1
1
openmpi/orte/mca/plm/base/plm_base_receive.c

381 строка
14 KiB
C
Исходник Обычный вид История

/* -*- C -*-
*
* Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2011 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) 2011 Los Alamos National Security, LLC.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*
*/
/*
* includes
*/
#include "orte_config.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#include "opal/mca/mca.h"
#include "opal/mca/base/mca_base_param.h"
#include "opal/dss/dss.h"
#include "opal/threads/threads.h"
#include "orte/constants.h"
#include "orte/types.h"
#include "orte/util/proc_info.h"
#include "orte/util/error_strings.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/ras/base/base.h"
#include "orte/util/name_fns.h"
#include "orte/mca/state/state.h"
#include "orte/runtime/orte_globals.h"
Start reducing our dependency on the event library by removing at least one instance where we use it to redirect the program counter. Rolf reported occasional hangs of mpirun in very specific circumstances after all daemons were done. A review of MTT results indicates this may have been happening more generally in a small fraction of cases. The problem was tracked to use of the grpcomm.onesided_barrier to control daemon/mpirun termination. This relied on messaging -and- required that the program counter jump from the errmgr back to grpcomm. On rare occasions, this jump did not occur, causing mpirun to hang. This patch looks more invasive than it is - most of the affected files simply had one or two lines removed. The essence of the change is: * pulled the job_complete and quit routines out of orterun and orted_main and put them in a common place * modified the errmgr to directly call the new routines when termination is detected * removed the grpcomm.onesided_barrier and its associated RML tag * add a new "num_routes" API to the routed framework that reports back the number of dependent routes. When route_lost is called, the daemon's list of "children" is checked and adjusted if that route went to a "leaf" in the routing tree * use connection termination between daemons to track rollup of the daemon tree. Daemons and HNP now terminate once num_routes returns zero Also picked up in this commit is the addition of a new bool flag to the app_context struct, and increasing the job_control field from 8 to 16 bits. Both trivial. This commit was SVN r23429.
2010-07-17 21:03:27 +00:00
#include "orte/runtime/orte_quit.h"
#include "orte/mca/plm/plm_types.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/plm/base/base.h"
static bool recv_issued=false;
int orte_plm_base_comm_start(void)
{
int rc;
if (recv_issued) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive start comm",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_PLM,
ORTE_RML_PERSISTENT,
orte_plm_base_recv,
NULL))) {
ORTE_ERROR_LOG(rc);
}
if (ORTE_PROC_IS_HNP) {
if (ORTE_SUCCESS != (rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_ORTED_CALLBACK,
ORTE_RML_PERSISTENT,
orte_plm_base_daemon_callback, NULL))) {
ORTE_ERROR_LOG(rc);
}
}
recv_issued = true;
return rc;
}
int orte_plm_base_comm_stop(void)
{
if (!recv_issued) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive stop comm",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PLM);
if (ORTE_PROC_IS_HNP) {
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK);
}
recv_issued = false;
return ORTE_SUCCESS;
}
/* process incoming messages in order of receipt */
void orte_plm_base_recv(int status, orte_process_name_t* sender,
opal_buffer_t* buffer, orte_rml_tag_t tag,
void* cbdata)
{
orte_plm_cmd_flag_t command;
orte_std_cntr_t count;
orte_jobid_t job;
orte_job_t *jdata, *parent;
opal_buffer_t *answer;
orte_vpid_t vpid;
orte_proc_t *proc;
orte_proc_state_t state;
orte_exit_code_t exit_code;
int32_t rc=ORTE_SUCCESS, ret;
orte_app_context_t *app, *child_app;
orte_process_name_t name;
pid_t pid;
bool running;
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive processing msg",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
count = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &command, &count, ORTE_PLM_CMD))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
switch (command) {
case ORTE_PLM_LAUNCH_JOB_CMD:
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive job launch command from %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(sender)));
/* unpack the job object */
count = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &jdata, &count, ORTE_JOB))) {
ORTE_ERROR_LOG(rc);
goto ANSWER_LAUNCH;
}
/* record the sender so we know who to respond to */
jdata->originator.jobid = sender->jobid;
jdata->originator.vpid = sender->vpid;
/* get the parent's job object */
if (NULL == (parent = orte_get_job_data_object(sender->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto ANSWER_LAUNCH;
}
/* if the prefix was set in the parent's job, we need to transfer
* that prefix to the child's app_context so any further launch of
* orteds can find the correct binary. There always has to be at
* least one app_context in both parent and child, so we don't
* need to check that here. However, be sure not to overwrite
* the prefix if the user already provided it!
*/
app = (orte_app_context_t*)opal_pointer_array_get_item(parent->apps, 0);
child_app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, 0);
if (NULL != app->prefix_dir &&
NULL == child_app->prefix_dir) {
child_app->prefix_dir = strdup(app->prefix_dir);
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive adding hosts",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* process any add-hostfile and add-host options that were provided */
if (ORTE_SUCCESS != (rc = orte_ras_base_add_hosts(jdata))) {
ORTE_ERROR_LOG(rc);
goto ANSWER_LAUNCH;
}
if( NULL == parent->bookmark ) {
/* find the sender's node in the job map */
if (NULL != (proc = (orte_proc_t*)opal_pointer_array_get_item(parent->procs, sender->vpid))) {
/* set the bookmark so the child starts from that place - this means
* that the first child process could be co-located with the proc
* that called comm_spawn, assuming slots remain on that node. Otherwise,
* the procs will start on the next available node
*/
jdata->bookmark = proc->node;
}
} else {
jdata->bookmark = parent->bookmark;
}
/* launch it */
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive calling spawn",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if (ORTE_SUCCESS != (rc = orte_plm.spawn(jdata))) {
ORTE_ERROR_LOG(rc);
goto ANSWER_LAUNCH;
}
break;
ANSWER_LAUNCH:
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive - error on launch: %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), rc));
/* setup the response */
answer = OBJ_NEW(opal_buffer_t);
/* pack the error code to be returned */
if (ORTE_SUCCESS != (ret = opal_dss.pack(answer, &rc, 1, OPAL_INT32))) {
ORTE_ERROR_LOG(ret);
}
/* send the response back to the sender */
if (0 > (ret = orte_rml.send_buffer_nb(sender, answer, ORTE_RML_TAG_PLM_PROXY, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(answer);
}
break;
case ORTE_PLM_UPDATE_PROC_STATE:
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive update proc state command from %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(sender)));
count = 1;
while (ORTE_SUCCESS == (rc = opal_dss.unpack(buffer, &job, &count, ORTE_JOBID))) {
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive got update_proc_state for job %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(job)));
name.jobid = job;
running = false;
/* get the job object */
if (NULL == (jdata = orte_get_job_data_object(job))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto CLEANUP;
}
count = 1;
while (ORTE_SUCCESS == (rc = opal_dss.unpack(buffer, &vpid, &count, ORTE_VPID))) {
if (ORTE_VPID_INVALID == vpid) {
/* flag indicates that this job is complete - move on */
break;
}
name.vpid = vpid;
/* unpack the pid */
count = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &pid, &count, OPAL_PID))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
/* unpack the state */
count = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &state, &count, ORTE_PROC_STATE))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
if (ORTE_PROC_STATE_RUNNING == state) {
running = true;
}
/* unpack the exit code */
count = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &exit_code, &count, ORTE_EXIT_CODE))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive got update_proc_state for vpid %lu state %s exit_code %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(unsigned long)vpid, orte_proc_state_to_str(state), (int)exit_code));
/* get the proc data object */
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, vpid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
ORTE_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
proc->state = state;
proc->pid = pid;
proc->exit_code = exit_code;
ORTE_ACTIVATE_PROC_STATE(&name, state);
}
if (running) {
jdata->num_daemons_reported++;
if (orte_report_launch_progress) {
if (0 == jdata->num_daemons_reported % 100 ||
jdata->num_daemons_reported == orte_process_info.num_procs) {
opal_output(orte_clean_output, "App launch reported: %d (out of %d) daemons - %d (out of %d) procs",
(int)jdata->num_daemons_reported, (int)orte_process_info.num_procs,
(int)jdata->num_launched, (int)jdata->num_procs);
}
}
}
/* prepare for next job */
count = 1;
}
if (ORTE_ERR_UNPACK_READ_PAST_END_OF_BUFFER != rc) {
ORTE_ERROR_LOG(rc);
} else {
rc = ORTE_SUCCESS;
}
break;
case ORTE_PLM_INIT_ROUTES_CMD:
count=1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &job, &count, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
goto CLEANUP;
}
name.jobid = job;
/* get the job object */
if (NULL == (jdata = orte_get_job_data_object(job))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto CLEANUP;
}
count=1;
while (ORTE_SUCCESS == opal_dss.unpack(buffer, &vpid, &count, ORTE_VPID)) {
if (ORTE_VPID_INVALID == vpid) {
break;
}
name.vpid = vpid;
ORTE_ACTIVATE_PROC_STATE(&name, ORTE_PROC_STATE_REGISTERED);
count=1;
}
/* pass the remainder of the buffer to the active module's
* init_routes API
*/
if (ORTE_SUCCESS != (rc = orte_routed.init_routes(job, buffer))) {
ORTE_ERROR_LOG(rc);
ORTE_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
break;
default:
ORTE_ERROR_LOG(ORTE_ERR_VALUE_OUT_OF_BOUNDS);
rc = ORTE_ERR_VALUE_OUT_OF_BOUNDS;
break;
}
CLEANUP:
if (ORTE_SUCCESS != rc) {
goto DEPART;
}
DEPART:
/* see if an error occurred - if so, wakeup the HNP so we can exit */
if (ORTE_PROC_IS_HNP && ORTE_SUCCESS != rc) {
jdata = NULL;
ORTE_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:receive done processing commands",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
}
/* where HNP messages come */
void orte_plm_base_receive_process_msg(int fd, short event, void *data)
{
assert(0);
}