1
1
openmpi/orte/mca/odls/default/odls_default_module.c

433 строки
14 KiB
C
Исходник Обычный вид История

/*
* 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 Evergrid, Inc. All rights reserved.
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <stdlib.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <errno.h>
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#include <signal.h>
#ifdef HAVE_FCNTL_H
#include <fcntl.h>
#endif
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#ifdef HAVE_SYS_PARAM_H
#include <sys/param.h>
#endif
#ifdef HAVE_NETDB_H
#include <netdb.h>
#endif
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif /* HAVE_SYS_STAT_H */
#if defined(HAVE_SCHED_YIELD)
/* Only if we have sched_yield() */
#ifdef HAVE_SCHED_H
#include <sched.h>
#endif
#else
/* Only do these if we don't have <sched.h> */
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
#endif /* HAVE_SCHED_YIELD */
#include "orte/util/show_help.h"
#include "orte/runtime/orte_wait.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/iof/base/iof_base_setup.h"
#include "orte/util/name_fns.h"
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 23:48:23 +04:00
#include "orte/mca/odls/base/odls_private.h"
#include "orte/mca/odls/default/odls_default.h"
/*
* External Interface
*/
static int orte_odls_default_launch_local_procs(opal_buffer_t *data);
static int orte_odls_default_kill_local_procs(orte_jobid_t job, bool set_state);
static int orte_odls_default_signal_local_procs(const orte_process_name_t *proc, int32_t signal);
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 23:48:23 +04:00
static void set_handler_default(int sig);
orte_odls_base_module_t orte_odls_default_module = {
orte_odls_base_default_get_add_procs_data,
orte_odls_default_launch_local_procs,
orte_odls_default_kill_local_procs,
orte_odls_default_signal_local_procs,
orte_odls_base_default_deliver_message,
orte_odls_base_default_require_sync,
orte_odls_base_default_collect_data
};
static bool odls_default_child_died(pid_t pid, unsigned int timeout, int *exit_status)
{
time_t end;
pid_t ret;
#if !defined(HAVE_SCHED_YIELD)
struct timeval t;
fd_set bogus;
#endif
end = time(NULL) + timeout;
do {
ret = waitpid(pid, exit_status, WNOHANG);
if (pid == ret) {
/* It died -- return success */
return true;
} else if (0 == ret) {
/* with NOHANG specified, if a process has already exited
* while waitpid was registered, then waitpid returns 0
* as there is no error - this is a race condition problem
* that occasionally causes us to incorrectly report a proc
* as refusing to die. Unfortunately, errno may not be reset
* by waitpid in this case, so we cannot check it - just assume
* the proc has indeed died
*/
return true;
} else if (-1 == ret && ECHILD == errno) {
/* The pid no longer exists, so we'll call this "good
enough for government work" */
return true;
}
#if defined(HAVE_SCHED_YIELD)
sched_yield();
#else
/* Bogus delay for 1 usec */
t.tv_sec = 0;
t.tv_usec = 1;
FD_ZERO(&bogus);
FD_SET(0, &bogus);
select(1, &bogus, NULL, NULL, &t);
#endif
} while (time(NULL) < end);
/* The child didn't die, so return false */
return false;
}
static int odls_default_kill_local(pid_t pid, int signum)
{
if (0 != kill(pid, signum)) {
if (ESRCH != errno) return errno;
}
return 0;
}
int orte_odls_default_kill_local_procs(orte_jobid_t job, bool set_state)
{
int rc;
if (ORTE_SUCCESS != (rc = orte_odls_base_default_kill_local_procs(job, set_state,
odls_default_kill_local, odls_default_child_died))) {
ORTE_ERROR_LOG(rc);
return rc;
}
return ORTE_SUCCESS;
}
/**
* Fork/exec the specified processes
*/
static int odls_default_fork_local_proc(
orte_app_context_t* context,
orte_odls_child_t *child,
char **environ_copy)
{
orte_iof_base_io_conf_t opts;
int rc;
sigset_t sigs;
int i, p[2];
/* should pull this information from MPIRUN instead of going with
default */
opts.usepty = OMPI_ENABLE_PTY_SUPPORT;
/* BWB - Fix post beta. Should setup stdin in orterun and make
part of the app_context. Do not change this without also
changing the reverse of this in
odls_default_wait_local_proc(). */
if (child->name->vpid == 0) {
opts.connect_stdin = true;
} else {
opts.connect_stdin = false;
}
if (ORTE_SUCCESS != (rc = orte_iof_base_setup_prefork(&opts))) {
ORTE_ERROR_LOG(rc);
child->state = ORTE_PROC_STATE_FAILED_TO_START;
child->exit_code = rc;
return rc;
}
/* A pipe is used to communicate between the parent and child to
2008-01-27 00:36:04 +03:00
indicate whether the exec ultimately succeeded or failed. The
child sets the pipe to be close-on-exec; the child only ever
writes anything to the pipe if there is an error (e.g.,
executable not found, exec() fails, etc.). The parent does a
blocking read on the pipe; if the pipe closed with no data,
then the exec() succeeded. If the parent reads something from
the pipe, then the child was letting us know that it failed. */
if (pipe(p) < 0) {
ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_PIPES);
child->state = ORTE_PROC_STATE_FAILED_TO_START;
child->exit_code = ORTE_ERR_SYS_LIMITS_PIPES;
return ORTE_ERR_SYS_LIMITS_PIPES;
}
/* Fork off the child */
child->pid = fork();
if(child->pid < 0) {
ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_CHILDREN);
child->state = ORTE_PROC_STATE_FAILED_TO_START;
child->exit_code = ORTE_ERR_SYS_LIMITS_CHILDREN;
return ORTE_ERR_SYS_LIMITS_CHILDREN;
}
if (child->pid == 0) {
long fd, fdmax = sysconf(_SC_OPEN_MAX);
/* Setup the pipe to be close-on-exec */
close(p[0]);
fcntl(p[1], F_SETFD, FD_CLOEXEC);
/* setup stdout/stderr so that any error messages that we may
print out will get displayed back at orterun.
NOTE: Definitely do this AFTER we check contexts so that any
error message from those two functions doesn't come out to the
user. IF we didn't do it in this order, THEN a user who gives
us a bad executable name or working directory would get N
error messages, where N=num_procs. This would be very annoying
for large jobs, so instead we set things up so that orterun
always outputs a nice, single message indicating what happened
*/
if (ORTE_SUCCESS != (i = orte_iof_base_setup_child(&opts,
&environ_copy))) {
write(p[1], &i, sizeof(int));
exit(1);
}
/* close all file descriptors w/ exception of
stdin/stdout/stderr and the pipe used for the IOF INTERNAL
messages */
for(fd=3; fd<fdmax; fd++) {
if (fd != opts.p_internal[1]) {
close(fd);
}
}
if (context->argv == NULL) {
context->argv = malloc(sizeof(char*)*2);
context->argv[0] = strdup(context->app);
context->argv[1] = NULL;
}
/* Set signal handlers back to the default. Do this close to
the exev() 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 launched process
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 launched process. */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* Exec the new executable */
execve(context->app, context->argv, environ_copy);
This commit represents a bunch of work on a Mercurial side branch. As such, the commit message back to the master SVN repository is fairly long. = ORTE Job-Level Output Messages = Add two new interfaces that should be used for all new code throughout the ORTE and OMPI layers (we already make the search-and-replace on the existing ORTE / OMPI layers): * orte_output(): (and corresponding friends ORTE_OUTPUT, orte_output_verbose, etc.) This function sends the output directly to the HNP for processing as part of a job-specific output channel. It supports all the same outputs as opal_output() (syslog, file, stdout, stderr), but for stdout/stderr, the output is sent to the HNP for processing and output. More on this below. * orte_show_help(): This function is a drop-in-replacement for opal_show_help(), with two differences in functionality: 1. the rendered text help message output is sent to the HNP for display (rather than outputting directly into the process' stderr stream) 1. the HNP detects duplicate help messages and does not display them (so that you don't see the same error message N times, once from each of your N MPI processes); instead, it counts "new" instances of the help message and displays a message every ~5 seconds when there are new ones ("I got X new copies of the help message...") opal_show_help and opal_output still exist, but they only output in the current process. The intent for the new orte_* functions is that they can apply job-level intelligence to the output. As such, we recommend that all new ORTE and OMPI code use the new orte_* functions, not thei opal_* functions. === New code === For ORTE and OMPI programmers, here's what you need to do differently in new code: * Do not include opal/util/show_help.h or opal/util/output.h. Instead, include orte/util/output.h (this one header file has declarations for both the orte_output() series of functions and orte_show_help()). * Effectively s/opal_output/orte_output/gi throughout your code. Note that orte_output_open() takes a slightly different argument list (as a way to pass data to the filtering stream -- see below), so you if explicitly call opal_output_open(), you'll need to slightly adapt to the new signature of orte_output_open(). * Literally s/opal_show_help/orte_show_help/. The function signature is identical. === Notes === * orte_output'ing to stream 0 will do similar to what opal_output'ing did, so leaving a hard-coded "0" as the first argument is safe. * For systems that do not use ORTE's RML or the HNP, the effect of orte_output_* and orte_show_help will be identical to their opal counterparts (the additional information passed to orte_output_open() will be lost!). Indeed, the orte_* functions simply become trivial wrappers to their opal_* counterparts. Note that we have not tested this; the code is simple but it is quite possible that we mucked something up. = Filter Framework = Messages sent view the new orte_* functions described above and messages output via the IOF on the HNP will now optionally be passed through a new "filter" framework before being output to stdout/stderr. The "filter" OPAL MCA framework is intended to allow preprocessing to messages before they are sent to their final destinations. The first component that was written in the filter framework was to create an XML stream, segregating all the messages into different XML tags, etc. This will allow 3rd party tools to read the stdout/stderr from the HNP and be able to know exactly what each text message is (e.g., a help message, another OMPI infrastructure message, stdout from the user process, stderr from the user process, etc.). Filtering is not active by default. Filter components must be specifically requested, such as: {{{ $ mpirun --mca filter xml ... }}} There can only be one filter component active. = New MCA Parameters = The new functionality described above introduces two new MCA parameters: * '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that help messages will be aggregated, as described above. If set to 0, all help messages will be displayed, even if they are duplicates (i.e., the original behavior). * '''orte_base_show_output_recursions''': An MCA parameter to help debug one of the known issues, described below. It is likely that this MCA parameter will disappear before v1.3 final. = Known Issues = * The XML filter component is not complete. The current output from this component is preliminary and not real XML. A bit more work needs to be done to configure.m4 search for an appropriate XML library/link it in/use it at run time. * There are possible recursion loops in the orte_output() and orte_show_help() functions -- e.g., if RML send calls orte_output() or orte_show_help(). We have some ideas how to fix these, but figured that it was ok to commit before feature freeze with known issues. The code currently contains sub-optimal workarounds so that this will not be a problem, but it would be good to actually solve the problem rather than have hackish workarounds before v1.3 final. This commit was SVN r18434.
2008-05-14 00:00:55 +04:00
orte_show_help("help-odls-default.txt", "orte-odls-default:execv-error",
true, context->app, strerror(errno));
exit(1);
} else {
/* connect endpoints IOF */
rc = orte_iof_base_setup_parent(child->name, &opts);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* Wait to read something from the pipe or close */
close(p[1]);
while (1) {
rc = read(p[0], &i, sizeof(int));
if (rc < 0) {
/* Signal interrupts are ok */
if (errno == EINTR) {
continue;
}
/* Other errno's are bad */
child->state = ORTE_PROC_STATE_FAILED_TO_START;
child->exit_code = ORTE_ERR_PIPE_READ_FAILURE;
OPAL_OUTPUT_VERBOSE((2, orte_odls_globals.output,
"%s odls:default:fork got code %d back from child",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), i));
return ORTE_ERR_PIPE_READ_FAILURE;
break;
} else if (0 == rc) {
/* Child was successful in exec'ing! */
break;
} else {
/* Doh -- child failed.
Let the calling function
know about the failure. The actual exit status of child proc
cannot be found here - all we can do is report the ORTE error
code that was reported back to us. The calling func needs to report the
failure to launch this process through the SMR or else
everyone else will hang.
*/
child->state = ORTE_PROC_STATE_FAILED_TO_START;
child->exit_code = i;
OPAL_OUTPUT_VERBOSE((2, orte_odls_globals.output,
"%s odls:default:fork got code %d back from child",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), i));
return i;
}
}
/* set the proc state to LAUNCHED */
child->state = ORTE_PROC_STATE_LAUNCHED;
child->alive = true;
}
return ORTE_SUCCESS;
}
/**
* Launch all processes allocated to the current node.
*/
int orte_odls_default_launch_local_procs(opal_buffer_t *data)
{
int rc;
orte_jobid_t job;
/* construct the list of children we are to launch */
if (ORTE_SUCCESS != (rc = orte_odls_base_default_construct_child_list(data, &job))) {
OPAL_OUTPUT_VERBOSE((2, orte_odls_globals.output,
"%s odls:default:launch:local failed to construct child list on error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_ERROR_NAME(rc)));
goto CLEANUP;
}
/* launch the local procs */
if (ORTE_SUCCESS != (rc = orte_odls_base_default_launch_local(job, odls_default_fork_local_proc))) {
OPAL_OUTPUT_VERBOSE((2, orte_odls_globals.output,
"%s odls:default:launch:local failed to launch on error %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_ERROR_NAME(rc)));
goto CLEANUP;
}
CLEANUP:
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);
}
static int send_signal(pid_t pid, int signal)
{
int rc = ORTE_SUCCESS;
OPAL_OUTPUT_VERBOSE((1, orte_odls_globals.output,
"%s sending signal %d to pid %ld",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
signal, (long)pid));
if (kill(pid, signal) != 0) {
switch(errno) {
case EINVAL:
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
rc = ORTE_ERR_BAD_PARAM;
break;
case ESRCH:
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
break;
case EPERM:
ORTE_ERROR_LOG(ORTE_ERR_PERM);
rc = ORTE_ERR_PERM;
break;
default:
ORTE_ERROR_LOG(ORTE_ERROR);
rc = ORTE_ERROR;
}
}
return rc;
}
static int orte_odls_default_signal_local_procs(const orte_process_name_t *proc, int32_t signal)
{
int rc;
if (ORTE_SUCCESS != (rc = orte_odls_base_default_signal_local_procs(proc, signal, send_signal))) {
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 23:48:23 +04:00
ORTE_ERROR_LOG(rc);
return rc;
}
return ORTE_SUCCESS;
}