54b2cf747e
The commit has been tested for C/R and Cray operations, and on Odin (SLURM, rsh) and RoadRunner (TM). I tried to update all environments, but obviously could not test them. I know that Windows needs some work, and have highlighted what is know to be needed in the odls process component. This represents a lot of work by Brian, Tim P, Josh, and myself, with much advice from Jeff and others. For posterity, I have appended a copy of the email describing the work that was done: As we have repeatedly noted, the modex operation in MPI_Init is the single greatest consumer of time during startup. To-date, we have executed that operation as an ORTE stage gate that held the process until a startup message containing all required modex (and OOB contact info - see #3 below) info could be sent to it. Each process would send its data to the HNP's registry, which assembled and sent the message when all processes had reported in. In addition, ORTE had taken responsibility for monitoring process status as it progressed through a series of "stage gates". The process reported its status at each gate, and ORTE would then send a "release" message once all procs had reported in. The incoming changes revamp these procedures in three ways: 1. eliminating the ORTE stage gate system and cleanly delineating responsibility between the OMPI and ORTE layers for MPI init/finalize. The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info. The allgather is implemented using the orte_grpcomm framework since the BTL's are not active at that point. At the moment, the grpcomm framework only has a "basic" component analogous to OMPI's "basic" coll framework - I would recommend that the MPI team create additional, more advanced components to improve performance of this step. The other stage gates have been replaced by orte_grpcomm barrier functions. We tried to use MPI barriers instead (since the BTL's are active at that point), but - as we discussed on the telecon - these are not currently true barriers so the job would hang when we fell through while messages were still in process. Note that the grpcomm barrier doesn't actually resolve that problem, but Brian has pointed out that we are unlikely to ever see it violated. Again, you might want to spend a little time on an advanced barrier algorithm as the one in "basic" is very simplistic. Summarizing this change: ORTE no longer tracks process state nor has direct responsibility for synchronizing jobs. This is now done via collective operations within the MPI layer, albeit using ORTE collective communication services. I -strongly- urge the MPI team to implement advanced collective algorithms to improve the performance of this critical procedure. 2. reducing the volume of data exchanged during modex. Data in the modex consisted of the process name, the name of the node where that process is located (expressed as a string), plus a string representation of all contact info. The nodename was required in order for the modex to determine if the process was local or not - in addition, some people like to have it to print pretty error messages when a connection failed. The size of this data has been reduced in three ways: (a) reducing the size of the process name itself. The process name consisted of two 32-bit fields for the jobid and vpid. This is far larger than any current system, or system likely to exist in the near future, can support. Accordingly, the default size of these fields has been reduced to 16-bits, which means you can have 32k procs in each of 32k jobs. Since the daemons must have a vpid, and we require one daemon/node, this also restricts the default configuration to 32k nodes. To support any future "mega-clusters", a configuration option --enable-jumbo-apps has been added. This option increases the jobid and vpid field sizes to 32-bits. Someday, if necessary, someone can add yet another option to increase them to 64-bits, I suppose. (b) replacing the string nodename with an integer nodeid. Since we have one daemon/node, the nodeid corresponds to the local daemon's vpid. This replaces an often lengthy string with only 2 (or at most 4) bytes, a substantial reduction. (c) when the mca param requesting that nodenames be sent to support pretty error messages, a second mca param is now used to request FQDN - otherwise, the domain name is stripped (by default) from the message to save space. If someone wants to combine those into a single param somehow (perhaps with an argument?), they are welcome to do so - I didn't want to alter what people are already using. While these may seem like small savings, they actually amount to a significant impact when aggregated across the entire modex operation. Since every proc must receive the modex data regardless of the collective used to send it, just reducing the size of the process name removes nearly 400MBytes of communication from a 32k proc job (admittedly, much of this comm may occur in parallel). So it does add up pretty quickly. 3. routing RML messages to reduce connections. The default messaging system remains point-to-point - i.e., each proc opens a socket to every proc it communicates with and sends its messages directly. A new option uses the orteds as routers - i.e., each proc only opens a single socket to its local orted. All messages are sent from the proc to the orted, which forwards the message to the orted on the node where the intended recipient proc is located - that orted then forwards the message to its local proc (the recipient). This greatly reduces the connection storm we have encountered during startup. It also has the benefit of removing the sharing of every proc's OOB contact with every other proc. The orted routing tables are populated during launch since every orted gets a map of where every proc is being placed. Each proc, therefore, only needs to know the contact info for its local daemon, which is passed in via the environment when the proc is fork/exec'd by the daemon. This alone removes ~50 bytes/process of communication that was in the current STG1 startup message - so for our 32k proc job, this saves us roughly 32k*50 = 1.6MBytes sent to 32k procs = 51GBytes of messaging. Note that you can use the new routing method by specifying -mca routed tree - if you so desire. This mode will become the default at some point in the future. There are a few minor additional changes in the commit that I'll just note in passing: * propagation of command line mca params to the orteds - fixes ticket #1073. See note there for details. * requiring of "finalize" prior to "exit" for MPI procs - fixes ticket #1144. See note there for details. * cleanup of some stale header files This commit was SVN r16364.
869 строки
27 KiB
C
869 строки
27 KiB
C
/*
|
|
* Copyright (c) 2004-2007 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) 2007 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
/**
|
|
* @file
|
|
* ORTE Checkpoint Tool for checkpointing a multiprocess job
|
|
*
|
|
*/
|
|
|
|
#include "orte_config.h"
|
|
|
|
#include <stdio.h>
|
|
#include <errno.h>
|
|
#ifdef HAVE_STDLIB_H
|
|
#include <stdlib.h>
|
|
#endif /* HAVE_STDLIB_H */
|
|
#ifdef HAVE_UNISTD_H
|
|
#include <unistd.h>
|
|
#endif /* HAVE_UNISTD_H */
|
|
#ifdef HAVE_FCNTL_H
|
|
#include <fcntl.h>
|
|
#endif /* HAVE_FCNTL_H */
|
|
#ifdef HAVE_SYS_TYPES_H
|
|
#include <sys/types.h>
|
|
#endif /* HAVE_SYS_TYPES_H */
|
|
#ifdef HAVE_SYS_STAT_H
|
|
#include <sys/stat.h> /* for mkfifo */
|
|
#endif /* HAVE_SYS_STAT_H */
|
|
#ifdef HAVE_SYS_WAIT_H
|
|
#include <sys/wait.h>
|
|
#endif /* HAVE_SYS_WAIT_H */
|
|
#ifdef HAVE_STRING_H
|
|
#include <string.h>
|
|
#endif /* HAVE_STRING_H */
|
|
|
|
|
|
#include "opal/util/cmd_line.h"
|
|
#include "opal/util/argv.h"
|
|
#include "opal/util/show_help.h"
|
|
#include "opal/util/opal_environ.h"
|
|
#include "opal/util/output.h"
|
|
#include "opal/util/os_path.h"
|
|
#include "opal/util/os_dirpath.h"
|
|
#include "opal/mca/base/base.h"
|
|
#include "opal/mca/base/mca_base_param.h"
|
|
#include "opal/mca/crs/crs.h"
|
|
#include "opal/mca/crs/base/base.h"
|
|
#include "opal/runtime/opal.h"
|
|
#include "opal/runtime/opal_cr.h"
|
|
|
|
#include "orte/orte_constants.h"
|
|
#include "orte/runtime/runtime.h"
|
|
#include "orte/runtime/orte_cr.h"
|
|
#include "orte/mca/gpr/gpr.h"
|
|
#include "orte/mca/rml/rml.h"
|
|
#include "orte/dss/dss.h"
|
|
#include "orte/util/univ_info.h"
|
|
#include "orte/util/sys_info.h"
|
|
#include "orte/util/proc_info.h"
|
|
#include "opal/util/os_path.h"
|
|
#include "orte/util/session_dir.h"
|
|
#include "orte/mca/snapc/snapc.h"
|
|
#include "orte/mca/snapc/base/base.h"
|
|
|
|
/******************
|
|
* Local Functions
|
|
******************/
|
|
static int ckpt_init(int argc, char *argv[]); /* Initalization routine */
|
|
static int ckpt_finalize(void); /* Finalization routine */
|
|
static int parse_args(int argc, char *argv[]);
|
|
static int notify_process_for_checkpoint(char **global_snapshot_handle, int *seq_num, int term);
|
|
static int contact_hnp(orte_process_name_t *peer, char *global_snapshot_handle, int term);
|
|
static int wait_for_checkpoint(orte_process_name_t *peer, char **global_snapshot_handle, int *seq_num);
|
|
static int find_universe(void);
|
|
static int pretty_print_status(int state, char * snapshot_ref);
|
|
static int pretty_print_reference(int seq, char * snapshot_ref);
|
|
|
|
/*****************************************
|
|
* Global Vars for Command line Arguments
|
|
*****************************************/
|
|
enum orte_checkpoint_stage_t {
|
|
ORTE_CKPT_STAGE_NULL,
|
|
ORTE_CKPT_STAGE_INIT_OPAL_UTIL,
|
|
ORTE_CKPT_STAGE_INIT_OPAL,
|
|
ORTE_CKPT_STAGE_INIT_ORTE,
|
|
ORTE_CKPT_STAGE_FINALIZE
|
|
|
|
};
|
|
typedef enum orte_checkpoint_stage_t orte_checkpoint_stage_t;
|
|
|
|
typedef struct {
|
|
bool help;
|
|
int pid;
|
|
bool term;
|
|
bool verbose;
|
|
char *req_universe_name; /**< User Requested Universe */
|
|
int stage; /* Has completed init fully */
|
|
bool nowait; /* Do not wait for checkpoint to complete before returning */
|
|
bool status; /* Display status messages while checkpoint is progressing */
|
|
int output;
|
|
int ckpt_status;
|
|
} orte_checkpoint_globals_t;
|
|
|
|
orte_checkpoint_globals_t orte_checkpoint_globals;
|
|
|
|
opal_cmd_line_init_t cmd_line_opts[] = {
|
|
{ NULL, NULL, NULL,
|
|
'h', NULL, "help",
|
|
0,
|
|
&orte_checkpoint_globals.help, OPAL_CMD_LINE_TYPE_BOOL,
|
|
"This help message" },
|
|
|
|
{ NULL, NULL, NULL,
|
|
'v', NULL, "verbose",
|
|
0,
|
|
&orte_checkpoint_globals.verbose, OPAL_CMD_LINE_TYPE_BOOL,
|
|
"Be Verbose" },
|
|
|
|
{ NULL, NULL, NULL,
|
|
'\0', NULL, "term",
|
|
0,
|
|
&orte_checkpoint_globals.term, OPAL_CMD_LINE_TYPE_BOOL,
|
|
"Terminate the application after checkpoint" },
|
|
|
|
{ NULL, NULL, NULL,
|
|
'w', NULL, "nowait",
|
|
0,
|
|
&orte_checkpoint_globals.nowait, OPAL_CMD_LINE_TYPE_BOOL,
|
|
"Do not wait for the application to finish checkpointing before returning" },
|
|
|
|
{ NULL, NULL, NULL,
|
|
's', NULL, "status",
|
|
0,
|
|
&orte_checkpoint_globals.status, OPAL_CMD_LINE_TYPE_BOOL,
|
|
"Display status messages describing the progression of the checkpoint" },
|
|
|
|
{ "universe", NULL, NULL,
|
|
'\0', NULL, "universe",
|
|
1,
|
|
&orte_checkpoint_globals.req_universe_name, OPAL_CMD_LINE_TYPE_STRING,
|
|
"Set the universe name as username@hostname:universe_name for this "
|
|
"application. This should be the universe of the application that you wish "
|
|
"to checkpoint." },
|
|
|
|
/* End of list */
|
|
{ NULL, NULL, NULL, '\0', NULL, NULL, 0,
|
|
NULL, OPAL_CMD_LINE_TYPE_NULL,
|
|
NULL }
|
|
};
|
|
|
|
int
|
|
main(int argc, char *argv[])
|
|
{
|
|
int ret, exit_status = ORTE_SUCCESS;
|
|
char *global_snapshot_handle;
|
|
int seq_num = -1;
|
|
|
|
/***************
|
|
* Initialize
|
|
***************/
|
|
if (ORTE_SUCCESS != (ret = ckpt_init(argc, argv))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*******************************
|
|
* Checkpoint the requested PID
|
|
*******************************/
|
|
if( orte_checkpoint_globals.verbose ) {
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"orte_checkpoint: Checkpointing...");
|
|
if(0 < orte_checkpoint_globals.pid) {
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"\t PID %d",
|
|
orte_checkpoint_globals.pid);
|
|
}
|
|
|
|
if(NULL != orte_checkpoint_globals.req_universe_name) {
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"\t Universe (%s)",
|
|
orte_checkpoint_globals.req_universe_name);
|
|
}
|
|
else {
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"\t Connected to Universe (%s)",
|
|
orte_universe_info.name);
|
|
}
|
|
|
|
if(orte_checkpoint_globals.term) {
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"\t Terminating after checkpoint\n");
|
|
}
|
|
}
|
|
|
|
if(ORTE_SUCCESS != (ret = notify_process_for_checkpoint(&global_snapshot_handle,
|
|
&seq_num,
|
|
orte_checkpoint_globals.term)) ) {
|
|
opal_show_help("help-orte-checkpoint.txt", "ckpt_failure", true,
|
|
orte_checkpoint_globals.pid, ret);
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
if( ORTE_SNAPC_CKPT_STATE_ERROR == orte_checkpoint_globals.ckpt_status ) {
|
|
opal_show_help("help-orte-checkpoint.txt", "ckpt_failure", true,
|
|
orte_checkpoint_globals.pid, ORTE_ERROR);
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if( orte_checkpoint_globals.status ) {
|
|
pretty_print_status(ORTE_SNAPC_CKPT_STATE_FINISHED, global_snapshot_handle);
|
|
}
|
|
|
|
if(!orte_checkpoint_globals.nowait) {
|
|
pretty_print_reference(seq_num, global_snapshot_handle);
|
|
}
|
|
|
|
cleanup:
|
|
/***************
|
|
* Cleanup
|
|
***************/
|
|
if (ORTE_SUCCESS != (ret = ckpt_finalize())) {
|
|
return ret;
|
|
}
|
|
|
|
return exit_status;
|
|
}
|
|
|
|
static int parse_args(int argc, char *argv[]) {
|
|
int i, ret, len, exit_status = ORTE_SUCCESS ;
|
|
opal_cmd_line_t cmd_line;
|
|
char **app_env = NULL, **global_env = NULL;
|
|
|
|
/* Init structure */
|
|
memset(&orte_checkpoint_globals, 0, sizeof(orte_checkpoint_globals_t));
|
|
orte_checkpoint_globals.help = false;
|
|
orte_checkpoint_globals.pid = -1;
|
|
orte_checkpoint_globals.term = false;
|
|
orte_checkpoint_globals.verbose = false;
|
|
orte_checkpoint_globals.req_universe_name = NULL;
|
|
orte_checkpoint_globals.stage = ORTE_CKPT_STAGE_NULL;
|
|
orte_checkpoint_globals.nowait = false;
|
|
orte_checkpoint_globals.status = false;
|
|
orte_checkpoint_globals.output = -1;
|
|
orte_checkpoint_globals.ckpt_status = ORTE_SNAPC_CKPT_STATE_NONE;
|
|
|
|
/* Parse the command line options */
|
|
opal_cmd_line_create(&cmd_line, cmd_line_opts);
|
|
mca_base_open();
|
|
mca_base_cmd_line_setup(&cmd_line);
|
|
ret = opal_cmd_line_parse(&cmd_line, true, argc, argv);
|
|
|
|
/**
|
|
* Put all of the MCA arguments in the environment
|
|
*/
|
|
mca_base_cmd_line_process_args(&cmd_line, &app_env, &global_env);
|
|
|
|
len = opal_argv_count(app_env);
|
|
for(i = 0; i < len; ++i) {
|
|
putenv(app_env[i]);
|
|
}
|
|
|
|
len = opal_argv_count(global_env);
|
|
for(i = 0; i < len; ++i) {
|
|
putenv(global_env[i]);
|
|
}
|
|
|
|
opal_setenv(mca_base_param_env_var("opal_cr_is_tool"),
|
|
"1",
|
|
true, &environ);
|
|
|
|
/**
|
|
* Now start parsing our specific arguments
|
|
*/
|
|
/* get the remaining bits */
|
|
opal_cmd_line_get_tail(&cmd_line, &argc, &argv);
|
|
|
|
if (OPAL_SUCCESS != ret ||
|
|
orte_checkpoint_globals.help ||
|
|
(0 >= argc && NULL == orte_checkpoint_globals.req_universe_name)) {
|
|
char *args = NULL;
|
|
args = opal_cmd_line_get_usage_msg(&cmd_line);
|
|
opal_show_help("help-orte-checkpoint.txt", "usage", true,
|
|
args);
|
|
free(args);
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* If the user did not supply a universe, then they must
|
|
* supply the PID of MPIRUN
|
|
*/
|
|
if(0 >= argc &&
|
|
NULL != orte_checkpoint_globals.req_universe_name) {
|
|
exit_status = ORTE_SUCCESS;
|
|
goto cleanup;
|
|
}
|
|
|
|
orte_checkpoint_globals.pid = atoi(argv[0]);
|
|
if ( 0 >= orte_checkpoint_globals.pid ) {
|
|
opal_show_help("help-orte-checkpoint.txt", "invalid_pid", true,
|
|
orte_checkpoint_globals.pid);
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* JJH: No wait is currently not implemented or tested
|
|
*/
|
|
if(orte_checkpoint_globals.nowait) {
|
|
orte_checkpoint_globals.nowait = false;
|
|
opal_show_help("help-orte-checkpoint.txt", "not_impl",
|
|
true,
|
|
"Disconnected checkpoint");
|
|
}
|
|
|
|
cleanup:
|
|
return exit_status;
|
|
}
|
|
|
|
static int
|
|
notify_process_for_checkpoint(char **global_snapshot_handle, int *seq_num, int term)
|
|
{
|
|
int ret, exit_status = ORTE_SUCCESS;
|
|
orte_process_name_t peer;
|
|
|
|
peer = *ORTE_PROC_MY_HNP;
|
|
|
|
/*
|
|
* Contact HNP via RML
|
|
* The notification will be received by the Global Snapshot Coordinator [SnapC]
|
|
* in the HNP(s)
|
|
* See orte_snapc(7) for more information.
|
|
*/
|
|
if( ORTE_SUCCESS != (ret = contact_hnp(&peer, *global_snapshot_handle, term)) ) {
|
|
opal_show_help("help-orte-checkpoint.txt", "unable_to_connect", true,
|
|
orte_checkpoint_globals.pid);
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* Wait for the global_snapshot_coordinator to notify us (via the RML)
|
|
* of the completion of the checkpoint.
|
|
* Unless the user wants us to return immediately
|
|
*/
|
|
if(!orte_checkpoint_globals.nowait) {
|
|
/*
|
|
* Wait for progress updates, stop waiting when 'Finished' status
|
|
*/
|
|
do {
|
|
if( ORTE_SUCCESS != (ret = wait_for_checkpoint(&peer, global_snapshot_handle, seq_num) ) ) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* If process said that it cannot checkpoint at this time return a
|
|
* pretty message.
|
|
*/
|
|
if( ORTE_SNAPC_CKPT_STATE_NO_CKPT == orte_checkpoint_globals.ckpt_status ) {
|
|
opal_show_help("help-orte-checkpoint.txt", "non-ckptable",
|
|
true,
|
|
orte_checkpoint_globals.pid);
|
|
exit_status = ORTE_ERROR;
|
|
break;
|
|
}
|
|
/*
|
|
* If we are to display the status progression
|
|
*/
|
|
if( orte_checkpoint_globals.status ) {
|
|
if(ORTE_SNAPC_CKPT_STATE_FINISHED != orte_checkpoint_globals.ckpt_status)
|
|
pretty_print_status(orte_checkpoint_globals.ckpt_status, *global_snapshot_handle);
|
|
}
|
|
/*
|
|
* Otherwise only display it if we are going to be terminated soon
|
|
*/
|
|
else {
|
|
/* Since ORTE kills us before we get the Finished message,
|
|
* print out the global snapshot handle when we start running
|
|
*/
|
|
if(orte_checkpoint_globals.term &&
|
|
ORTE_SNAPC_CKPT_STATE_RUNNING == orte_checkpoint_globals.ckpt_status ) {
|
|
pretty_print_status(orte_checkpoint_globals.ckpt_status, *global_snapshot_handle);
|
|
}
|
|
}
|
|
} while(ORTE_SNAPC_CKPT_STATE_FINISHED != orte_checkpoint_globals.ckpt_status &&
|
|
ORTE_SNAPC_CKPT_STATE_ERROR != orte_checkpoint_globals.ckpt_status );
|
|
}
|
|
|
|
cleanup:
|
|
return exit_status;
|
|
}
|
|
|
|
/*
|
|
* This function attempts to:
|
|
* 1. Find the universe that matches one or both of the following:
|
|
* - --universe specified by the user (if any)
|
|
* - PID specified by the user (if any)
|
|
* 2. Attach orte_checkpoint to that universe, so we can talk to
|
|
* it's GPR.
|
|
*/
|
|
static int find_universe(void) {
|
|
int ret, exit_status = ORTE_SUCCESS;
|
|
char *fulldirpath = NULL,
|
|
*prefix = NULL,
|
|
*frontend = NULL,
|
|
*univ_name = NULL,
|
|
*full_univ = NULL;
|
|
|
|
/*
|
|
* If the user specified a universe, trust it as correct
|
|
*/
|
|
if( NULL != orte_checkpoint_globals.req_universe_name ) {
|
|
exit_status = ORTE_SUCCESS;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* Otherwise check to see if we can find it by looking for the
|
|
* default constructed with the PID
|
|
*/
|
|
asprintf(&univ_name, "%s-%d", ORTE_DEFAULT_UNIVERSE, orte_checkpoint_globals.pid);
|
|
|
|
if( ORTE_SUCCESS != (ret = orte_session_dir_get_name(&fulldirpath,
|
|
&prefix,
|
|
&frontend,
|
|
orte_system_info.user,
|
|
orte_system_info.nodename,
|
|
NULL, /* Unknown batchid */
|
|
univ_name,
|
|
NULL, /* Unknown Jobid */
|
|
NULL) /* Unknown process ID */
|
|
)) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"orte_checkpoint: find_universe: Trying to find the session directory\n\t\t(%s)\n",
|
|
fulldirpath);
|
|
|
|
/*
|
|
* Check that the directory is accessable.
|
|
*/
|
|
if( ORTE_SUCCESS != (ret = opal_os_dirpath_access(fulldirpath, 0) )) {
|
|
opal_show_help("help-orte-checkpoint.txt", "no_universe", true,
|
|
orte_checkpoint_globals.pid, fulldirpath);
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/*
|
|
* Set the MCA parameter
|
|
* username@hostname:universe_name
|
|
*/
|
|
asprintf(&full_univ, "%s@%s:%s",
|
|
orte_system_info.user,
|
|
orte_system_info.nodename,
|
|
univ_name);
|
|
opal_setenv(mca_base_param_env_var("universe"),
|
|
full_univ,
|
|
true, &environ);
|
|
|
|
cleanup:
|
|
if(NULL != full_univ)
|
|
free(full_univ);
|
|
if(NULL != fulldirpath)
|
|
free(fulldirpath);
|
|
if(NULL != prefix)
|
|
free(prefix);
|
|
if(NULL != frontend)
|
|
free(frontend);
|
|
if(NULL != univ_name)
|
|
free(univ_name);
|
|
|
|
return exit_status;
|
|
}
|
|
|
|
static int ckpt_init(int argc, char *argv[]) {
|
|
int exit_status = ORTE_SUCCESS, ret;
|
|
|
|
/*
|
|
* Make sure to init util before parse_args
|
|
* to ensure installdirs is setup properly
|
|
* before calling mca_base_open();
|
|
*/
|
|
if( ORTE_SUCCESS != (ret = opal_init_util()) ) {
|
|
return ret;
|
|
}
|
|
orte_checkpoint_globals.stage = ORTE_CKPT_STAGE_INIT_OPAL_UTIL;
|
|
|
|
/*
|
|
* Parse Command Line Arguments
|
|
*/
|
|
if (ORTE_SUCCESS != (ret = parse_args(argc, argv))) {
|
|
return ret;
|
|
}
|
|
|
|
/*
|
|
* Setup OPAL Output handle from the verbose argument
|
|
*/
|
|
if( orte_checkpoint_globals.verbose ) {
|
|
orte_checkpoint_globals.output = opal_output_open(NULL);
|
|
opal_output_set_verbosity(orte_checkpoint_globals.output, 10);
|
|
} else {
|
|
orte_checkpoint_globals.output = 0; /* Default=STDOUT */
|
|
}
|
|
|
|
/*
|
|
* We are trying to attach to another process' GPR so we need to
|
|
* attach no matter if it is identified as private or not.
|
|
*/
|
|
opal_setenv(mca_base_param_env_var("universe_console"),
|
|
"1",
|
|
true, &environ);
|
|
|
|
/* Disable the checkpoint notification routine for this
|
|
* tool. As we will never need to checkpoint this tool.
|
|
* Note: This must happen before opal_init().
|
|
*/
|
|
opal_cr_set_enabled(false);
|
|
|
|
/* Select the none component, since we don't actually use a checkpointer */
|
|
opal_setenv(mca_base_param_env_var("crs"),
|
|
"none",
|
|
true, &environ);
|
|
|
|
/***************************
|
|
* We need all of OPAL
|
|
***************************/
|
|
if (ORTE_SUCCESS != (ret = opal_init())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
orte_checkpoint_globals.stage = ORTE_CKPT_STAGE_INIT_OPAL;
|
|
|
|
/***************************
|
|
* And ORTE, but need to do a bit of a dance first
|
|
***************************/
|
|
/* register handler for errnum -> string converstion */
|
|
opal_error_register("ORTE", ORTE_ERR_BASE, ORTE_ERR_MAX, orte_err2str);
|
|
|
|
/* Register all MCA Params */
|
|
if (ORTE_SUCCESS != (ret = orte_register_params(true))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/* Ensure the system_info structure is instantiated and initialized */
|
|
if (ORTE_SUCCESS != (ret = orte_sys_info())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/* Ensure the process info structure is instantiated and initialized */
|
|
if (ORTE_SUCCESS != (ret = orte_proc_info())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/***************************
|
|
* Find the universe that we need to connect to, if it exists
|
|
***************************/
|
|
if (ORTE_SUCCESS != (ret = find_universe())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
/* JJH XXX
|
|
* JJH XXX In actuality, we only want to setup upto the RML
|
|
* JJH XXX so we can talk to the HNP over the wire, but don't
|
|
* JJH XXX become a job of the universe.
|
|
* JJH XXX This is a bandaid until we do it right.
|
|
* JJH XXX
|
|
*/
|
|
if (ORTE_SUCCESS != (ret = orte_init(ORTE_INFRASTRUCTURE)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
orte_checkpoint_globals.stage = ORTE_CKPT_STAGE_INIT_ORTE;
|
|
|
|
cleanup:
|
|
return exit_status;
|
|
}
|
|
|
|
static int ckpt_finalize(void) {
|
|
int exit_status = ORTE_SUCCESS, ret;
|
|
|
|
if( ORTE_CKPT_STAGE_INIT_OPAL_UTIL == orte_checkpoint_globals.stage) {
|
|
if (ORTE_SUCCESS != (ret = opal_finalize_util())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
}
|
|
else if( ORTE_CKPT_STAGE_INIT_OPAL == orte_checkpoint_globals.stage) {
|
|
if (ORTE_SUCCESS != (ret = opal_finalize())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
}
|
|
else {
|
|
if (ORTE_SUCCESS != (ret = orte_finalize())) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
}
|
|
|
|
cleanup:
|
|
return exit_status;
|
|
}
|
|
|
|
static int contact_hnp(orte_process_name_t *peer, char *global_snapshot_handle, int term) {
|
|
int ret, exit_status = ORTE_SUCCESS;
|
|
orte_buffer_t *buffer;
|
|
size_t command;
|
|
pid_t hnp_pid;
|
|
orte_std_cntr_t n;
|
|
/* JJH XXX currently we assume jobid = 1, don't do this in the future */
|
|
orte_jobid_t jobid = 1;
|
|
|
|
if (NULL == (buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/***********************************
|
|
* Notify HNP that we want to chat about a checkpoint
|
|
***********************************/
|
|
command = ORTE_SNAPC_GLOBAL_INIT_CMD;
|
|
if (ORTE_SUCCESS != (ret = orte_dss.pack(buffer, &command, 1, ORTE_CKPT_CMD)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
if ( 0 > (ret = orte_rml.send_buffer(peer, buffer, ORTE_RML_TAG_CKPT, 0)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/********************
|
|
* Receive the PID of the HNP, just to be doubly sure we are talking to
|
|
* the right HNP.
|
|
********************/
|
|
if( 0 > (ret = orte_rml.recv_buffer(peer, buffer, ORTE_RML_TAG_CKPT, 0)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
n = 1;
|
|
if ( ORTE_SUCCESS != (ret = orte_dss.unpack(buffer, &hnp_pid, &n, ORTE_PID)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"orte_checkpoint: contact_hnp: Head Node Process PID = %d\n",
|
|
hnp_pid);
|
|
opal_output_verbose(10, orte_checkpoint_globals.output,
|
|
"orte_checkpoint: contact_hnp: Expected PID = %d\n",
|
|
orte_checkpoint_globals.pid);
|
|
|
|
if(hnp_pid != orte_checkpoint_globals.pid) {
|
|
opal_show_help("help-orte-checkpoint.txt", "invalid_pid", true,
|
|
orte_checkpoint_globals.pid);
|
|
exit_status = ORTE_ERROR;
|
|
orte_snapc_base_global_coord_send_ack(peer, false);
|
|
goto cleanup;
|
|
}
|
|
|
|
if( ORTE_SUCCESS != (ret = orte_snapc_base_global_coord_send_ack(peer, true)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/**************************
|
|
* Send over the term flag
|
|
**************************/
|
|
OBJ_RELEASE(buffer);
|
|
if (NULL == (buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (ORTE_SUCCESS != (ret = orte_dss.pack(buffer, &term, 1, ORTE_BOOL))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (0 > (ret = orte_rml.send_buffer(peer, buffer, ORTE_RML_TAG_CKPT, 0))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/**************************
|
|
* Send over the jobid flag
|
|
**************************/
|
|
OBJ_RELEASE(buffer);
|
|
if (NULL == (buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (ORTE_SUCCESS != (ret = orte_dss.pack(buffer, &jobid, 1, ORTE_SIZE))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (0 > (ret = orte_rml.send_buffer(peer, buffer, ORTE_RML_TAG_CKPT, 0))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
cleanup:
|
|
OBJ_RELEASE(buffer);
|
|
|
|
return exit_status;
|
|
}
|
|
|
|
static int wait_for_checkpoint(orte_process_name_t *peer, char **global_snapshot_handle, int *seq_num) {
|
|
int ret, exit_status = ORTE_SUCCESS;
|
|
orte_buffer_t *loc_buffer;
|
|
orte_std_cntr_t n = 1;
|
|
size_t str_len = 0;
|
|
int ckpt_status = ORTE_SNAPC_CKPT_STATE_NONE;
|
|
|
|
if (NULL == (loc_buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
/******************
|
|
* Receive the checkpoint status
|
|
******************/
|
|
OBJ_RELEASE(loc_buffer);
|
|
if (NULL == (loc_buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if( ORTE_SUCCESS != (ret = orte_rml.recv_buffer(peer, loc_buffer, ORTE_RML_TAG_CKPT, 0))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
n = 1;
|
|
if ( ORTE_SUCCESS != (ret = orte_dss.unpack(loc_buffer, &ckpt_status, &n, ORTE_INT)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
orte_checkpoint_globals.ckpt_status = ckpt_status;
|
|
|
|
/* ACK */
|
|
if( ORTE_SUCCESS != (ret = orte_snapc_base_global_coord_send_ack(peer, true)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/* If we cannot checkpoint, then just skip to the end */
|
|
if( ORTE_SNAPC_CKPT_STATE_NO_CKPT == ckpt_status) {
|
|
*global_snapshot_handle = NULL;
|
|
goto cleanup;
|
|
}
|
|
|
|
/******************
|
|
* Receive the size of the global snapshot handle
|
|
******************/
|
|
OBJ_RELEASE(loc_buffer);
|
|
if (NULL == (loc_buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if( ORTE_SUCCESS != (ret = orte_rml.recv_buffer(peer, loc_buffer, ORTE_RML_TAG_CKPT, 0))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
n = 1;
|
|
if ( ORTE_SUCCESS != (ret = orte_dss.unpack(loc_buffer, &str_len, &n, ORTE_SIZE)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/* ACK */
|
|
if( ORTE_SUCCESS != (ret = orte_snapc_base_global_coord_send_ack(peer, true)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/******************
|
|
* Receive the global snapshot handle
|
|
******************/
|
|
OBJ_RELEASE(loc_buffer);
|
|
if (NULL == (loc_buffer = OBJ_NEW(orte_buffer_t))) {
|
|
exit_status = ORTE_ERROR;
|
|
goto cleanup;
|
|
}
|
|
|
|
if( ORTE_SUCCESS != (ret = orte_rml.recv_buffer(peer, loc_buffer, ORTE_RML_TAG_CKPT, 0))) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
n = 1;
|
|
if ( ORTE_SUCCESS != (ret = orte_dss.unpack(loc_buffer, global_snapshot_handle, &n, ORTE_STRING)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
n = 1;
|
|
if ( ORTE_SUCCESS != (ret = orte_dss.unpack(loc_buffer, seq_num, &n, ORTE_INT)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
/* ACK */
|
|
if( ORTE_SUCCESS != (ret = orte_snapc_base_global_coord_send_ack(peer, true)) ) {
|
|
exit_status = ret;
|
|
goto cleanup;
|
|
}
|
|
|
|
cleanup:
|
|
OBJ_RELEASE(loc_buffer);
|
|
|
|
return exit_status;
|
|
}
|
|
|
|
static int pretty_print_status(int state, char * snapshot_ref) {
|
|
char * state_str = NULL;
|
|
|
|
state_str = orte_snapc_ckpt_state_str(state);
|
|
|
|
opal_output(orte_checkpoint_globals.output,
|
|
"%*s - Global Snapshot Reference: %s\n",
|
|
25, state_str, snapshot_ref);
|
|
if( NULL != state_str) {
|
|
free(state_str);
|
|
}
|
|
|
|
return ORTE_SUCCESS;
|
|
}
|
|
|
|
static int pretty_print_reference(int seq, char * snapshot_ref) {
|
|
|
|
printf("Snapshot Ref.: %3d %s\n",
|
|
seq, snapshot_ref);
|
|
|
|
return ORTE_SUCCESS;
|
|
}
|