1
1
openmpi/orte/mca/iof/tool/iof_tool.c
Ralph Castain 6509f60929 Complete the memprobe support. This provides a new scaling tool called "mpi_memprobe" that samples the memory footprint of the local daemon and the client procs, and then reports the results. The output contains the footprint of the daemon on each node, plus the average footprint of the client procs on that node.
Samples are taken after MPI_Init, and then again after MPI_Barrier. This allows the user to see memory consumption caused by add_procs, as well as any modex contribution from forming connections if pmix_base_async_modex is given.

Using the probe simply involves executing it via mpirun, with however many copies you want per node. Example:

$ mpirun -npernode 2 ./mpi_memprobe
Sampling memory usage after MPI_Init
Data for node rhc001
	Daemon: 12.483398
	Client: 6.514648

Data for node rhc002
	Daemon: 11.865234
	Client: 4.643555

Sampling memory usage after MPI_Barrier
Data for node rhc001
	Daemon: 12.520508
	Client: 6.576660

Data for node rhc002
	Daemon: 11.879883
	Client: 4.703125

Note that the client value on node rhc001 is larger - this is where rank=0 is housed, and apparently it gets a larger footprint for some reason.

Signed-off-by: Ralph Castain <rhc@open-mpi.org>
2017-01-05 10:32:17 -08:00

305 строки
9.4 KiB
C

/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2013 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <errno.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#include <string.h>
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/util/name_fns.h"
#include "orte/runtime/orte_globals.h"
#include "orte/runtime/orte_wait.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/iof/base/base.h"
#include "iof_tool.h"
static int init(void);
static int tool_push(const orte_process_name_t* dst_name, orte_iof_tag_t src_tag, int fd);
static int tool_pull(const orte_process_name_t* src_name,
orte_iof_tag_t src_tag,
int fd);
static int tool_close(const orte_process_name_t* peer,
orte_iof_tag_t source_tag);
static int tool_output(const orte_process_name_t* peer,
orte_iof_tag_t source_tag,
const char *msg);
static int finalize(void);
static int tool_ft_event(int state);
orte_iof_base_module_t orte_iof_tool_module = {
.init = init,
.push = tool_push,
.pull = tool_pull,
.close = tool_close,
.output = tool_output,
.finalize = finalize,
.ft_event = tool_ft_event
};
static int init(void)
{
/* post a non-blocking RML receive to get messages
from the HNP IOF component */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_IOF_PROXY,
ORTE_RML_PERSISTENT,
orte_iof_tool_recv,
NULL);
mca_iof_tool_component.closed = false;
return ORTE_SUCCESS;
}
/**
* Push data from the specified file descriptor
* to the indicated SINK set of peers.
*/
static int tool_push(const orte_process_name_t* dst_name, orte_iof_tag_t src_tag, int fd)
{
/* at this time, we do not allow tools to push data into the
* stdin of a job. This is due to potential confusion over which
* stdin is being read/used, and the impossibility of resolving
* potential interleaving of the data
*/
return ORTE_ERR_NOT_SUPPORTED;
}
/*
* Callback when non-blocking RML send completes.
*/
static void send_cb(int status, orte_process_name_t *peer,
opal_buffer_t *buf, orte_rml_tag_t tag,
void *cbdata)
{
/* nothing to do here - just release buffer and return */
OBJ_RELEASE(buf);
}
/**
* Pull data from the specified set of SOURCE peers and
* dump to the indicated file descriptor.
*/
static int tool_pull(const orte_process_name_t* src_name,
orte_iof_tag_t src_tag,
int fd)
{
/* if we are a tool, then we need to request the HNP to please
* forward the data from the specified process to us. Note that
* the HNP will return an error if the specified stream of any
* intended recipient is not open. By default, stdout/err/diag
* are all left open. However, the user can also direct us to
* close any or all of those streams, so the success of this call
* will depend upon how the user executed the application
*/
opal_buffer_t *buf;
orte_iof_tag_t tag;
orte_process_name_t hnp;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_iof_base_framework.framework_output,
"%s pulling output for proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(src_name)));
buf = OBJ_NEW(opal_buffer_t);
/* setup the tag to pull from HNP */
tag = src_tag | ORTE_IOF_PULL;
/* pack the tag - we do this first so that flow control messages can
* consist solely of the tag
*/
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &tag, 1, ORTE_IOF_TAG))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* pack the name of the source */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, src_name, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* pack our name as the sink */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, ORTE_PROC_MY_NAME, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* send the buffer to the correct HNP */
ORTE_HNP_NAME_FROM_JOB(&hnp, src_name->jobid);
orte_rml.send_buffer_nb(orte_mgmt_conduit,
&hnp, buf, ORTE_RML_TAG_IOF_HNP,
send_cb, NULL);
return ORTE_SUCCESS;
}
static int tool_close(const orte_process_name_t* src_name,
orte_iof_tag_t src_tag)
{
/* if we are a tool, then we need to request the HNP to stop
* forwarding data from this process/stream
*/
opal_buffer_t *buf;
orte_iof_tag_t tag;
orte_process_name_t hnp;
int rc;
OPAL_OUTPUT_VERBOSE((1, orte_iof_base_framework.framework_output,
"%s closing output for proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(src_name)));
buf = OBJ_NEW(opal_buffer_t);
/* setup the tag to stop the copy */
tag = src_tag | ORTE_IOF_CLOSE;
/* pack the tag - we do this first so that flow control messages can
* consist solely of the tag
*/
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &tag, 1, ORTE_IOF_TAG))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* pack the name of the source */
if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, src_name, 1, ORTE_NAME))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buf);
return rc;
}
/* flag that the close is incomplete */
mca_iof_tool_component.closed = false;
/* send the buffer to the correct HNP */
ORTE_HNP_NAME_FROM_JOB(&hnp, src_name->jobid);
orte_rml.send_buffer_nb(orte_mgmt_conduit,
&hnp, buf, ORTE_RML_TAG_IOF_HNP,
send_cb, NULL);
return ORTE_SUCCESS;
}
static int finalize(void)
{
opal_list_item_t* item;
orte_iof_write_output_t *output;
orte_iof_write_event_t *wev;
int num_written;
bool dump;
/* check if anything is still trying to be written out */
wev = orte_iof_base.iof_write_stdout->wev;
if (!opal_list_is_empty(&wev->outputs)) {
dump = false;
/* make one last attempt to write this out */
while (NULL != (item = opal_list_remove_first(&wev->outputs))) {
output = (orte_iof_write_output_t*)item;
if (!dump) {
num_written = write(wev->fd, output->data, output->numbytes);
if (num_written < output->numbytes) {
/* don't retry - just cleanout the list and dump it */
dump = true;
}
}
OBJ_RELEASE(output);
}
}
OBJ_RELEASE(orte_iof_base.iof_write_stdout);
if (!orte_xml_output) {
/* we only opened stderr channel if we are NOT doing xml output */
wev = orte_iof_base.iof_write_stderr->wev;
if (!opal_list_is_empty(&wev->outputs)) {
dump = false;
/* make one last attempt to write this out */
while (NULL != (item = opal_list_remove_first(&wev->outputs))) {
output = (orte_iof_write_output_t*)item;
if (!dump) {
num_written = write(wev->fd, output->data, output->numbytes);
if (num_written < output->numbytes) {
/* don't retry - just cleanout the list and dump it */
dump = true;
}
}
OBJ_RELEASE(output);
}
}
OBJ_RELEASE(orte_iof_base.iof_write_stderr);
}
/* Cancel the RML receive */
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_IOF_PROXY);
return ORTE_SUCCESS;
}
static int tool_output(const orte_process_name_t* peer,
orte_iof_tag_t source_tag,
const char *msg)
{
/* output this to our local output */
if (ORTE_IOF_STDOUT & source_tag || orte_xml_output) {
orte_iof_base_write_output(peer, source_tag, (const unsigned char*)msg, strlen(msg), orte_iof_base.iof_write_stdout->wev);
} else {
orte_iof_base_write_output(peer, source_tag, (const unsigned char*)msg, strlen(msg), orte_iof_base.iof_write_stderr->wev);
}
return ORTE_SUCCESS;
}
/*
* FT event
*/
static int tool_ft_event(int state)
{
return ORTE_ERR_NOT_IMPLEMENTED;
}