1
1
openmpi/orte/util/hostfile/hostfile.c

951 строка
34 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2008 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
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <errno.h>
#include <string.h>
#include <sys/stat.h>
#include "opal/class/opal_list.h"
#include "opal/util/argv.h"
#include "opal/util/output.h"
#include "opal/mca/mca.h"
#include "opal/mca/base/base.h"
#include "opal/util/if.h"
#include "opal/mca/installdirs/installdirs.h"
#include "orte/util/show_help.h"
#include "orte/util/proc_info.h"
#include "orte/util/name_fns.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ras/ras_types.h"
#include "orte/runtime/orte_globals.h"
#include "orte/util/hostfile/hostfile_lex.h"
#include "orte/util/hostfile/hostfile.h"
static const char *cur_hostfile_name = NULL;
static void hostfile_parse_error(int token)
{
switch (token) {
case ORTE_HOSTFILE_STRING:
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-hostfile.txt", "parse_error_string",
true,
cur_hostfile_name,
orte_util_hostfile_line,
token,
orte_util_hostfile_value.sval);
break;
case ORTE_HOSTFILE_IPV4:
case ORTE_HOSTFILE_IPV6:
case ORTE_HOSTFILE_INT:
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-hostfile.txt", "parse_error_int",
true,
cur_hostfile_name,
orte_util_hostfile_line,
token,
orte_util_hostfile_value.ival);
break;
default:
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-hostfile.txt", "parse_error",
true,
cur_hostfile_name,
orte_util_hostfile_line,
token );
break;
}
}
/**
* Return the integer following an = (actually may only return positive ints)
*/
static int hostfile_parse_int(void)
{
if (ORTE_HOSTFILE_EQUAL != orte_util_hostfile_lex())
return -1;
if (ORTE_HOSTFILE_INT != orte_util_hostfile_lex())
return -1;
return orte_util_hostfile_value.ival;
}
/**
* Return the string following an = (option to a keyword)
*/
static char *hostfile_parse_string(void)
{
int rc;
if (ORTE_HOSTFILE_EQUAL != orte_util_hostfile_lex()){
return NULL;
}
rc = orte_util_hostfile_lex();
if (ORTE_HOSTFILE_STRING != rc){
return NULL;
}
return strdup(orte_util_hostfile_value.sval);
}
static orte_node_t* hostfile_lookup(opal_list_t* nodes, const char* name)
{
opal_list_item_t* item;
for(item = opal_list_get_first(nodes);
item != opal_list_get_end(nodes);
item = opal_list_get_next(item)) {
orte_node_t* node = (orte_node_t*)item;
if(strcmp(node->name, name) == 0) {
opal_list_remove_item(nodes, item);
return node;
}
}
return NULL;
}
static int hostfile_parse_line(int token, opal_list_t* updates, opal_list_t* exclude, bool keep_all)
{
int rc;
orte_node_t* node;
bool got_count = false;
bool got_max = false;
char* value;
char** argv;
char* node_name = NULL;
char* node_alias = NULL;
char* username = NULL;
int cnt;
int number_of_slots = 0;
char buff[64];
if (ORTE_HOSTFILE_STRING == token ||
ORTE_HOSTFILE_HOSTNAME == token ||
ORTE_HOSTFILE_INT == token ||
ORTE_HOSTFILE_IPV4 == token ||
ORTE_HOSTFILE_IPV6 == token) {
if(ORTE_HOSTFILE_INT == token) {
snprintf(buff, 64, "%d", orte_util_hostfile_value.ival);
value = buff;
} else {
value = orte_util_hostfile_value.sval;
}
argv = opal_argv_split (value, '@');
cnt = opal_argv_count (argv);
if (1 == cnt) {
node_name = strdup(argv[0]);
} else if (2 == cnt) {
username = strdup(argv[0]);
node_name = strdup(argv[1]);
} else {
opal_output(0, "WARNING: Unhandled user@host-combination\n"); /* XXX */
}
opal_argv_free (argv);
/* if the first letter of the name is '^', then this is a node
* to be excluded. Remove the ^ character so the nodename is
* usable, and put it on the exclude list
*/
if ('^' == node_name[0]) {
int i, len;
len = strlen(node_name);
for (i=1; i < len; i++) {
node_name[i-1] = node_name[i];
}
node_name[len-1] = '\0'; /* truncate */
OPAL_OUTPUT_VERBOSE((3, orte_debug_output,
"%s hostfile: node %s is being excluded",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node_name));
/* convert this into something globally unique */
if (strcmp(node_name, "localhost") == 0 || opal_ifislocal(node_name)) {
/* Nodename has been allocated, that is for sure */
if (orte_show_resolved_nodenames &&
0 != strcmp(node_name, orte_process_info.nodename)) {
node_alias = strdup(node_name);
}
free (node_name);
node_name = strdup(orte_process_info.nodename);
}
/* Do we need to make a new node object? First check to see
if it's already in the exclude list */
if (NULL == (node = hostfile_lookup(exclude, node_name))) {
node = OBJ_NEW(orte_node_t);
node->name = node_name;
}
/* Note that we need to add this back to the exclude list.
If it was found, we just removed it (in hostfile_lookup()),
so this puts it back. If it was not found, then we have to
add it to the exclude list anyway. */
opal_list_append(exclude, &node->super);
return ORTE_SUCCESS;
}
/* this is not a node to be excluded, so we need to process it and
* add it to the "include" list. See if this host is actually us.
*/
if (strcmp(node_name, "localhost") == 0 || opal_ifislocal(node_name)) {
/* Nodename has been allocated, that is for sure */
if (orte_show_resolved_nodenames &&
0 != strcmp(node_name, orte_process_info.nodename)) {
node_alias = strdup(node_name);
}
free (node_name);
node_name = strdup(orte_process_info.nodename);
}
OPAL_OUTPUT_VERBOSE((3, orte_debug_output,
"%s hostfile: node %s is being included - keep all is %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node_name,
keep_all ? "TRUE" : "FALSE"));
/* Do we need to make a new node object? First check to see
* if we are keeping everything or if it's already in the updates
* list. Because we check keep_all first, if that is set we will
* not do the hostfile_lookup call, and thus won't remove the
* pre-existing node from the updates list
*/
if (keep_all || NULL == (node = hostfile_lookup(updates, node_name))) {
node = OBJ_NEW(orte_node_t);
node->name = node_name;
}
/* do we need to record an alias for this node? */
if (NULL != node_alias) {
/* add to list of aliases for this node - only add if unique */
opal_argv_append_unique_nosize(&node->alias, node_alias, false);
free(node_alias);
}
} else if (ORTE_HOSTFILE_RELATIVE == token) {
/* store this for later processing */
node = OBJ_NEW(orte_node_t);
node->name = strdup(orte_util_hostfile_value.sval);
} else if (ORTE_HOSTFILE_RANK == token) {
/* we can ignore the rank, but we need to extract the node name. we
* first need to shift over to the other side of the equal sign as
* this is where the node name will be
*/
while (!orte_util_hostfile_done &&
ORTE_HOSTFILE_EQUAL != token) {
token = orte_util_hostfile_lex();
}
if (orte_util_hostfile_done) {
/* bad syntax somewhere */
return ORTE_ERROR;
}
/* next position should be the node name */
token = orte_util_hostfile_lex();
if(ORTE_HOSTFILE_INT == token) {
snprintf(buff, 64, "%d", orte_util_hostfile_value.ival);
value = buff;
} else {
value = orte_util_hostfile_value.sval;
}
argv = opal_argv_split (value, '@');
cnt = opal_argv_count (argv);
if (1 == cnt) {
node_name = strdup(argv[0]);
} else if (2 == cnt) {
username = strdup(argv[0]);
node_name = strdup(argv[1]);
} else {
opal_output(0, "WARNING: Unhandled user@host-combination\n"); /* XXX */
}
opal_argv_free (argv);
/* Do we need to make a new node object? First check to see
* if we are keeping everything or if it's already in the updates
* list. Because we check keep_all first, if that is set we will
* not do the hostfile_lookup call, and thus won't remove the
* pre-existing node from the updates list
*/
if (keep_all || NULL == (node = hostfile_lookup(updates, node_name))) {
node = OBJ_NEW(orte_node_t);
node->name = node_name;
}
/* add a slot */
node->slots++;
/* do we need to record an alias for this node? */
if (NULL != node_alias) {
/* add to list of aliases for this node - only add if unique */
opal_argv_append_unique_nosize(&node->alias, node_alias, false);
free(node_alias);
}
/* skip to end of line */
while (!orte_util_hostfile_done &&
ORTE_HOSTFILE_NEWLINE != token) {
token = orte_util_hostfile_lex();
}
opal_list_append(updates, &node->super);
return ORTE_SUCCESS;
} else {
hostfile_parse_error(token);
return ORTE_ERROR;
}
got_count = false;
while (!orte_util_hostfile_done) {
token = orte_util_hostfile_lex();
switch (token) {
case ORTE_HOSTFILE_DONE:
goto done;
case ORTE_HOSTFILE_NEWLINE:
goto done;
case ORTE_HOSTFILE_USERNAME:
node->username = hostfile_parse_string();
break;
case ORTE_HOSTFILE_BOARDS:
rc = hostfile_parse_int();
if (rc < 0) {
orte_show_help("help-hostfile.txt", "boards",
true,
cur_hostfile_name, rc);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
node->boards = rc;
break;
case ORTE_HOSTFILE_SOCKETS_PER_BOARD:
rc = hostfile_parse_int();
if (rc < 0) {
orte_show_help("help-hostfile.txt", "sockets",
true,
cur_hostfile_name, rc);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
node->sockets_per_board = rc;
break;
case ORTE_HOSTFILE_CORES_PER_SOCKET:
rc = hostfile_parse_int();
if (rc < 0) {
orte_show_help("help-hostfile.txt", "cores",
true,
cur_hostfile_name, rc);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
node->cores_per_socket = rc;
break;
case ORTE_HOSTFILE_CPU_SET:
if (NULL != node->cpu_set) {
free(node->cpu_set);
}
node->cpu_set = hostfile_parse_string();
break;
case ORTE_HOSTFILE_COUNT:
case ORTE_HOSTFILE_CPU:
case ORTE_HOSTFILE_SLOTS:
rc = hostfile_parse_int();
if (rc < 0) {
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-hostfile.txt", "slots",
true,
cur_hostfile_name, rc);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
node->slots += rc;
got_count = true;
/* Ensure that slots_max >= slots */
if (node->slots_max != 0 && node->slots_max < node->slots) {
node->slots_max = node->slots;
}
break;
case ORTE_HOSTFILE_SLOTS_MAX:
rc = hostfile_parse_int();
if (rc < 0) {
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-hostfile.txt", "max_slots",
true,
cur_hostfile_name, ((size_t) rc));
OBJ_RELEASE(node);
return ORTE_ERROR;
}
/* Only take this update if it puts us >= node_slots */
if (rc >= node->slots) {
if (node->slots_max != rc) {
node->slots_max = rc;
got_max = true;
}
} else {
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-hostfile.txt", "max_slots_lt",
true,
cur_hostfile_name, node->slots, rc);
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
break;
default:
hostfile_parse_error(token);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
if (number_of_slots > node->slots) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
OBJ_RELEASE(node);
return ORTE_ERROR;
}
}
done:
if (!got_count) {
if (got_max) {
node->slots = node->slots_max;
} else {
++node->slots;
}
}
opal_list_append(updates, &node->super);
return ORTE_SUCCESS;
}
/**
* Parse the specified file into a node list.
*/
static int hostfile_parse(const char *hostfile, opal_list_t* updates, opal_list_t* exclude, bool keep_all)
{
int token;
int rc = ORTE_SUCCESS;
cur_hostfile_name = hostfile;
orte_util_hostfile_done = false;
orte_util_hostfile_in = fopen(hostfile, "r");
if (NULL == orte_util_hostfile_in) {
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-hostfile.txt", "no-hostfile", true, hostfile);
rc = ORTE_ERR_NOT_FOUND;
goto unlock;
}
while (!orte_util_hostfile_done) {
token = orte_util_hostfile_lex();
switch (token) {
case ORTE_HOSTFILE_DONE:
orte_util_hostfile_done = true;
break;
case ORTE_HOSTFILE_NEWLINE:
break;
/*
* This looks odd, since we have several forms of host-definitions:
* hostname just plain as it is, being a ORTE_HOSTFILE_STRING
* IP4s and user@IPv4s
* hostname.domain and user@hostname.domain
*/
case ORTE_HOSTFILE_STRING:
case ORTE_HOSTFILE_INT:
case ORTE_HOSTFILE_HOSTNAME:
case ORTE_HOSTFILE_IPV4:
case ORTE_HOSTFILE_IPV6:
case ORTE_HOSTFILE_RELATIVE:
case ORTE_HOSTFILE_RANK:
rc = hostfile_parse_line(token, updates, exclude, keep_all);
if (ORTE_SUCCESS != rc) {
goto unlock;
}
break;
default:
hostfile_parse_error(token);
goto unlock;
}
}
fclose(orte_util_hostfile_in);
orte_util_hostfile_in = NULL;
unlock:
cur_hostfile_name = NULL;
return rc;
}
/**
* Parse the provided hostfile and add the nodes to the list.
*/
int orte_util_add_hostfile_nodes(opal_list_t *nodes,
bool *override_oversubscribed,
char *hostfile)
{
opal_list_t exclude;
opal_list_item_t *item, *itm;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_debug_output,
"%s hostfile: checking hostfile %s for nodes",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
/* parse the hostfile and add the contents to the list */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, nodes, &exclude, false))) {
goto cleanup;
}
/* parse the nodes to check for any relative node directives */
for (item = opal_list_get_first(nodes);
item != opal_list_get_end(nodes);
item = opal_list_get_next(item)) {
orte_node_t *node=(orte_node_t*)item;
if ('+' == node->name[0]) {
orte_show_help("help-hostfile.txt", "hostfile:relative-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
}
/* remove from the list of nodes those that are in the exclude list */
while(NULL != (item = opal_list_remove_first(&exclude))) {
orte_node_t *exnode = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
orte_node_t *node=(orte_node_t*)itm;
if (0 == strcmp(exnode->name, node->name)) {
/* match - remove it */
opal_list_remove_item(nodes, itm);
OBJ_RELEASE(itm);
break;
}
}
OBJ_RELEASE(item);
}
/* indicate that ORTE should override any oversubscribed conditions
* based on local hardware limits since the user (a) might not have
* provided us any info on the #slots for a node, and (b) the user
* might have been wrong! If we don't check the number of local physical
* processors, then we could be too aggressive on our sched_yield setting
* and cause performance problems.
*/
*override_oversubscribed = true;
cleanup:
OBJ_DESTRUCT(&exclude);
return rc;
}
/* Parse the provided hostfile and filter the nodes that are
* on the input list, removing those that
* are not found in the hostfile
*/
int orte_util_filter_hostfile_nodes(opal_list_t *nodes,
char *hostfile)
{
opal_list_t newnodes, exclude;
opal_list_item_t *item1, *item2, *next, *item3;
orte_node_t *node_from_list, *node_from_file, *node_from_pool, *node3;
int rc = ORTE_SUCCESS;
char *cptr;
int num_empty, nodeidx;
bool want_all_empty = false;
opal_list_t keep;
OPAL_OUTPUT_VERBOSE((1, orte_debug_output,
"%s hostfile: filtering nodes through hostfile %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
/* parse the hostfile and create local list of findings */
OBJ_CONSTRUCT(&newnodes, opal_list_t);
OBJ_CONSTRUCT(&exclude, opal_list_t);
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, &newnodes, &exclude, false))) {
OBJ_DESTRUCT(&newnodes);
return rc;
}
/* remove from the list of newnodes those that are in the exclude list
* since we could have added duplicate names above due to the */
while (NULL != (item1 = opal_list_remove_first(&exclude))) {
node_from_file = (orte_node_t*)item1;
/* check for matches on nodes */
for (item2 = opal_list_get_first(&newnodes);
item2 != opal_list_get_end(&newnodes);
item2 = opal_list_get_next(item2)) {
orte_node_t *node = (orte_node_t*)item2;
if (0 == strcmp(node_from_file->name, node->name)) {
/* match - remove it */
opal_list_remove_item(&newnodes, item2);
OBJ_RELEASE(item2);
break;
}
}
OBJ_RELEASE(item1);
}
/* now check our nodes and keep those that match. We can
* destruct our hostfile list as we go since this won't be needed
*/
OBJ_CONSTRUCT(&keep, opal_list_t);
while (NULL != (item2 = opal_list_remove_first(&newnodes))) {
node_from_file = (orte_node_t*)item2;
next = opal_list_get_next(item2);
/* see if this is a relative node syntax */
if ('+' == node_from_file->name[0]) {
/* see if we specified empty nodes */
if ('e' == node_from_file->name[1] ||
'E' == node_from_file->name[1]) {
/* request for empty nodes - do they want
* all of them?
*/
if (NULL != (cptr = strchr(node_from_file->name, ':'))) {
/* the colon indicates a specific # are requested */
cptr++; /* step past : */
num_empty = strtol(cptr, NULL, 10);
} else {
/* want them all - set num_empty to max */
num_empty = INT_MAX;
want_all_empty = true;
}
/* search the list of nodes provided to us and find those
* that are empty
*/
item1 = opal_list_get_first(nodes);
while (0 < num_empty && item1 != opal_list_get_end(nodes)) {
node_from_list = (orte_node_t*)item1;
next = opal_list_get_next(item1); /* keep our place */
if (0 == node_from_list->slots_inuse) {
/* check to see if this node is explicitly called
* out later - if so, don't use it here
*/
for (item3 = opal_list_get_first(&newnodes);
item3 != opal_list_get_end(&newnodes);
item3 = opal_list_get_next(item3)) {
node3 = (orte_node_t*)item3;
if (0 == strcmp(node3->name, node_from_list->name)) {
/* match - don't use it */
goto skipnode;
}
}
/* remove item from list */
opal_list_remove_item(nodes, item1);
/* xfer to keep list */
opal_list_append(&keep, item1);
--num_empty;
}
skipnode:
item1 = next;
}
/* did they get everything they wanted? */
if (!want_all_empty && 0 < num_empty) {
orte_show_help("help-hostfile.txt", "hostfile:not-enough-empty",
true, num_empty);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
} else if ('n' == node_from_file->name[1] ||
'N' == node_from_file->name[1]) {
/* they want a specific relative node #, so
* look it up on global pool
*/
nodeidx = strtol(&node_from_file->name[2], NULL, 10);
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, nodeidx))) {
/* this is an error */
orte_show_help("help-hostfile.txt", "hostfile:relative-node-not-found",
true, nodeidx, node_from_file->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* search the list of nodes provided to us and find it */
for (item1 = opal_list_get_first(nodes);
item1 != opal_list_get_end(nodes);
item1 = opal_list_get_next(nodes)) {
node_from_list = (orte_node_t*)item1;
if (0 == strcmp(node_from_list->name, node_from_pool->name)) {
/* match - remove item from list */
opal_list_remove_item(nodes, item1);
/* xfer to keep list */
opal_list_append(&keep, item1);
break;
}
}
} else {
/* invalid relative node syntax */
orte_show_help("help-hostfile.txt", "hostfile:invalid-relative-node-syntax",
true, node_from_file->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
} else {
/* we are looking for a specific node on the list
* search the provided list of nodes to see if this
* one is found
*/
for (item1 = opal_list_get_first(nodes);
item1 != opal_list_get_end(nodes);
item1 = opal_list_get_next(item1)) {
node_from_list = (orte_node_t*)item1;
/* since the name in the hostfile might not match
* our local name, and yet still be intended to match,
* we have to check for local interfaces
*/
if (0 == strcmp(node_from_file->name, node_from_list->name) ||
(opal_ifislocal(node_from_list->name) &&
opal_ifislocal(node_from_file->name))) {
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node_from_file->slots < node_from_list->slots) {
node_from_list->slots_alloc = node_from_file->slots;
}
/* remove the node from the list */
opal_list_remove_item(nodes, item1);
/* xfer it to keep list */
opal_list_append(&keep, item1);
break;
}
}
}
/* cleanup the newnode list */
OBJ_RELEASE(item2);
}
/* if we still have entries on our hostfile list, then
* there were requested hosts that were not in our allocation.
* This is an error - report it to the user and return an error
*/
if (0 != opal_list_get_size(&newnodes)) {
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-hostfile.txt", "not-all-mapped-alloc",
true, hostfile);
while (NULL != (item1 = opal_list_remove_first(&newnodes))) {
OBJ_RELEASE(item1);
}
OBJ_DESTRUCT(&newnodes);
return ORTE_ERR_SILENT;
}
/* clear the rest of the nodes list */
while (NULL != (item1 = opal_list_remove_first(nodes))) {
OBJ_RELEASE(item1);
}
/* the nodes list has been cleared - rebuild it in order */
while (NULL != (item1 = opal_list_remove_first(&keep))) {
opal_list_append(nodes, item1);
}
cleanup:
OBJ_DESTRUCT(&newnodes);
return rc;
}
int orte_util_get_ordered_host_list(opal_list_t *nodes,
char *hostfile)
{
opal_list_t exclude;
opal_list_item_t *item, *itm, *item2, *item1;
char *cptr;
int num_empty, i, nodeidx, startempty=0;
bool want_all_empty=false;
orte_node_t *node_from_pool, *newnode;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_debug_output,
"%s hostfile: creating ordered list of hosts from hostfile %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hostfile));
OBJ_CONSTRUCT(&exclude, opal_list_t);
/* parse the hostfile and add the contents to the list, keeping duplicates */
if (ORTE_SUCCESS != (rc = hostfile_parse(hostfile, nodes, &exclude, true))) {
goto cleanup;
}
/* parse the nodes to process any relative node directives */
item2 = opal_list_get_first(nodes);
while (item2 != opal_list_get_end(nodes)) {
orte_node_t *node=(orte_node_t*)item2;
/* save the next location in case this one gets removed */
item1 = opal_list_get_next(item2);
if ('+' != node->name[0]) {
item2 = item1;
continue;
}
/* see if we specified empty nodes */
if ('e' == node->name[1] ||
'E' == node->name[1]) {
/* request for empty nodes - do they want
* all of them?
*/
if (NULL != (cptr = strchr(node->name, ':'))) {
/* the colon indicates a specific # are requested */
cptr++; /* step past : */
num_empty = strtol(cptr, NULL, 10);
} else {
/* want them all - set num_empty to max */
num_empty = INT_MAX;
want_all_empty = true;
}
/* insert empty nodes into newnodes list in place of the current item.
* since item1 is the next item, we insert in front of it
*/
if (!orte_hnp_is_allocated && 0 == startempty) {
startempty = 1;
}
for (i=startempty; 0 < num_empty && i < orte_node_pool->size; i++) {
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
continue;
}
if (0 == node_from_pool->slots_inuse) {
newnode = OBJ_NEW(orte_node_t);
newnode->name = strdup(node_from_pool->name);
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node->slots < node_from_pool->slots) {
newnode->slots_alloc = node->slots;
} else {
newnode->slots_alloc = node_from_pool->slots;
}
opal_list_insert_pos(nodes, item1, &newnode->super);
/* track number added */
--num_empty;
}
}
/* bookmark where we stopped in case they ask for more */
startempty = i;
/* did they get everything they wanted? */
if (!want_all_empty && 0 < num_empty) {
orte_show_help("help-hostfile.txt", "hostfile:not-enough-empty",
true, num_empty);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* since we have expanded the provided node, remove
* it from list
*/
opal_list_remove_item(nodes, item2);
OBJ_RELEASE(item2);
} else if ('n' == node->name[1] ||
'N' == node->name[1]) {
/* they want a specific relative node #, so
* look it up on global pool
*/
nodeidx = strtol(&node->name[2], NULL, 10);
/* if the HNP is not allocated, then we need to
* adjust the index as the node pool is offset
* by one
*/
if (!orte_hnp_is_allocated) {
nodeidx++;
}
/* see if that location is filled */
if (NULL == (node_from_pool = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, nodeidx))) {
/* this is an error */
orte_show_help("help-hostfile.txt", "hostfile:relative-node-not-found",
true, nodeidx, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* create the node object */
newnode = OBJ_NEW(orte_node_t);
newnode->name = strdup(node_from_pool->name);
/* if the slot count here is less than the
* total slots avail on this node, set it
* to the specified count - this allows people
* to subdivide an allocation
*/
if (node->slots < node_from_pool->slots) {
newnode->slots_alloc = node->slots;
} else {
newnode->slots_alloc = node_from_pool->slots;
}
/* insert it before item1 */
opal_list_insert_pos(nodes, item1, &newnode->super);
/* since we have expanded the provided node, remove
* it from list
*/
opal_list_remove_item(nodes, item2);
OBJ_RELEASE(item2);
} else {
/* invalid relative node syntax */
orte_show_help("help-hostfile.txt", "hostfile:invalid-relative-node-syntax",
true, node->name);
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* move to next */
item2 = item1;
}
/* remove from the list of nodes those that are in the exclude list */
while(NULL != (item = opal_list_remove_first(&exclude))) {
orte_node_t *exnode = (orte_node_t*)item;
/* check for matches on nodes */
for (itm = opal_list_get_first(nodes);
itm != opal_list_get_end(nodes);
itm = opal_list_get_next(itm)) {
orte_node_t *node=(orte_node_t*)itm;
if (0 == strcmp(exnode->name, node->name)) {
/* match - remove it */
opal_list_remove_item(nodes, itm);
OBJ_RELEASE(itm);
/* have to cycle through the entire list as we could
* have duplicates
*/
}
}
OBJ_RELEASE(item);
}
cleanup:
OBJ_DESTRUCT(&exclude);
return rc;
}