2013-01-31 15:19:39 +00:00
|
|
|
/*
|
|
|
|
|
* Copyright (c) 2012-2013 Los Alamos National Security, Inc. All rights reserved.
|
|
|
|
|
* $COPYRIGHT$
|
|
|
|
|
*
|
|
|
|
|
* Additional copyrights may follow
|
|
|
|
|
*
|
|
|
|
|
* $HEADER$
|
|
|
|
|
*
|
|
|
|
|
*/
|
|
|
|
|
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
#include "opal_config.h"
|
|
|
|
|
#include "opal/constants.h"
|
2013-01-31 15:19:39 +00:00
|
|
|
|
2013-09-06 19:03:19 +00:00
|
|
|
#include <time.h>
|
2013-01-31 15:19:39 +00:00
|
|
|
#include <string.h>
|
|
|
|
|
#include <sys/types.h>
|
|
|
|
|
#ifdef HAVE_LIMITS_H
|
|
|
|
|
#include <limits.h>
|
|
|
|
|
#endif
|
|
|
|
|
#include <stdio.h>
|
|
|
|
|
#ifdef HAVE_FCNTL_H
|
|
|
|
|
#include <fcntl.h>
|
|
|
|
|
#endif
|
|
|
|
|
#ifdef HAVE_UNISTD_H
|
|
|
|
|
#include <unistd.h>
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
#include "opal/class/opal_pointer_array.h"
|
|
|
|
|
#include "opal/util/argv.h"
|
2013-02-26 20:21:10 +00:00
|
|
|
#include "opal/util/output.h"
|
2013-02-01 14:21:06 +00:00
|
|
|
#include "opal_stdint.h"
|
2013-01-31 15:19:39 +00:00
|
|
|
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
#include "opal/mca/db/base/base.h"
|
2013-01-31 15:19:39 +00:00
|
|
|
#include "db_print.h"
|
|
|
|
|
|
|
|
|
|
static int init(void);
|
|
|
|
|
static void finalize(void);
|
|
|
|
|
static int add_log(const char *table,
|
|
|
|
|
const opal_value_t *kvs, int nkvs);
|
|
|
|
|
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
opal_db_base_module_t opal_db_print_module = {
|
2013-01-31 15:19:39 +00:00
|
|
|
init,
|
|
|
|
|
finalize,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
2013-07-22 22:57:16 +00:00
|
|
|
NULL,
|
2013-01-31 15:19:39 +00:00
|
|
|
add_log
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static opal_pointer_array_t tables;
|
|
|
|
|
static FILE *fpout=NULL;
|
|
|
|
|
|
|
|
|
|
static int init(void)
|
|
|
|
|
{
|
|
|
|
|
OBJ_CONSTRUCT(&tables, opal_pointer_array_t);
|
|
|
|
|
opal_pointer_array_init(&tables, 1, INT_MAX, 1);
|
|
|
|
|
|
|
|
|
|
if (0 == strcmp(mca_db_print_component.filename, "-")) {
|
|
|
|
|
fpout = stdout;
|
|
|
|
|
} else if (0 == strcmp(mca_db_print_component.filename, "+")) {
|
|
|
|
|
fpout = stderr;
|
|
|
|
|
} else if (NULL == (fpout = fopen(mca_db_print_component.filename, "w"))) {
|
|
|
|
|
opal_output(0, "ERROR: cannot open log file %s", mca_db_print_component.filename);
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
return OPAL_ERROR;
|
2013-01-31 15:19:39 +00:00
|
|
|
}
|
|
|
|
|
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
return OPAL_SUCCESS;
|
2013-01-31 15:19:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void finalize(void)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
char *ptr;
|
|
|
|
|
|
|
|
|
|
for (i=0; i < tables.size; i++) {
|
|
|
|
|
if (NULL != (ptr = (char*)opal_pointer_array_get_item(&tables, i))) {
|
|
|
|
|
free(ptr);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
OBJ_DESTRUCT(&tables);
|
|
|
|
|
if (NULL != fpout &&
|
|
|
|
|
stdout != fpout &&
|
|
|
|
|
stderr != fpout) {
|
|
|
|
|
fclose(fpout);
|
|
|
|
|
fpout = NULL;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int add_log(const char *table,
|
|
|
|
|
const opal_value_t *kvs, int nkvs)
|
|
|
|
|
{
|
|
|
|
|
char **cmdargs=NULL, *vstr;
|
|
|
|
|
time_t nowtime;
|
2013-02-01 14:21:06 +00:00
|
|
|
struct tm nowtm;
|
2013-01-31 15:19:39 +00:00
|
|
|
char tbuf[1024];
|
|
|
|
|
int i;
|
|
|
|
|
bool found;
|
|
|
|
|
|
2013-03-27 21:11:47 +00:00
|
|
|
opal_output_verbose(2, opal_db_base_framework.framework_output,
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
"Logging data for table %s", table);
|
2013-01-31 15:19:39 +00:00
|
|
|
|
|
|
|
|
found = false;
|
|
|
|
|
for (i=0; i < tables.size; i++) {
|
|
|
|
|
if (NULL == (vstr = (char*)opal_pointer_array_get_item(&tables, i))) {
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
if (0 == strcmp(vstr, table)) {
|
|
|
|
|
/* already handled this one */
|
|
|
|
|
found = true;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!found) {
|
|
|
|
|
/* record that we have it */
|
|
|
|
|
vstr = strdup(table);
|
|
|
|
|
opal_pointer_array_add(&tables, vstr);
|
|
|
|
|
/* create and print the column headers */
|
|
|
|
|
for (i=0; i < nkvs; i++) {
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, kvs[i].key);
|
|
|
|
|
}
|
|
|
|
|
vstr = opal_argv_join(cmdargs, '|');
|
|
|
|
|
fprintf(fpout, "%s\n", vstr);
|
|
|
|
|
free(vstr);
|
|
|
|
|
opal_argv_free(cmdargs);
|
|
|
|
|
cmdargs = NULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* cycle through the provided values and print them */
|
|
|
|
|
for (i=0; i < nkvs; i++) {
|
|
|
|
|
switch (kvs[i].type) {
|
|
|
|
|
case OPAL_STRING:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%s", kvs[i].data.string);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
2013-02-01 14:21:06 +00:00
|
|
|
case OPAL_SIZE:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIsize_t "", kvs[i].data.size);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_INT:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%d", kvs[i].data.integer);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_INT8:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIi8 "", kvs[i].data.int8);
|
2013-01-31 15:19:39 +00:00
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_INT16:
|
2013-02-01 14:21:06 +00:00
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIi16 "", kvs[i].data.int16);
|
2013-01-31 15:19:39 +00:00
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
2013-02-01 14:21:06 +00:00
|
|
|
case OPAL_INT32:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIi32 "", kvs[i].data.int32);
|
2013-01-31 15:19:39 +00:00
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_INT64:
|
2013-02-01 14:21:06 +00:00
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIi64 "", kvs[i].data.int64);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_UINT:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%u", kvs[i].data.uint);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_UINT8:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIu8 "", kvs[i].data.uint8);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_UINT16:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIu16 "", kvs[i].data.uint16);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_UINT32:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIu32 "", kvs[i].data.uint32);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_UINT64:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%" PRIu64 "", kvs[i].data.uint64);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_PID:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%lu", (unsigned long)kvs[i].data.pid);
|
2013-01-31 15:19:39 +00:00
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_FLOAT:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "%f", kvs[i].data.fval);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
case OPAL_TIMEVAL:
|
|
|
|
|
/* we only care about seconds */
|
|
|
|
|
nowtime = kvs[i].data.tv.tv_sec;
|
2013-02-01 14:21:06 +00:00
|
|
|
(void)localtime_r(&nowtime, &nowtm);
|
|
|
|
|
strftime(tbuf, sizeof(tbuf), "%Y-%m-%d %H:%M:%S", &nowtm);
|
|
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
snprintf(tbuf, sizeof(tbuf), "Unsupported type: %s",
|
|
|
|
|
opal_dss.lookup_data_type(kvs[i].type));
|
2013-01-31 15:19:39 +00:00
|
|
|
opal_argv_append_nosize(&cmdargs, tbuf);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* assemble the value string */
|
|
|
|
|
vstr = opal_argv_join(cmdargs, ',');
|
|
|
|
|
|
|
|
|
|
/* print it */
|
|
|
|
|
fprintf(fpout, "%s\n", vstr);
|
|
|
|
|
free(vstr);
|
|
|
|
|
|
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data.
A few changes were required to support this move:
1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution).
2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time.
3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering.
4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options.
No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework.
This commit was SVN r28112.
2013-02-26 17:50:04 +00:00
|
|
|
return OPAL_SUCCESS;
|
2013-01-31 15:19:39 +00:00
|
|
|
}
|