, 0);
|
||||
mca_base_param_lookup_int(ret, &ompi_rte_debug_flag);
|
||||
|
||||
/*
|
||||
* Initialize the event library
|
||||
*/
|
||||
if (OMPI_SUCCESS != (ret = ompi_event_init())) {
|
||||
ompi_output(0, "ompi_rte_init: ompi_event_init failed with error status: %d\n", ret);
|
||||
return ret;
|
||||
}
|
||||
|
||||
/*
|
||||
* Out of Band Messaging
|
||||
*/
|
||||
if (OMPI_SUCCESS != (ret = mca_oob_base_open())) {
|
||||
/* JMS show_help */
|
||||
printf("show_help: ompi_rte_init failed in oob_base_open\n");
|
||||
return ret;
|
||||
}
|
||||
user_threads = true;
|
||||
hidden_threads = false;
|
||||
if (OMPI_SUCCESS != (ret = mca_oob_base_init(&user_threads,
|
||||
&hidden_threads))) {
|
||||
/* JMS show_help */
|
||||
printf("show_help: ompi_rte_init failed in mca_oob_base_init()\n");
|
||||
return ret;
|
||||
}
|
||||
*allow_multi_user_threads &= user_threads;
|
||||
*have_hidden_threads |= hidden_threads;
|
||||
|
||||
/*
|
||||
* Name Server
|
||||
*/
|
||||
@ -65,9 +65,13 @@ int ompi_rte_universe_exists(char *host, char *name, char *tmpdir, char *oob_con
|
||||
return OMPI_ERR_FATAL;
|
||||
}
|
||||
|
||||
/* /\* ...and ping to verify it's alive *\/ */
|
||||
/* if (OMPI_SUCCESS != mca_oob_ping(&seed)) { */
|
||||
/* return OMPI_ERR_CONNECTION_FAILED; */
|
||||
/* } */
|
||||
/* /\* ...and ping to verify it's alive *\/ */
|
||||
/* if (OMPI_SUCCESS != mca_oob_ping(&seed)) { */
|
||||
/* return OMPI_ERR_CONNECTION_FAILED; */
|
||||
/* } */
|
||||
|
||||
/* set the my_universe field */
|
||||
ompi_process_info.my_universe = strdup(ompi_universe_info.name);
|
||||
return OMPI_SUCCESS;
|
||||
}
|
||||
}
|
||||
|
||||
@ -41,7 +41,6 @@ extern "C" {
|
||||
bool persistence;
|
||||
char *scope;
|
||||
bool silent_mode;
|
||||
bool script_mode;
|
||||
bool web_server;
|
||||
char *socket_contact_info;
|
||||
char *oob_contact_info;
|
||||
@ -95,7 +94,8 @@ extern "C" {
|
||||
* be called by every application using the RTE interface, including
|
||||
* MPI applications and mpirun.
|
||||
*/
|
||||
int ompi_rte_init(bool *allow_multi_user_threads, bool *have_hidden_threads);
|
||||
int ompi_rte_init_stage1(bool *allow_multi_user_threads, bool *have_hidden_threads);
|
||||
int ompi_rte_init_stage2(bool *allow_multi_user_threads, bool *have_hidden_threads);
|
||||
|
||||
/**
|
||||
* Finalize the Open MPI run time environment
|
||||
|
||||
@ -107,7 +107,8 @@ main(int argc, char *argv[])
|
||||
return ret;
|
||||
}
|
||||
|
||||
if (OMPI_SUCCESS != ompi_rte_init(&multi_thread, &hidden_thread)) {
|
||||
if (OMPI_SUCCESS != ompi_rte_init_stage1(&multi_thread, &hidden_thread) ||
|
||||
OMPI_SUCCESS != ompi_rte_init_stage2(&multi_thread, &hidden_thread)) {
|
||||
/* BWB show_help */
|
||||
printf("show_help: ompi_rte_init failed\n");
|
||||
return ret;
|
||||
|
||||
Загрузка…
x
Ссылка в новой задаче
Block a user