1
1
openmpi/orte/tools/orte-clean/orte-clean.c
Ralph Castain 54b2cf747e These changes were mostly captured in a prior RFC (except for #2 below) and are aimed specifically at improving startup performance and setting up the remaining modifications described in that RFC.
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 The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info.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The other stage gates have been replaced by orte_grpcomm barrier functions.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.
2007-10-05 19:48:23 +00:00

382 строки
12 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 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2007 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include <stdio.h>
#include <errno.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif /* HAVE_SYS_STAT_H */
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif /* HAVE_SYS_TYPES_H */
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif /* HAVE_SYS_WAIT_H */
#ifdef HAVE_SYS_PARAM_H
#include <sys/param.h>
#endif /* HAVE_SYS_PARAM_H */
#ifdef HAVE_STRING_H
#include <string.h>
#endif /* HAVE_STRING_H */
#ifdef HAVE_DIRENT_H
#include <dirent.h>
#endif /* HAVE_DIRENT_H */
#include <signal.h>
#include "orte/orte_constants.h"
#include "opal/util/cmd_line.h"
#include "opal/util/argv.h"
#include "opal/util/show_help.h"
#include "opal/util/output.h"
#include "opal/util/opal_environ.h"
#include "opal/util/os_dirpath.h"
#include "opal/mca/base/base.h"
#include "opal/mca/base/mca_base_param.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/util/universe_setup_file_io.h"
#include "orte/mca/gpr/gpr.h"
#include "orte/mca/rml/rml.h"
#include "opal/runtime/opal.h"
#if OPAL_ENABLE_FT == 1
#include "opal/runtime/opal_cr.h"
#endif
#include "orte/runtime/runtime.h"
/******************
* Local Functions
******************/
static int orte_clean_init(void);
static int parse_args(int argc, char *argv[]);
#if !defined(__WINDOWS__)
static void kill_procs(void);
#endif /* !defined(__WINDOWS__) */
/*****************************************
* Global Vars for Command line Arguments
*****************************************/
typedef struct {
bool help;
bool verbose;
} orte_clean_globals_t;
orte_clean_globals_t orte_clean_globals;
opal_cmd_line_init_t cmd_line_opts[] = {
{ NULL, NULL, NULL,
'h', NULL, "help",
0,
&orte_clean_globals.help, OPAL_CMD_LINE_TYPE_BOOL,
"This help message" },
{ NULL, NULL, NULL,
'v', NULL, "verbose",
0,
&orte_clean_globals.verbose, OPAL_CMD_LINE_TYPE_BOOL,
"Generate verbose output" },
/* End of list */
{ NULL, NULL, NULL,
'\0', NULL, NULL,
0,
NULL, OPAL_CMD_LINE_TYPE_NULL,
NULL }
};
/*
* This utility will do a brute force clean of a node. It will
* attempt to clean up any files in the user's session directory.
* It will also look for any orted and orterun processes that are
* not part of this job, and kill them off.
*/
int
main(int argc, char *argv[])
{
int ret, exit_status = ORTE_SUCCESS;
/***************
* Initialize
***************/
if (ORTE_SUCCESS != (ret = parse_args(argc, argv))) {
return ret;
}
if (ORTE_SUCCESS != (ret = orte_clean_init())) {
exit_status = ret;
goto cleanup;
}
/*
* Clean out all /tmp directories except for our own.
*/
orte_universe_clean_directories(orte_universe_info.name, orte_clean_globals.verbose);
#if !defined(__WINDOWS__)
kill_procs();
#endif /* !defined(__WINDOWS__) */
orte_finalize();
opal_finalize();
cleanup:
return exit_status;
}
/*
* Parse the command line arguments using the functions command
* line utility functions.
*/
static int parse_args(int argc, char *argv[]) {
int ret;
opal_cmd_line_t cmd_line;
orte_clean_globals_t tmp = { false, false };
/* Parse the command line options */
/* NOTE: There is a bug in the PGI 6.2 series that causes the
compiler to choke when copying structs containing bool members
by value. So do a memcpy here instead. */
memcpy(&orte_clean_globals, &tmp, sizeof(tmp));
/*
* Initialize list of available command line options.
*/
opal_cmd_line_create(&cmd_line, cmd_line_opts);
ret = opal_cmd_line_parse(&cmd_line, true, argc, argv);
opal_setenv(mca_base_param_env_var("opal_cr_is_tool"),
"1", true, NULL);
/**
* Now start parsing our specific arguments
*/
if (OPAL_SUCCESS != ret ||
orte_clean_globals.help) {
char *args = NULL;
args = opal_cmd_line_get_usage_msg(&cmd_line);
opal_show_help("help-orte-clean.txt", "usage", true,
args);
free(args);
return ORTE_ERROR;
}
OBJ_DESTRUCT(&cmd_line);
return ORTE_SUCCESS;
}
static int orte_clean_init(void) {
int exit_status = ORTE_SUCCESS, ret;
/*
* 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, NULL);
#if OPAL_ENABLE_FT == 1
/* 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);
#endif
#if 0
/***************************
* We need all of OPAL
* RHC: this is included in orte_init,
* so I don't think this is needed right now.
***************************/
if (ORTE_SUCCESS != (ret = opal_init())) {
exit_status = ret;
goto cleanup;
}
#endif
if (ORTE_SUCCESS != (ret = orte_init(ORTE_INFRASTRUCTURE))) {
exit_status = ret;
goto cleanup;
}
cleanup:
return exit_status;
}
#if !defined(__WINDOWS__)
/*
* This function makes a call to "ps" to find out the processes that
* are running on this node. It then attempts to kill off any orteds
* and orteruns that are not related to this job.
*/
static
void kill_procs(void) {
int ortedpid;
char procname[MAXPATHLEN]; /* only really need 8, but being safe */
char pidstr[MAXPATHLEN]; /* only really need 8, but being safe */
char user[MAXPATHLEN];
int procpid;
FILE *psfile;
bool kill_orteruns = false;
int orunpid = 0;
/*
* This is the command that is used to get the information about
* all the processes that are running. The output looks like the
* following:
* COMMAND PID USER
* tcsh 12556 rolfv
* ps 14424 rolfv
* etc.
* Currently, we do not make use of the USER field, but we may later
* on so we grab it also.
*/
/*
* The configure determines if there is a valid ps command for us to
* use. If it is set to unknown, then we skip this section.
*/
char command[] = ORTE_CLEAN_PS_CMD;
if (!(strcmp("unknown", command))) {
return;
}
/*
* Try to get the pid of our orterun process from our universe name.
* This works in the case where one is using the default universe name
* which appends the pid after the 'default-universe-' string. In
* this way, we avoid killing our own mpirun process. Note that if
* we cannot determine our orterun pid, then we skip killing the
* orterun processes to avoid odd behavior for the user.
*/
if (!(strncmp(ORTE_DEFAULT_UNIVERSE, orte_universe_info.name,
sizeof(ORTE_DEFAULT_UNIVERSE)-1))) {
char *tptr;
/*
* Set a pointer to the pid part of the name. The pointer
* is adjusted by the name along with one extra to remove the
* dash before the pid. Then convert to a pid. If the strtol()
* returns zero, then we got an error on the conversion and we
* will skip killing the orteruns.
*/
tptr = orte_universe_info.name + sizeof(ORTE_DEFAULT_UNIVERSE);
if (0 != (orunpid = (int)strtol(tptr, (char **)NULL, 10))) {
kill_orteruns = true;
}
}
/*
* Get our parent pid which is the pid of the orted.
*/
ortedpid = getppid();
/*
* There is a race condition here. The problem is that we are looking
* for any processes named orted. However, one may erroneously find more
* orteds then there really are because the orted is doing a series of
* fork/execs. If we run with more than one orte-clean on a node, then
* one of the orte-cleans may catch the other one while it has forked,
* but not exec'ed. It will therefore kill an orte-clean. Now one
* can argue it is silly to run more than one orte-clean on a node, and
* this is true. We will have to figure out how to prevent this. For
* now, we use a big hammer and just sleep a second to decrease the
* probability.
*/
sleep(1);
psfile = popen(command, "r");
/*
* Read the first line of the output. We just throw it away
* as it is the header consisting of the words COMMAND, PID and
* USER.
*/
if ((fscanf(psfile, "%s%s%s", procname, pidstr, user)) == EOF) {
return;
}
while ((fscanf(psfile, "%s%s%s", procname, pidstr, user)) != EOF) {
procpid = atoi(pidstr);
/*
* Look for any orteds that are not our parent and attempt to
* kill them. We currently do not worry whether we are the
* owner or not. If we are not, we will just fail to send
* the signal and that is OK. This also allows a root process
* to kill all orteds.
*/
if (!strcmp("orted", procname)) {
if (procpid != ortedpid) {
if (orte_clean_globals.verbose) {
opal_output(0, "orte-clean: found potential rogue orted process"
" (pid=%d,user=%s), sending SIGKILL...\n",
procpid, user);
}
/*
* We ignore the return code here as we do not really
* care whether this worked or not.
*/
(void)kill(procpid, SIGKILL);
}
}
/*
* Now check for any orteruns.
*/
if (kill_orteruns) {
if (!strcmp("orterun", procname)) {
if (procpid != orunpid) {
if (orte_clean_globals.verbose) {
opal_output(0, "orte-clean: found potential rogue orterun process"
" (pid=%d,user=%s), sending SIGKILL...\n",
procpid, user);
}
/*
* We ignore the return code here as we do not really
* care whether this worked or not.
*/
(void)kill(procpid, SIGKILL);
}
}
}
}
return;
}
#endif /* !defined(__WINDOWS__) */