620 lines
17 KiB
C
620 lines
17 KiB
C
#include <errno.h>
|
|
#include <stdlib.h>
|
|
|
|
#include "cluster.h"
|
|
#include "controller.capnp.h"
|
|
#include "objmgr.h"
|
|
#include "request.h"
|
|
#include "worker.h"
|
|
#include "xce/cluster_def.h"
|
|
#include "xce/types.h"
|
|
|
|
extern void collect_shared_resvals(cluster_t*, collect_check_fn_t,
|
|
int*, xce_val_with_idx_t***);
|
|
|
|
static int xcesched_push_joblist(cluster_t* c,
|
|
xlist_t* l)
|
|
{
|
|
xce_data_t data;
|
|
xce_jobs_data_t jobs = {
|
|
.n_total = l->count,
|
|
.pos = 0,
|
|
.count = l->count,
|
|
.jobs = (xce_job_t**)l->data
|
|
};
|
|
int rc = 0;
|
|
|
|
memset(&data, 0x0, sizeof(data));
|
|
|
|
data.table = Data_data_schedjobs;
|
|
data.data.jobs = &jobs;
|
|
|
|
rc = xce_push_data_req(c->sched.ep, c->myhost, &data,
|
|
check_controller, c, 100);
|
|
return rc;
|
|
}
|
|
|
|
static int xcesched_push_workerlist(cluster_t* c,
|
|
xlist_t* l)
|
|
{
|
|
xce_data_t data;
|
|
xce_workers_data_t workers = {
|
|
.n_total = l->count,
|
|
.pos = 0,
|
|
.count = l->count,
|
|
.workers = (xce_worker_t**)l->data
|
|
};
|
|
int rc = 0;
|
|
|
|
memset(&data, 0x0, sizeof(data));
|
|
|
|
data.table = Data_data_schedworkers;
|
|
data.data.workers = &workers;
|
|
|
|
rc = xce_push_data_req(c->sched.ep, c->myhost, &data,
|
|
check_controller, c, 100);
|
|
return rc;
|
|
}
|
|
|
|
static int xcesched_push_configs(cluster_t* c)
|
|
{
|
|
xce_configdata_t configdata;
|
|
int rc = 0;
|
|
|
|
memset(&configdata, 0x0, sizeof(configdata));
|
|
|
|
configdata.data.conf4sched = (xce_conf4sched_t*)calloc(1, sizeof(xce_conf4sched_t));
|
|
if (configdata.data.conf4sched == NULL) {
|
|
return -ENOMEM;
|
|
}
|
|
|
|
collect_confvars(c, NULL, &(configdata.data.conf4sched->n_confvars),
|
|
&(configdata.data.conf4sched->confvars));
|
|
collect_queues(c, NULL, &(configdata.data.conf4sched->n_queues),
|
|
&(configdata.data.conf4sched->queues));
|
|
collect_resources(c, NULL, &(configdata.data.conf4sched->n_resources),
|
|
&(configdata.data.conf4sched->resources));
|
|
collect_shared_resvals(c, NULL, &(configdata.data.conf4sched->nresvals),
|
|
&(configdata.data.conf4sched->resvals));
|
|
|
|
configdata.data.conf4sched->n_ugroups = c->myconf->cluster_config->nugroups;
|
|
configdata.data.conf4sched->ugroups = c->myconf->cluster_config->ugroups;
|
|
configdata.data.conf4sched->n_hgroups = c->myconf->cluster_config->nhgroups;
|
|
configdata.data.conf4sched->hgroups = c->myconf->cluster_config->hgroups;
|
|
|
|
/* add resources */
|
|
configdata.data_which = ConfigData_data_schedconfig;
|
|
configdata.data.conf4sched->clustername = c->myconf->cluster_config->clustername;
|
|
|
|
rc = xce_pushconfig_req(c->sched.ep, c->myhost,
|
|
&configdata);
|
|
|
|
free_resval_with_idx_list(&(configdata.data.conf4sched->resvals),
|
|
configdata.data.conf4sched->nresvals);
|
|
|
|
deref_objs(configdata.data.conf4sched->n_confvars,
|
|
(void**)configdata.data.conf4sched->confvars);
|
|
deref_objs(configdata.data.conf4sched->n_queues,
|
|
(void**)configdata.data.conf4sched->queues);
|
|
deref_objs(configdata.data.conf4sched->n_resources,
|
|
(void**)configdata.data.conf4sched->resources);
|
|
|
|
FREEUP(configdata.data.conf4sched->confvars);
|
|
FREEUP(configdata.data.conf4sched->queues);
|
|
FREEUP(configdata.data.conf4sched->resources);
|
|
|
|
FREEUP(configdata.data.conf4sched);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int check_job_(xce_job_t* j)
|
|
{
|
|
if (j == NULL) {
|
|
return 0;
|
|
}
|
|
|
|
return 1;
|
|
}
|
|
|
|
static int xcesched_push_jobs(cluster_t* c)
|
|
{
|
|
xce_data_t data;
|
|
xce_jobs_data_t jobs = {
|
|
.n_total = 0,
|
|
.pos = 0,
|
|
.count = 0,
|
|
.jobs = NULL
|
|
};
|
|
int rc = 0;
|
|
|
|
memset(&data, 0x0, sizeof(data));
|
|
|
|
data.table = Data_data_schedjobs;
|
|
data.data.jobs = &jobs;
|
|
|
|
collect_jobs(c, (collect_check_fn_t)check_job_,
|
|
&(jobs.n_total), &(jobs.jobs));
|
|
jobs.count = jobs.n_total;
|
|
|
|
rc = xce_push_data_req(c->sched.ep, c->myhost, &data,
|
|
check_controller, c, 100);
|
|
deref_objs(jobs.n_total, (void**)jobs.jobs);
|
|
FREEUP(jobs.jobs);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int xcesched_push_workers(cluster_t* c)
|
|
{
|
|
xce_data_t data;
|
|
xce_workers_data_t workers = {
|
|
.n_total = 0,
|
|
.pos = 0,
|
|
.count = 0,
|
|
.workers = NULL
|
|
};
|
|
int rc = 0;
|
|
int i;
|
|
|
|
memset(&data, 0x0, sizeof(data));
|
|
|
|
data.table = Data_data_schedworkers;
|
|
data.data.workers = &workers;
|
|
|
|
collect_workers(c, (collect_check_fn_t)check_worker_,
|
|
&(workers.n_total), &(workers.workers));
|
|
workers.count = workers.n_total;
|
|
|
|
rc = xce_push_data_req(c->sched.ep, c->myhost, &data,
|
|
check_controller, c, 100);
|
|
|
|
/* free up collected resource value which used
|
|
passed to scheduler
|
|
*/
|
|
for (i = 0; i < workers.n_total; i++) {
|
|
xce_worker_t* w = workers.workers[i];
|
|
|
|
free_resval_with_idx_list(&(w->resvals_to_sched),
|
|
w->nresvals_to_sched);
|
|
}
|
|
|
|
deref_objs(workers.n_total, (void**)workers.workers);
|
|
FREEUP(workers.workers);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_new_fun(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
xlist_t* l = va_arg(ap, xlist_t*);
|
|
int rc = 0;
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = -EINVAL;
|
|
if (l != NULL) {
|
|
rc = xcesched_push_joblist(c, l);
|
|
}
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_start_fun(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
job_pidinfo_t* info = va_arg(ap, job_pidinfo_t*);
|
|
int rc = 0;
|
|
/* clang-format off */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedStartJob,
|
|
.data = {
|
|
.sched_start_job = info->jobid
|
|
}
|
|
};
|
|
/* clang-format on */
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_exit_fun(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
xce_jobfinish_t* info = va_arg(ap, xce_jobfinish_t*);
|
|
int rc = 0;
|
|
/* clang-format off */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedExitJob,
|
|
.data = {
|
|
.sched_exit_job = info->jobid
|
|
}
|
|
};
|
|
/* clang-format on */
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_fail_fun(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
job_failinfo_t* info = va_arg(ap, job_failinfo_t*);
|
|
int rc = 0;
|
|
/* clang-format off */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedFailJob,
|
|
.data = {
|
|
.sched_fail_job = info->jobid
|
|
}
|
|
};
|
|
/* clang-format on */
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_update_fun(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
job_t* job = va_arg(ap, job_t*);
|
|
int rc = 0;
|
|
/* clang-format off */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedUpdateJob,
|
|
.data = {
|
|
.sched_update_job = &(job->job)
|
|
}
|
|
};
|
|
/* clang-format on */
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_job_hook_fun(int type,
|
|
int situation,
|
|
int detail,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
switch (detail) {
|
|
case HOOK_DETAIL_JOB_NEW:
|
|
return sched_job_new_fun(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_JOB_START:
|
|
return sched_job_start_fun(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_JOB_EXIT:
|
|
return sched_job_exit_fun(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_JOB_UPDATE:
|
|
return sched_job_update_fun(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_JOB_FAIL:
|
|
return sched_job_fail_fun(type, situation,
|
|
data, ap);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
static int sched_worker_change(int type,
|
|
int situation,
|
|
void* data,
|
|
cluster_t* c,
|
|
worker_t* w)
|
|
{
|
|
int rc = 0;
|
|
xlist_t wlist;
|
|
|
|
TRACE("%s worker %s idx %d situation %d",
|
|
__func__, w->host.name,
|
|
w->idx, situation);
|
|
|
|
collect_resval_with_idx(&(w->host.resvals),
|
|
&(w->host.nresvals_to_sched),
|
|
&(w->host.resvals_to_sched));
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
xlist_init(&wlist, 2);
|
|
xlist_store(&wlist, w);
|
|
rc = xcesched_push_workerlist(c, &wlist);
|
|
xlist_fini(&wlist);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
free_resval_with_idx_list(&(w->host.resvals_to_sched),
|
|
w->host.nresvals_to_sched);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_worker_join(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
char* name = va_arg(ap, char*);
|
|
worker_t* w = NULL;
|
|
int rc;
|
|
|
|
if (situation == HOOK_BEFORE_EVENT) {
|
|
/* nothing to do */
|
|
return 0;
|
|
}
|
|
|
|
w = workermgr_lock(&(c->workermgr), name,
|
|
OBJREF_READ);
|
|
|
|
rc = sched_worker_change(type, situation, data,
|
|
c, w);
|
|
|
|
/* FIXME: need to mark the jobs in this worker */
|
|
obj_deref(w);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_worker_dead(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
worker_t* w = va_arg(ap, worker_t*);
|
|
|
|
/* FIXME: need to handle job status in this worker */
|
|
if (situation == HOOK_BEFORE_EVENT) {
|
|
return 0;
|
|
}
|
|
|
|
return sched_worker_change(type, situation, data,
|
|
c, w);
|
|
}
|
|
|
|
static int sched_worker_update(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
worker_t* w = va_arg(ap, worker_t*);
|
|
|
|
if (situation == HOOK_BEFORE_EVENT) {
|
|
return 0;
|
|
}
|
|
|
|
return sched_worker_change(type, situation, data,
|
|
c, w);
|
|
}
|
|
static int sched_worker_hook_fun(int type,
|
|
int situation,
|
|
int detail,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
switch (detail) {
|
|
case HOOK_DETAIL_WORKER_JOIN:
|
|
return sched_worker_join(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_WORKER_DEAD:
|
|
return sched_worker_dead(type, situation,
|
|
data, ap);
|
|
case HOOK_DETAIL_WORKER_UPDATE:
|
|
return sched_worker_update(type, situation,
|
|
data, ap);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int sched_resource_hook_fun(int type,
|
|
int situation,
|
|
int detail,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
xce_resval_list_t* resvals = va_arg(ap, xce_resval_list_t*);
|
|
/* clang-format on */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedUpdateResvals,
|
|
.data = {
|
|
.sched_update_resvals = resvals }
|
|
};
|
|
/* clang-format off */
|
|
int rc = 0;
|
|
|
|
if (detail != HOOK_DETAIL_RESOURCE_UPDATE) {
|
|
return 0;
|
|
}
|
|
|
|
xce_use(resvals);
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int sched_queue_update(int type,
|
|
int situation,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
cluster_t* c = va_arg(ap, cluster_t*);
|
|
queue_t* q = va_arg(ap, queue_t*);
|
|
int rc = 0;
|
|
/* clang-format off */
|
|
xce_notify_data_t xdata = {
|
|
.event = NotifyData_data_schedUpdateQueue,
|
|
.data = {
|
|
.sched_update_queue = &(q->queue)
|
|
}
|
|
};
|
|
/* clang-format on */
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
switch (situation) {
|
|
case HOOK_AFTER_EVENT:
|
|
rc = notify(c, c->sched.ep, &xdata);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
return rc;
|
|
}
|
|
static int sched_queue_hook_fun(int type,
|
|
int situation,
|
|
int detail,
|
|
void* data,
|
|
va_list ap)
|
|
{
|
|
switch (detail) {
|
|
case HOOK_DETAIL_QUEUE_UPDATE:
|
|
return sched_queue_update(type, situation, data, ap);
|
|
default:
|
|
break;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
int on_scheduler_join(cluster_t* c)
|
|
{
|
|
static int hook_installed = 0;
|
|
|
|
mylock_lock(&(c->sched.lock));
|
|
if (c->sched.ep != NULL) {
|
|
xcesched_push_configs(c);
|
|
xcesched_push_workers(c);
|
|
xcesched_push_jobs(c);
|
|
}
|
|
mylock_unlock(&(c->sched.lock));
|
|
|
|
if (!hook_installed) {
|
|
hook_installed = 1;
|
|
|
|
xce_add_hook(&(c->hooksys),
|
|
XCE_CONTROLLER_HOOK_JOB,
|
|
(hook_fn_t)sched_job_hook_fun,
|
|
NULL);
|
|
xce_add_hook(&(c->hooksys),
|
|
XCE_CONTROLLER_HOOK_WORKER,
|
|
(hook_fn_t)sched_worker_hook_fun,
|
|
NULL);
|
|
xce_add_hook(&(c->hooksys),
|
|
XCE_CONTROLLER_HOOK_RESOURCE,
|
|
(hook_fn_t)sched_resource_hook_fun,
|
|
NULL);
|
|
xce_add_hook(&(c->hooksys),
|
|
XCE_CONTROLLER_HOOK_QUEUE,
|
|
(hook_fn_t)sched_queue_hook_fun,
|
|
NULL);
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
int handle_scheduler_leave(cluster_t* c)
|
|
{
|
|
/* we need to mark the cluster status */
|
|
TRACE("%s scheduler leave",
|
|
__func__);
|
|
mylock_lock(&(c->sched.lock));
|
|
c->sched.ep = NULL;
|
|
mylock_unlock(&(c->sched.lock));
|
|
return 0;
|
|
}
|