Skip to content

Fix the tree-spawn-with-rollup #4628

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Dec 18, 2017
Merged

Fix the tree-spawn-with-rollup #4628

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 0 additions & 8 deletions orte/mca/plm/base/plm_base_launch_support.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1383,14 +1383,6 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
if (ORTE_ERR_UNPACK_READ_PAST_END_OF_BUFFER != rc) {
ORTE_ERROR_LOG(rc);
ORTE_ACTIVATE_JOB_STATE(jdatorted, ORTE_JOB_STATE_FAILED_TO_START);
} else if (NULL != orte_tree_launch_cmd) {
/* if a tree-launch is underway, send the cmd back */
relay = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(relay, orte_tree_launch_cmd);
orte_rml.send_buffer_nb(orte_mgmt_conduit,
sender, relay,
ORTE_RML_TAG_DAEMON,
orte_rml_send_callback, NULL);
}
}

Expand Down
40 changes: 31 additions & 9 deletions orte/mca/plm/rsh/plm_rsh_module.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,6 +88,7 @@
#include "orte/mca/ess/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/grpcomm/base/base.h"
#include "orte/mca/oob/base/base.h"
#include "orte/mca/rmaps/rmaps.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/rml/base/rml_contact.h"
Expand Down Expand Up @@ -605,7 +606,6 @@ static int setup_launch(int *argcptr, char ***argvptr,
(mca_plm_rsh_component.using_qrsh && mca_plm_rsh_component.daemonize_qrsh)) &&
((!mca_plm_rsh_component.using_llspawn) ||
(mca_plm_rsh_component.using_llspawn && mca_plm_rsh_component.daemonize_llspawn))) {
opal_argv_append(&argc, &argv, "--daemonize");
}

/*
Expand All @@ -617,9 +617,20 @@ static int setup_launch(int *argcptr, char ***argvptr,
proc_vpid_index);

/* ensure that only the ssh plm is selected on the remote daemon */
opal_argv_append_nosize(&argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&argv, "plm");
opal_argv_append_nosize(&argv, "rsh");
opal_argv_append(&argc, &argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append(&argc, &argv, "plm");
opal_argv_append(&argc, &argv, "rsh");

/* if we are tree-spawning, tell our child daemons the
* uri of their parent (me) */
if (!mca_plm_rsh_component.no_tree_spawn) {
opal_argv_append(&argc, &argv, "--tree-spawn");
orte_oob_base_get_addr(&param);
opal_argv_append(&argc, &argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append(&argc, &argv, "orte_parent_uri");
opal_argv_append(&argc, &argv, param);
free(param);
}

/* unless told otherwise... */
if (mca_plm_rsh_component.pass_environ_mca_params) {
Expand Down Expand Up @@ -795,11 +806,22 @@ static int remote_spawn(opal_buffer_t *launch)
/* if we hit any errors, tell the HNP it was us */
target.vpid = ORTE_PROC_MY_NAME->vpid;

/* extract the prefix from the launch buffer */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(launch, &prefix, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
if (NULL != launch) {
/* extract the prefix from the launch buffer */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(launch, &prefix, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
} else {
/* check to see if enable-orterun-prefix-by-default was given - if
* this is being done by a singleton, then orterun will not be there
* to put the prefix in the app. So make sure we check to find it */
if ((bool)ORTE_WANT_ORTERUN_PREFIX_BY_DEFAULT) {
prefix = strdup(opal_install_dirs.prefix);
} else {
prefix = NULL;
}
}

/* get the updated routing list */
Expand Down
46 changes: 26 additions & 20 deletions orte/orted/orted_main.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,7 +120,7 @@ static void rollup(int status, orte_process_name_t* sender,
static opal_buffer_t *bucket, *mybucket = NULL;
static int ncollected = 0;

static char *orte_parent_uri;
static char *orte_parent_uri = NULL;

static struct {
bool debug;
Expand Down Expand Up @@ -187,6 +187,10 @@ opal_cmd_line_init_t orte_cmd_line_opts[] = {
&orted_globals.set_sid, OPAL_CMD_LINE_TYPE_BOOL,
"Direct the orted to separate from the current session"},

{ NULL, '\0', "tree-spawn", "tree-spawn", 0,
&orted_globals.tree_spawn, OPAL_CMD_LINE_TYPE_BOOL,
"Tree-based spawn in progress" },

{ "tmpdir_base", '\0', NULL, "tmpdir", 1,
NULL, OPAL_CMD_LINE_TYPE_STRING,
"Set the root for the session directory tree" },
Expand Down Expand Up @@ -667,22 +671,19 @@ int orte_daemon(int argc, char *argv[])
MCA_BASE_VAR_SCOPE_CONSTANT,
&orte_parent_uri);
if (NULL != orte_parent_uri) {
orte_process_name_t parent;
opal_value_t val;

/* set the contact info into our local database */
ret = orte_rml_base_parse_uris(orte_parent_uri, &parent, NULL);
ret = orte_rml_base_parse_uris(orte_parent_uri, ORTE_PROC_MY_PARENT, NULL);
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
free (orte_parent_uri);
orte_parent_uri = NULL;
goto DONE;
}
OBJ_CONSTRUCT(&val, opal_value_t);
val.key = OPAL_PMIX_PROC_URI;
val.type = OPAL_STRING;
val.data.string = orte_parent_uri;
if (OPAL_SUCCESS != (ret = opal_pmix.store_local(&parent, &val))) {
if (OPAL_SUCCESS != (ret = opal_pmix.store_local(ORTE_PROC_MY_PARENT, &val))) {
ORTE_ERROR_LOG(ret);
OBJ_DESTRUCT(&val);
goto DONE;
Expand All @@ -691,21 +692,22 @@ int orte_daemon(int argc, char *argv[])
val.data.string = NULL;
OBJ_DESTRUCT(&val);

/* don't need this value anymore */
free(orte_parent_uri);
orte_parent_uri = NULL;

/* tell the routed module that we have a path
* back to the HNP
*/
if (ORTE_SUCCESS != (ret = orte_routed.update_route(NULL, ORTE_PROC_MY_HNP, &parent))) {
if (ORTE_SUCCESS != (ret = orte_routed.update_route(NULL, ORTE_PROC_MY_HNP, ORTE_PROC_MY_PARENT))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* and a path to our parent */
if (ORTE_SUCCESS != (ret = orte_routed.update_route(NULL, ORTE_PROC_MY_PARENT, ORTE_PROC_MY_PARENT))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
/* set the lifeline to point to our parent so that we
* can handle the situation if that lifeline goes away
*/
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(NULL, &parent))) {
if (ORTE_SUCCESS != (ret = orte_routed.set_lifeline(NULL, ORTE_PROC_MY_PARENT))) {
ORTE_ERROR_LOG(ret);
goto DONE;
}
Expand All @@ -717,12 +719,15 @@ int orte_daemon(int argc, char *argv[])
*/
if (!ORTE_PROC_IS_HNP) {
orte_process_name_t target;
target.jobid = ORTE_PROC_MY_NAME->jobid;

if (orte_fwd_mpirun_port || orte_static_ports) {
/* setup the rollup callback */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK,
ORTE_RML_PERSISTENT, rollup, NULL);
/* setup the rollup callback */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK,
ORTE_RML_PERSISTENT, rollup, NULL);

/* define the target jobid */
target.jobid = ORTE_PROC_MY_NAME->jobid;
if (orte_fwd_mpirun_port || orte_static_ports || NULL != orte_parent_uri) {
/* we start by sending to ourselves */
target.vpid = ORTE_PROC_MY_NAME->vpid;
/* since we will be waiting for any children to send us
* their rollup info before sending to our parent, save
Expand Down Expand Up @@ -789,7 +794,6 @@ int orte_daemon(int argc, char *argv[])
}
OPAL_LIST_RELEASE(modex);
} else {
opal_output(0, "VAL KEY: %s", (NULL == val->key) ? "NULL" : val->key);
/* single value */
flag = 1;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &flag, 1, OPAL_INT32))) {
Expand Down Expand Up @@ -965,6 +969,8 @@ int orte_daemon(int argc, char *argv[])
i += 2;
}
}
/* now launch any child daemons of ours */
orte_plm.remote_spawn(orte_tree_launch_cmd);
}

if (orte_debug_daemons_flag) {
Expand Down Expand Up @@ -1053,8 +1059,6 @@ static void rollup(int status, orte_process_name_t* sender,
int32_t i, flag, cnt;
opal_value_t *kv;

/* xfer the contents of the rollup to our bucket */
opal_dss.copy_payload(bucket, buffer);
ncollected++;

/* if the sender is ourselves, then we save that buffer
Expand All @@ -1064,6 +1068,8 @@ static void rollup(int status, orte_process_name_t* sender,
mybucket = OBJ_NEW(opal_buffer_t);
opal_dss.copy_payload(mybucket, buffer);
} else {
/* xfer the contents of the rollup to our bucket */
opal_dss.copy_payload(bucket, buffer);
/* the first entry in the bucket will be from our
* direct child - harvest it for connection info */
cnt = 1;
Expand Down