reformat code and job array

This commit is contained in:
Rongsong Shen 2025-11-07 16:14:27 +08:00
parent a665ccd4fe
commit 9ff857db9f
64 changed files with 1497 additions and 2476 deletions

View file

@ -7,20 +7,16 @@
#include "xce/errnox.h"
#include "xce/types.h"
static int run_tasks_on_host(cluster_t* c,
endpoint_t* ep,
uint64_t jobid,
static int run_tasks_on_host(cluster_t* c, endpoint_t* ep, uint64_t jobid,
xce_taskhost_t* info)
{
worker_t* w;
ioctx_t* ioctx;
int rc;
xce_taskrun_t runmsg = {
.nhosts = 1,
.jobid = jobid,
.clientidx = ep->idx,
.taskhosts = (xce_taskhost_t*[]) { info }
};
xce_taskrun_t runmsg = { .nhosts = 1,
.jobid = jobid,
.clientidx = ep->idx,
.taskhosts = (xce_taskhost_t*[]) { info } };
w = workermgr_lock(&(c->workermgr), info->host, OBJREF_READ);
if (w == NULL) {
@ -45,16 +41,14 @@ static int run_tasks_on_host(cluster_t* c,
return rc;
}
static int check_task_hosts(cluster_t* c,
xce_taskrun_t* runmsg)
static int check_task_hosts(cluster_t* c, xce_taskrun_t* runmsg)
{
int i;
for (i = 0; i < runmsg->nhosts; i++) {
worker_t* w;
w = workermgr_lock(&(c->workermgr),
runmsg->taskhosts[i]->host,
w = workermgr_lock(&(c->workermgr), runmsg->taskhosts[i]->host,
OBJREF_READ);
if (w == NULL) {
return -ENOENT;
@ -69,9 +63,7 @@ static int check_task_hosts(cluster_t* c,
return 0;
}
int do_runtasks(cluster_t* c,
endpoint_t* ep,
xce_request_t* req)
int do_runtasks(cluster_t* c, endpoint_t* ep, xce_request_t* req)
{
int rc = 0;
xce_reply_t reply;
@ -94,22 +86,16 @@ int do_runtasks(cluster_t* c,
for (i = 0; i < runmsg->nhosts; i++) {
int rc1;
rc1 = run_tasks_on_host(c, ep,
runmsg->jobid,
runmsg->taskhosts[i]);
rc1 = run_tasks_on_host(c, ep, runmsg->jobid, runmsg->taskhosts[i]);
if (rc1 < 0) {
xce_task_failinfo_t failinfo = {
.jobid = runmsg->jobid,
.host = runmsg->taskhosts[i]->host,
.taskidx = -1, /* all tasks failed */
.code = rc1,
.detail = NULL
};
xce_notify_data_t data = {
.event = NotifyData_data_taskFailInfo,
.data = {
.taskfailinfo = &failinfo }
};
xce_task_failinfo_t failinfo
= { .jobid = runmsg->jobid,
.host = runmsg->taskhosts[i]->host,
.taskidx = -1, /* all tasks failed */
.code = rc1,
.detail = NULL };
xce_notify_data_t data = { .event = NotifyData_data_taskFailInfo,
.data = { .taskfailinfo = &failinfo } };
/* notify client that run tasks on host fail */
notify(c, ep, &data);
@ -121,16 +107,12 @@ int do_runtasks(cluster_t* c,
return rc;
}
int on_task_start(cluster_t* c,
xce_task_pidinfo_t* info)
int on_task_start(cluster_t* c, xce_task_pidinfo_t* info)
{
endpoint_t* ep;
xce_notify_data_t data = {
.event = NotifyData_data_taskPidInfo,
.data = {
.taskinfo = info }
};
int clientidx = info->clientidx;
xce_notify_data_t data = { .event = NotifyData_data_taskPidInfo,
.data = { .taskinfo = info } };
int clientidx = info->clientidx;
if (clientidx < 0) {
ERROR("incorrect client idx %d", clientidx);
@ -148,16 +130,12 @@ int on_task_start(cluster_t* c,
return 0;
}
int on_task_fail(cluster_t* c,
xce_task_failinfo_t* info)
int on_task_fail(cluster_t* c, xce_task_failinfo_t* info)
{
endpoint_t* ep;
xce_notify_data_t data = {
.event = NotifyData_data_taskFailInfo,
.data = {
.taskfailinfo = info }
};
int clientidx = info->clientidx;
xce_notify_data_t data = { .event = NotifyData_data_taskFailInfo,
.data = { .taskfailinfo = info } };
int clientidx = info->clientidx;
if (clientidx < 0) {
ERROR("incorrect client idx %d", clientidx);
@ -175,16 +153,12 @@ int on_task_fail(cluster_t* c,
return 0;
}
int on_task_finish(cluster_t* c,
xce_task_finish_t* info)
int on_task_finish(cluster_t* c, xce_task_finish_t* info)
{
endpoint_t* ep;
xce_notify_data_t data = {
.event = NotifyData_data_taskFinish,
.data = {
.taskfinish = info }
};
int clientidx = info->clientidx;
xce_notify_data_t data = { .event = NotifyData_data_taskFinish,
.data = { .taskfinish = info } };
int clientidx = info->clientidx;
if (clientidx < 0) {
ERROR("incorrect client idx %d", clientidx);