1
0
mirror of https://github.com/newnius/YAO-scheduler.git synced 2025-06-07 22:31:55 +00:00
YAO-scheduler/src/job_manager.go

165 lines
3.3 KiB
Go
Raw Normal View History

2019-03-04 09:19:55 +00:00
package main
import (
"time"
"log"
2019-03-20 03:14:07 +00:00
"net/url"
"strings"
"io/ioutil"
"encoding/json"
2019-04-12 09:21:09 +00:00
"fmt"
2019-03-04 09:19:55 +00:00
)
type JobManager struct {
2019-03-20 03:14:07 +00:00
allocator *AllocatorFIFO
job Job
jobStatus JobStatus
resources []MsgAgent
2019-03-04 09:19:55 +00:00
}
2019-03-20 03:14:07 +00:00
func (jm *JobManager) start() {
log.Println("start job ", jm.job.Name)
2019-03-25 07:36:30 +00:00
jm.jobStatus = JobStatus{Name: jm.job.Name, tasks: map[string]TaskStatus{}}
2019-03-20 03:14:07 +00:00
/* request for resources */
for i := range jm.job.Tasks {
var resource MsgAgent
for {
resource = jm.allocator.requestResource(jm.job.Tasks[i])
if len(resource.Status) > 0 {
break
}
}
log.Println("Receive resource", resource)
jm.resources = append(jm.resources, resource)
}
2019-04-12 09:21:09 +00:00
jm.allocator.ack(&jm.job)
2019-03-04 09:19:55 +00:00
/* bring up containers */
2019-03-20 03:14:07 +00:00
for i := range jm.job.Tasks {
2019-04-12 09:21:09 +00:00
var GPUs []string
for _, GPU := range jm.resources[i].Status {
GPUs = append(GPUs, GPU.UUID)
}
2019-03-20 03:14:07 +00:00
v := url.Values{}
2019-04-12 09:21:09 +00:00
v.Set("image", jm.job.Tasks[i].Image)
2019-03-20 03:14:07 +00:00
v.Set("cmd", jm.job.Tasks[i].Cmd)
2019-04-12 09:21:09 +00:00
v.Set("name", jm.job.Tasks[i].Name)
v.Set("workspace", jm.job.Workspace)
v.Set("gpus", strings.Join(GPUs, ","))
fmt.Print(v.Encode())
2019-03-20 03:14:07 +00:00
resp, err := doRequest("POST", "http://kafka_node1:8000/create", strings.NewReader(v.Encode()), "application/x-www-form-urlencoded", "")
if err != nil {
log.Println(err)
return
}
defer resp.Body.Close()
body, err := ioutil.ReadAll(resp.Body)
if err != nil {
log.Println(err)
return
}
log.Println(string(body))
var res MsgCreate
err = json.Unmarshal([]byte(string(body)), &res)
if err != nil {
log.Println(err)
return
}
jm.jobStatus.tasks[jm.job.Tasks[i].Name] = TaskStatus{Id: res.Id}
}
2019-04-12 09:21:09 +00:00
jm.allocator.running(&jm.job)
2019-03-04 09:19:55 +00:00
/* monitor job execution */
for {
2019-03-20 03:14:07 +00:00
res := jm.status()
2019-03-25 07:36:30 +00:00
flag := false
2019-03-20 03:14:07 +00:00
for i := range res.Status {
if res.Status[i].Status == "running" {
log.Println(jm.job.Name, "-", i, " is running")
2019-03-25 07:36:30 +00:00
flag = true
2019-03-20 03:14:07 +00:00
} else {
log.Println(jm.job.Name, "-", i, " ", res.Status[i].Status)
/* save logs etc. */
/* return resource */
jm.allocator.returnResource(jm.resources[i])
}
}
2019-03-25 07:36:30 +00:00
if !flag {
break
}
2019-03-20 03:14:07 +00:00
time.Sleep(time.Second * 10)
2019-03-04 09:19:55 +00:00
}
2019-03-25 07:36:30 +00:00
jm.allocator.finish(&jm.job)
2019-03-20 03:14:07 +00:00
log.Println("finish job", jm.job.Name)
}
func (jm *JobManager) logs(taskName string) MsgLog {
spider := Spider{}
spider.Method = "GET"
2019-03-25 07:36:30 +00:00
spider.URL = "http://kafka_node1:8000/logs?id=" + taskName
2019-03-20 03:14:07 +00:00
err := spider.do()
if err != nil {
return MsgLog{Code: 1, Error: err.Error()}
}
resp := spider.getResponse()
defer resp.Body.Close()
body, err := ioutil.ReadAll(resp.Body)
if err != nil {
return MsgLog{Code: 1, Error: err.Error()}
}
var res MsgLog
err = json.Unmarshal([]byte(string(body)), &res)
if err != nil {
log.Println(err)
return MsgLog{Code: 1, Error: "Unknown"}
}
return res
}
func (jm *JobManager) status() MsgJobStatus {
var tasksStatus []TaskStatus
for _, taskStatus := range jm.jobStatus.tasks {
spider := Spider{}
spider.Method = "GET"
2019-03-25 07:36:30 +00:00
spider.URL = "http://kafka_node1:8000/status?id=" + taskStatus.Id
2019-03-20 03:14:07 +00:00
err := spider.do()
if err != nil {
continue
}
resp := spider.getResponse()
defer resp.Body.Close()
body, err := ioutil.ReadAll(resp.Body)
if err != nil {
continue
}
var res MsgTaskStatus
err = json.Unmarshal([]byte(string(body)), &res)
if err != nil {
continue
}
tasksStatus = append(tasksStatus, res.Status)
}
2019-03-04 09:19:55 +00:00
2019-03-20 03:14:07 +00:00
return MsgJobStatus{Status: tasksStatus}
2019-03-04 09:19:55 +00:00
}