1
1
openmpi/orte/mca/rmgr/proxy/rmgr_proxy.c

392 строки
11 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2006 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.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include <errno.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#ifdef HAVE_STRING_H
#include <string.h>
#endif /* HAVE_STRING_H */
#include "orte/orte_constants.h"
#include "opal/util/output.h"
#include "opal/util/trace.h"
#include "orte/mca/rds/rds.h"
#include "orte/mca/ras/ras.h"
#include "orte/mca/rmaps/rmaps.h"
#include "orte/mca/pls/pls.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/smr/smr.h"
#include "orte/mca/rmgr/base/rmgr_private.h"
#include "orte/mca/rmgr/proxy/rmgr_proxy.h"
static int orte_rmgr_proxy_setup_job(
orte_app_context_t** app_context,
orte_std_cntr_t num_context,
orte_jobid_t* jobid);
static int orte_rmgr_proxy_spawn_job(
orte_app_context_t** app_context,
orte_std_cntr_t num_context,
orte_jobid_t* jobid,
orte_std_cntr_t num_connect,
orte_process_name_t *connect,
orte_rmgr_cb_fn_t cbfn,
orte_proc_state_t cb_conditions);
orte_rmgr_base_module_t orte_rmgr_proxy_module = {
NULL, /* don't need special init */
orte_rmgr_proxy_setup_job,
orte_rmgr_proxy_spawn_job,
orte_rmgr_base_connect,
orte_rmgr_base_disconnect,
NULL, /* finalize */
/** SUPPORT FUNCTIONS ***/
orte_rmgr_base_get_app_context,
orte_rmgr_base_put_app_context,
orte_rmgr_base_check_context_cwd,
orte_rmgr_base_check_context_app,
orte_rmgr_base_set_vpid_range,
orte_rmgr_base_get_vpid_range
};
/*
* Setup the job segment and initialize the application context. Could
* do this in the proxy - but allowing the HNP to do this moves the registry
* and name service actions to the HNP for efficiency.
*/
static int orte_rmgr_proxy_setup_job(
orte_app_context_t** app_context,
orte_std_cntr_t num_context,
orte_jobid_t* jobid)
{
orte_buffer_t cmd;
orte_buffer_t rsp;
orte_std_cntr_t count;
orte_rmgr_cmd_t command=ORTE_RMGR_SETUP_JOB_CMD;
int rc;
OPAL_TRACE(1);
/* construct command */
OBJ_CONSTRUCT(&cmd, orte_buffer_t);
/* pack the command */
if (ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &command, 1, ORTE_RMGR_CMD))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
}
/* pack the number of app_contexts */
if(ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, &num_context, 1, ORTE_STD_CNTR))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
}
/* and pack them */
if(ORTE_SUCCESS != (rc = orte_dss.pack(&cmd, app_context, num_context, ORTE_APP_CONTEXT))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
}
/* send the command */
if(0 > (rc = orte_rml.send_buffer(ORTE_RML_NAME_SEED, &cmd, ORTE_RML_TAG_RMGR, 0))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&cmd);
return rc;
}
OBJ_DESTRUCT(&cmd);
/* wait for response */
OBJ_CONSTRUCT(&rsp, orte_buffer_t);
if(0 > (rc = orte_rml.recv_buffer(ORTE_RML_NAME_SEED, &rsp, ORTE_RML_TAG_RMGR))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&rsp);
return rc;
}
/* get the returned command */
count = 1;
if (ORTE_SUCCESS != (rc = orte_dss.unpack(&rsp, &command, &count, ORTE_RMGR_CMD))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&rsp);
return rc;
}
/* and check it to ensure valid comm */
if (ORTE_RMGR_SETUP_JOB_CMD != command) {
Add ability to trap and propagate SIGUSR1/2 to remote processes. There are a number of small changes that hit a bunch of files: 1. Changed the RMGR and PLS APIs to add "signal_job" and "signal_proc" entry points. Only the "signal_job" entries are implemented - none of the components have implementations for "signal_proc" at this time. Thus, you can signal all of the procs in a job, but cannot currently signal only one specific proc. 2. Implemented those new API functions in all components except xgrid (Brian will do so very soon). Only the rsh/ssh and fork modules have been tested, however, and only under OS-X. 3. Added signal traps and callback functions for SIGUSR1/2 to orterun/mpirun that catch those signals and call the appropriate commands to propagate them out to all processes in the job. 4. Added a new test directory under the orte branch to (eventually) hold unit and system level tests for just the run-time. Since our test branch of the repository is under restricted access, people working on the RTE were continually developing their own system-level tests - thus making it hard to help diagnose problems. I have moved the more commonly-used functions here, and added one specifically for testing the SIGUSR1/2 functionality. I will be contacting people directly to seek help with testing the changes on more environments. Other than compile issues, you should see absolutely no change in behavior on any of your systems - this additional functionality is transparent to anyone who does not issue a SIGUSR1/2 to mpirun. Ralph This commit was SVN r10258.
2006-06-08 22:27:17 +04:00
OBJ_DESTRUCT(&rsp);
ORTE_ERROR_LOG(ORTE_ERR_COMM_FAILURE);
return ORTE_ERR_COMM_FAILURE;
Add ability to trap and propagate SIGUSR1/2 to remote processes. There are a number of small changes that hit a bunch of files: 1. Changed the RMGR and PLS APIs to add "signal_job" and "signal_proc" entry points. Only the "signal_job" entries are implemented - none of the components have implementations for "signal_proc" at this time. Thus, you can signal all of the procs in a job, but cannot currently signal only one specific proc. 2. Implemented those new API functions in all components except xgrid (Brian will do so very soon). Only the rsh/ssh and fork modules have been tested, however, and only under OS-X. 3. Added signal traps and callback functions for SIGUSR1/2 to orterun/mpirun that catch those signals and call the appropriate commands to propagate them out to all processes in the job. 4. Added a new test directory under the orte branch to (eventually) hold unit and system level tests for just the run-time. Since our test branch of the repository is under restricted access, people working on the RTE were continually developing their own system-level tests - thus making it hard to help diagnose problems. I have moved the more commonly-used functions here, and added one specifically for testing the SIGUSR1/2 functionality. I will be contacting people directly to seek help with testing the changes on more environments. Other than compile issues, you should see absolutely no change in behavior on any of your systems - this additional functionality is transparent to anyone who does not issue a SIGUSR1/2 to mpirun. Ralph This commit was SVN r10258.
2006-06-08 22:27:17 +04:00
}
/* get the jobid */
count = 1;
if(ORTE_SUCCESS != (rc = orte_dss.unpack(&rsp, jobid, &count, ORTE_JOBID))) {
ORTE_ERROR_LOG(rc);
}
OBJ_DESTRUCT(&rsp);
return rc;
}
static void orte_rmgr_proxy_wireup_stdin(orte_jobid_t jobid)
{
int rc;
orte_process_name_t* name;
OPAL_TRACE(1);
if (ORTE_SUCCESS != (rc = orte_ns.create_process_name(&name, 0, jobid, 0))) {
ORTE_ERROR_LOG(rc);
return;
}
if (ORTE_SUCCESS != (rc = orte_iof.iof_push(name, ORTE_NS_CMP_JOBID, ORTE_IOF_STDIN, 0))) {
ORTE_ERROR_LOG(rc);
}
}
static void orte_rmgr_proxy_callback(orte_gpr_notify_data_t *data, void *cbdata)
{
orte_rmgr_cb_fn_t cbfunc;
union {
orte_rmgr_cb_fn_t func;
void * ptr;
} cbfunc_union;
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
orte_gpr_value_t **values, *value;
orte_gpr_keyval_t** keyvals;
orte_jobid_t jobid;
orte_std_cntr_t i, j, k;
int rc;
OPAL_TRACE(1);
/* ISO C forbids conversion of object pointer to function
pointer. So we do this, which is the same thing, but without
the warning from GCC */
cbfunc_union.ptr = cbdata;
cbfunc = cbfunc_union.func;
/* we made sure in the subscriptions that at least one
* value is always returned
* get the jobid from the segment name in the first value
*/
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
values = (orte_gpr_value_t**)(data->values)->addr;
if (ORTE_SUCCESS != (rc =
orte_schema.extract_jobid_from_segment_name(&jobid,
values[0]->segment))) {
ORTE_ERROR_LOG(rc);
return;
}
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
for(i = 0, k=0; k < data->cnt &&
i < (data->values)->size; i++) {
if (NULL != values[i]) {
k++;
value = values[i];
/* determine the state change */
keyvals = value->keyvals;
for(j=0; j<value->cnt; j++) {
orte_gpr_keyval_t* keyval = keyvals[j];
if(strcmp(keyval->key, ORTE_PROC_NUM_AT_INIT) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_INIT);
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_LAUNCHED) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_LAUNCHED);
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_RUNNING) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_RUNNING);
continue;
}
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
if(strcmp(keyval->key, ORTE_PROC_NUM_AT_STG1) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_AT_STG1);
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_AT_STG2) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_AT_STG2);
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_AT_STG3) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_AT_STG3);
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_FINALIZED) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_FINALIZED);
continue;
}
if(strcmp(keyval->key, ORTE_PROC_NUM_TERMINATED) == 0) {
(*cbfunc)(jobid,ORTE_PROC_STATE_TERMINATED);
continue;
}
}
}
}
}
/**
* define a callback point for completing the wireup of the stdin for io forwarding
*/
static void orte_rmgr_proxy_wireup_callback(orte_gpr_notify_data_t *data, void *cbdata)
{
orte_gpr_value_t **values;
orte_jobid_t jobid;
int rc;
OPAL_TRACE(1);
/* we made sure in the subscriptions that at least one
* value is always returned
* get the jobid from the segment name in the first value
*/
values = (orte_gpr_value_t**)(data->values)->addr;
if (ORTE_SUCCESS != (rc = orte_schema.extract_jobid_from_segment_name(&jobid, values[0]->segment))) {
ORTE_ERROR_LOG(rc);
return;
}
orte_rmgr_proxy_wireup_stdin(jobid);
}
/*
* Shortcut for the multiple steps involved in spawning a new job.
*/
static int orte_rmgr_proxy_spawn_job(
orte_app_context_t** app_context,
orte_std_cntr_t num_context,
orte_jobid_t* jobid,
orte_std_cntr_t num_connect,
orte_process_name_t *connect,
orte_rmgr_cb_fn_t cbfunc,
orte_proc_state_t cb_conditions)
{
int rc;
orte_process_name_t name = {0, ORTE_JOBID_INVALID, 0};
OPAL_TRACE(1);
/*
* Perform resource discovery.
*/
if (ORTE_SUCCESS != (rc = orte_rds.query())) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* Setup job and allocate resources
*/
if (ORTE_SUCCESS !=
(rc = orte_rmgr_proxy_setup_job(app_context,num_context,jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_ras.allocate_job(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_rmaps.map_job(*jobid, NULL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* setup I/O forwarding
*/
name.jobid = *jobid;
if (ORTE_SUCCESS != (rc = orte_iof.iof_pull(&name, ORTE_NS_CMP_JOBID, ORTE_IOF_STDOUT, 1))) {
ORTE_ERROR_LOG(rc);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_iof.iof_pull(&name, ORTE_NS_CMP_JOBID, ORTE_IOF_STDERR, 2))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* setup the launch system's stage gate counters and subscriptions */
if (ORTE_SUCCESS != (rc = orte_rmgr_base_proc_stage_gate_init(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/** setup the subscription so we can complete the wireup when all processes reach LAUNCHED */
rc = orte_smr.job_stage_gate_subscribe(*jobid, orte_rmgr_proxy_wireup_callback, NULL, ORTE_PROC_STATE_LAUNCHED);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* Define the ERRMGR's callbacks as required
*/
if (ORTE_SUCCESS != (rc = orte_errmgr.register_job(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/*
* setup callback
*/
if(NULL != cbfunc) {
union {
orte_rmgr_cb_fn_t func;
void * ptr;
} cbfunc_union;
void *cbdata;
/* ISO C forbids conversion of object pointer to function
pointer. So we do this, which is the same thing, but without
the warning from GCC */
cbfunc_union.func = cbfunc;
cbdata = cbfunc_union.ptr;
rc = orte_smr.job_stage_gate_subscribe(*jobid, orte_rmgr_proxy_callback, cbdata, cb_conditions);
if(ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
/*
* launch the job
*/
if (ORTE_SUCCESS != (rc = orte_pls.launch_job(*jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
return ORTE_SUCCESS;
}