/**
* Main control point for the cleanup planner ros node
*/
int main(int argc, char** argv)
{
ros::init(argc, argv, "cleanup_planner");
ros::NodeHandle n;
CleanupPlannerNode cleanup_node(n);
cleanup_node.spin();
return 0;
}
/**
* Method called whenever a given peer disconnects.
*
* @param cls closure
* @param peer peer identity this notification is about
*/
static void
core_disconnect_handler (void *cls,
const struct GNUNET_PeerIdentity * peer)
{
struct Node *n;
if (GNUNET_YES == is_me(peer))
return;
GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Disconnected from peer %s\n"),
GNUNET_i2s (peer));
if (NULL != (n = GNUNET_CONTAINER_multipeermap_get (nodes_requested, peer)))
cleanup_node (nodes_requested, peer, n);
if (NULL != (n = GNUNET_CONTAINER_multipeermap_get (nodes_active, peer)))
cleanup_node (nodes_active, peer, n);
if (NULL != (n = GNUNET_CONTAINER_multipeermap_get (nodes_inactive, peer)))
cleanup_node (nodes_inactive, peer, n);
}
void orte_state_base_track_procs(int fd, short argc, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
orte_job_t *jdata;
orte_proc_t *pdata;
int i;
opal_output_verbose(5, orte_state_base_framework.framework_output,
"%s state:base:track_procs called for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state));
/* get the job object for this proc */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto cleanup;
}
pdata = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
if (ORTE_PROC_STATE_RUNNING == state) {
/* update the proc state */
pdata->state = state;
jdata->num_launched++;
if (jdata->num_launched == jdata->num_procs) {
if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_READY_FOR_DEBUGGERS);
} else {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_RUNNING);
}
}
} else if (ORTE_PROC_STATE_REGISTERED == state) {
/* update the proc state */
pdata->state = state;
jdata->num_reported++;
if (jdata->num_reported == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_REGISTERED);
}
} else if (ORTE_PROC_STATE_IOF_COMPLETE == state) {
/* update the proc state */
pdata->state = state;
/* Release only the stdin IOF file descriptor for this child, if one
* was defined. File descriptors for the other IOF channels - stdout,
* stderr, and stddiag - were released when their associated pipes
* were cleared and closed due to termination of the process
*/
if (NULL != orte_iof.close) {
orte_iof.close(proc, ORTE_IOF_STDIN);
}
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_IOF_COMPLETE);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_WAITPID)) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_WAITPID_FIRED == state) {
/* update the proc state */
pdata->state = state;
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_WAITPID);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_IOF_COMPLETE)) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_TERMINATED == state) {
/* update the proc state */
ORTE_FLAG_UNSET(pdata, ORTE_PROC_FLAG_ALIVE);
pdata->state = state;
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_LOCAL)) {
/* Clean up the session directory as if we were the process
* itself. This covers the case where the process died abnormally
* and didn't cleanup its own session directory.
*/
orte_session_dir_finalize(proc);
}
/* if we are trying to terminate and our routes are
* gone, then terminate ourselves IF no local procs
* remain (might be some from another job)
*/
if (orte_orteds_term_ordered &&
0 == orte_routed.num_routes()) {
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (pdata = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i)) &&
ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_ALIVE)) {
/* at least one is still alive */
goto cleanup;
}
}
/* call our appropriate exit procedure */
OPAL_OUTPUT_VERBOSE((5, orte_state_base_framework.framework_output,
"%s state:base all routes and children gone - exiting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
goto cleanup;
}
/* return the allocated slot for reuse */
cleanup_node(pdata);
/* track job status */
jdata->num_terminated++;
if (jdata->num_terminated == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
}
cleanup:
OBJ_RELEASE(caddy);
}
void orte_state_base_track_procs(int fd, short argc, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
orte_job_t *jdata;
orte_proc_t *pdata;
opal_output_verbose(5, orte_state_base_framework.framework_output,
"%s state:base:track_procs called for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state));
/* get the job object for this proc */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto cleanup;
}
pdata = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
if (ORTE_PROC_STATE_RUNNING == state) {
/* update the proc state */
pdata->state = state;
jdata->num_launched++;
if (jdata->num_launched == jdata->num_procs) {
if (jdata->controls & ORTE_JOB_CONTROL_DEBUGGER_DAEMON) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_READY_FOR_DEBUGGERS);
} else {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_RUNNING);
}
}
} else if (ORTE_PROC_STATE_REGISTERED == state) {
/* update the proc state */
pdata->state = state;
jdata->num_reported++;
if (jdata->num_reported == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_REGISTERED);
}
} else if (ORTE_PROC_STATE_IOF_COMPLETE == state) {
/* update the proc state */
pdata->state = state;
/* Release only the stdin IOF file descriptor for this child, if one
* was defined. File descriptors for the other IOF channels - stdout,
* stderr, and stddiag - were released when their associated pipes
* were cleared and closed due to termination of the process
*/
if (NULL != orte_iof.close) {
orte_iof.close(proc, ORTE_IOF_STDIN);
}
pdata->iof_complete = true;
if (pdata->waitpid_recvd) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_WAITPID_FIRED == state) {
/* update the proc state */
pdata->state = state;
pdata->waitpid_recvd = true;
if (pdata->iof_complete) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_TERMINATED == state) {
/* update the proc state */
pdata->alive = false;
pdata->state = state;
if (pdata->local_proc) {
/* Clean up the session directory as if we were the process
* itself. This covers the case where the process died abnormally
* and didn't cleanup its own session directory.
*/
orte_session_dir_finalize(proc);
}
/* return the allocated slot for reuse */
cleanup_node(pdata);
/* track job status */
jdata->num_terminated++;
if (jdata->num_terminated == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
}
}
cleanup:
OBJ_RELEASE(caddy);
}
void orte_state_base_track_procs(int fd, short argc, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
orte_job_t *jdata;
orte_proc_t *pdata;
int i;
char *rtmod;
orte_process_name_t parent, target, *npptr;
opal_output_verbose(5, orte_state_base_framework.framework_output,
"%s state:base:track_procs called for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state));
/* get our "lifeline" routed module */
rtmod = orte_rml.get_routed(orte_mgmt_conduit);
/* get the job object for this proc */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto cleanup;
}
pdata = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
if (ORTE_PROC_STATE_RUNNING == state) {
/* update the proc state */
if (pdata->state < ORTE_PROC_STATE_TERMINATED) {
pdata->state = state;
}
jdata->num_launched++;
if (jdata->num_launched == jdata->num_procs) {
if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_READY_FOR_DEBUGGERS);
} else {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_RUNNING);
}
}
} else if (ORTE_PROC_STATE_REGISTERED == state) {
/* update the proc state */
if (pdata->state < ORTE_PROC_STATE_TERMINATED) {
pdata->state = state;
}
jdata->num_reported++;
if (jdata->num_reported == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_REGISTERED);
}
} else if (ORTE_PROC_STATE_IOF_COMPLETE == state) {
/* update the proc state */
if (pdata->state < ORTE_PROC_STATE_TERMINATED) {
pdata->state = state;
}
/* Release only the stdin IOF file descriptor for this child, if one
* was defined. File descriptors for the other IOF channels - stdout,
* stderr, and stddiag - were released when their associated pipes
* were cleared and closed due to termination of the process
*/
if (NULL != orte_iof.close) {
orte_iof.close(proc, ORTE_IOF_STDIN);
}
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_IOF_COMPLETE);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_WAITPID)) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_WAITPID_FIRED == state) {
/* update the proc state */
if (pdata->state < ORTE_PROC_STATE_TERMINATED) {
pdata->state = state;
}
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_WAITPID);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_IOF_COMPLETE)) {
ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
}
} else if (ORTE_PROC_STATE_TERMINATED == state) {
/* update the proc state */
ORTE_FLAG_UNSET(pdata, ORTE_PROC_FLAG_ALIVE);
if (pdata->state < ORTE_PROC_STATE_TERMINATED) {
pdata->state = state;
}
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_LOCAL)) {
/* tell the PMIx subsystem to cleanup this client */
opal_pmix.server_deregister_client(proc, NULL, NULL);
/* Clean up the session directory as if we were the process
* itself. This covers the case where the process died abnormally
* and didn't cleanup its own session directory.
*/
orte_session_dir_finalize(proc);
}
/* if we are trying to terminate and our routes are
* gone, then terminate ourselves IF no local procs
* remain (might be some from another job)
*/
if (orte_orteds_term_ordered &&
0 == orte_routed.num_routes(rtmod)) {
for (i=0; i < orte_local_children->size; i++) {
if (NULL != (pdata = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i)) &&
ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_ALIVE)) {
/* at least one is still alive */
goto cleanup;
}
}
/* call our appropriate exit procedure */
OPAL_OUTPUT_VERBOSE((5, orte_state_base_framework.framework_output,
"%s state:base all routes and children gone - exiting",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
goto cleanup;
}
/* return the allocated slot for reuse */
cleanup_node(pdata);
/* track job status */
jdata->num_terminated++;
if (jdata->num_terminated == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
/* if they requested notification upon completion, provide it */
if (orte_get_attribute(&jdata->attributes, ORTE_JOB_NOTIFY_COMPLETION, NULL, OPAL_BOOL)) {
/* notify_completion => notify the parent of the termination
* of this child job. So get the parent jobid info */
npptr = &parent;
if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_LAUNCH_PROXY, (void**)&npptr, OPAL_NAME)) {
/* notify everyone who asked for it */
target.jobid = jdata->jobid;
target.vpid = ORTE_VPID_WILDCARD;
_send_notification(OPAL_ERR_JOB_TERMINATED, &target, ORTE_NAME_WILDCARD);
} else {
target.jobid = jdata->jobid;
target.vpid = ORTE_VPID_WILDCARD;
_send_notification(OPAL_ERR_JOB_TERMINATED, &target, &parent);
}
}
} else if (ORTE_PROC_STATE_TERMINATED < pdata->state &&
!orte_job_term_ordered) {
/* if this was an abnormal term, notify the other procs of the termination */
parent.jobid = jdata->jobid;
parent.vpid = ORTE_VPID_WILDCARD;
_send_notification(OPAL_ERR_PROC_ABORTED, &pdata->name, &parent);
}
}
cleanup:
OBJ_RELEASE(caddy);
}
static void track_procs(int fd, short args, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
orte_process_name_t *proc = &caddy->name;
orte_proc_state_t state = caddy->proc_state;
orte_job_t *jdata;
orte_proc_t *pdata;
opal_output_verbose(2, orte_state_base_framework.framework_output,
"%s state:staged_hnp:track_procs called for proc %s state %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc),
orte_proc_state_to_str(state));
/* get the job object for this proc */
if (NULL == (jdata = orte_get_job_data_object(proc->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
OBJ_RELEASE(caddy);
return;
}
pdata = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, proc->vpid);
if (ORTE_PROC_STATE_RUNNING == state) {
/* update the proc state */
pdata->state = state;
jdata->num_launched++;
if (jdata->num_launched == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_RUNNING);
}
}
/* if this is a registration, check to see if it came from
* inside MPI_Init - if it did, that is not acceptable
*/
if (ORTE_PROC_STATE_REGISTERED == state) {
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_AS_MPI) &&
!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_GANG_LAUNCHED)) {
/* we can't support this - issue an error and abort */
orte_show_help("help-state-staged-hnp.txt", "mpi-procs-not-supported", true);
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_SILENT_ABORT);
}
/* update the proc state */
pdata->state = state;
jdata->num_reported++;
if (jdata->num_reported == jdata->num_procs) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_REGISTERED);
}
OBJ_RELEASE(caddy);
return;
}
if (ORTE_PROC_STATE_IOF_COMPLETE == state) {
/* update the proc state */
pdata->state = state;
/* Release only the stdin IOF file descriptor for this child, if one
* was defined. File descriptors for the other IOF channels - stdout,
* stderr, and stddiag - were released when their associated pipes
* were cleared and closed due to termination of the process
*/
if (NULL != orte_iof.close) {
orte_iof.close(proc, ORTE_IOF_STDIN);
}
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_IOF_COMPLETE);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_WAITPID)) {
goto terminated;
}
OBJ_RELEASE(caddy);
return;
}
if (ORTE_PROC_STATE_WAITPID_FIRED == state) {
/* update the proc state */
pdata->state = state;
ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_WAITPID);
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_IOF_COMPLETE)) {
goto terminated;
}
OBJ_RELEASE(caddy);
return;
}
/* if the proc terminated, see if any other procs are
* waiting to run. We assume that the app_contexts are
* in priority order, with the highest priority being
* at position 0 in the app_context array for this job
*/
if (ORTE_PROC_STATE_TERMINATED == state) {
terminated:
/* update the proc state */
ORTE_FLAG_UNSET(pdata, ORTE_PROC_FLAG_ALIVE);
pdata->state = ORTE_PROC_STATE_TERMINATED;
if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_LOCAL)) {
/* Clean up the session directory as if we were the process
* itself. This covers the case where the process died abnormally
* and didn't cleanup its own session directory.
*/
orte_session_dir_finalize(proc);
}
/* return the allocated slot for reuse */
cleanup_node(pdata);
/* track job status */
jdata->num_terminated++;
if (jdata->num_terminated == jdata->num_procs) {
/* no other procs are waiting, so end this job */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
} else if (jdata->num_mapped < jdata->num_procs) {
/* schedule the job for re-mapping so that procs
* waiting for resources can execute
*/
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);
}
/* otherwise, do nothing until more procs terminate */
OBJ_RELEASE(caddy);
return;
}
}
