提交代码

This commit is contained in:
zhouqunjie 2023-11-29 16:48:16 +08:00
parent a8959f0e07
commit 5db33d3353
4 changed files with 99 additions and 55 deletions

View File

@ -4,10 +4,10 @@ NacosConfig:
ServerConfigs:
# - IpAddr: 127.0.0.1
# Port: 8848
- IpAddr: nacos.jcce.dev
- IpAddr: 119.45.100.73
Port: 8848
ClientConfig:
NamespaceId: test
NamespaceId: zhouqj
TimeoutMs: 5000
NotLoadCacheAtStart: true
LogDir:

View File

@ -44,10 +44,17 @@ func (l *CloudMq) Consume(val string) error {
return err
}
//检测是否指定了集群列表
schdl.SpecifyClusters()
//检测是否指定了nsID
schdl.SpecifyNsID()
//通过标签匹配筛选出集群范围
schdl.MatchLabels()
//todo 屏蔽调度算法,
//todo 屏蔽原调度算法,因为监控数据暂未上报,临时采用随机调度
schdl.TempAssign()
//err = schdl.AssignAndSchedule()
//if err != nil {
// return err

View File

@ -49,6 +49,7 @@ func (cs *cloudScheduler) pickOptimalStrategy(task *algo.Task, providers ...*alg
}
func (cs *cloudScheduler) getNewStructForDb(task *response.TaskInfo, participantId int64, replica int32) (interface{}, error) {
bytes, err := json.Marshal(task.Metadata)
var bytesNew []byte
//replicas := task.Metadata.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64)
@ -61,13 +62,16 @@ func (cs *cloudScheduler) getNewStructForDb(task *response.TaskInfo, participant
deployment := v1.Deployment{}
json.Unmarshal(bytes, &deployment)
deployment.Spec.Replicas = &replica
deployment.Namespace = cloud.NsID
bytesNew, _ = json.Marshal(deployment)
case "StatefulSet":
statefulSet := v1.StatefulSet{}
json.Unmarshal(bytes, &statefulSet)
statefulSet.Spec.Replicas = &replica
statefulSet.Namespace = cloud.NsID
bytesNew, _ = json.Marshal(statefulSet)
}
cloud.Replica = replica
cloud.Id = utils.GenSnowflakeID()

View File

@ -16,25 +16,26 @@ package scheduler
import (
"encoding/json"
"fmt"
"github.com/pkg/errors"
"github.com/zeromicro/go-zero/core/logx"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algo"
tool "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/rpc/client/participantservice"
"gorm.io/gorm"
"math/rand"
"strconv"
"strings"
"time"
)
type Replicas int64
type ParticipantId int64
type scheduler struct {
task *response.TaskInfo
participantIds []int64
scheduleService scheduleService
dbEngin *gorm.DB
result map[ParticipantId]Replicas
result map[int64]string //pID:子任务yamlstring 键值对
participantRpc participantservice.ParticipantService
}
@ -44,48 +45,40 @@ func NewScheduler(scheduleService scheduleService, val string, dbEngin *gorm.DB,
if err != nil {
return nil, errors.New("create scheduler failed : " + err.Error())
}
return &scheduler{task: task, scheduleService: scheduleService, dbEngin: dbEngin, participantRpc: participantRpc, result: make(map[ParticipantId]Replicas, 0)}, nil
return &scheduler{task: task, scheduleService: scheduleService, dbEngin: dbEngin, participantRpc: participantRpc, result: make(map[int64]string, 0)}, nil
}
func (s *scheduler) SepcifyClusters() {
func (s *scheduler) SpecifyClusters() {
// 如果已指定集群名通过数据库查询后返回p端ip列表
if len(s.task.Clusters) != 0 {
s.dbEngin.Raw("select id from sc_participant_phy_info where `name` in (?)", s.task.Clusters).Scan(&s.participantIds)
return
}
}
func (s *scheduler) SpecifyNsID() {
// 未指定集群名只指定nsID
if len(s.task.Clusters) == 0 {
if len(s.task.NsID) != 0 {
var clusters string
s.dbEngin.Raw("select clusters from sc_tenant_info where `tenant_name` = ?", s.task.NsID).Scan(&clusters)
clusterArr := strings.Split(clusters, ",")
s.dbEngin.Raw("select id from sc_participant_phy_info where `name` in (?)", clusterArr).Scan(&s.participantIds)
}
} else {
return
}
}
func (s *scheduler) MatchLabels() {
var ids []int64
count := 0
// 已指定 ParticipantId 直接不走标签匹配
if s.task.ParticipantId != 0 {
return
}
// 如果已指定集群名通过数据库查询后返回p端ip列表
if len(s.task.Clusters) != 0 {
for i, _ := range s.task.Clusters {
clusterName := s.task.Clusters[i]
var participantId int64
s.dbEngin.Raw("select id from sc_participant_phy_info where `name` = ?", clusterName).Scan(&participantId)
s.participantIds = append(s.participantIds, participantId)
}
return
}
// 如果未指定集群名,通过用户名(nsID)筛选出用户对应的集群
if len(s.task.NsID) != 0 {
var clusters string
s.dbEngin.Raw("select clusters from sc_tenant_info where `tenant_name` = ?", s.task.NsID).Scan(&clusters)
clusterArr := strings.Split(clusters, ",")
for i, _ := range clusterArr {
clusterName := clusterArr[i]
var participantId int64
s.dbEngin.Raw("select id from sc_participant_phy_info where `name` = ?", clusterName).Scan(&participantId)
s.participantIds = append(s.participantIds, participantId)
}
} else {
//如果均未指定,则通过标签匹配
// 集群和nsID都未指定则通过标签匹配
if len(s.task.Clusters) == 0 && len(s.task.NsID) == 0 {
//如果集群列表或nsID均未指定
for key := range s.task.MatchLabels {
var participantIds []int64
s.dbEngin.Raw("select participant_id from sc_participant_label_info where `key` = ? and value = ?", key, s.task.MatchLabels[key]).Scan(&participantIds)
@ -96,19 +89,54 @@ func (s *scheduler) MatchLabels() {
count++
}
s.participantIds = ids
} else {
return
}
}
// TempAssign todo 屏蔽原调度算法
func (s *scheduler) TempAssign() error {
//需要判断task中的资源类型针对metadata中的多个kind做不同处理
//输入副本数和集群列表最终结果输出为pID对应副本数量列表针对多个kind需要做拆分和重新拼接组合
var resources []interface{}
tool.Convert(s.task.Metadata, &resources)
for _, resource := range resources {
//如果是Deployment需要对副本数做分发
if resource.(map[string]interface{})["kind"].(string) == "Deployment" || resource.(map[string]interface{})["kind"].(string) == "Replicaset" {
//replicas := int(resource.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64))
rand.Seed(time.Now().UnixNano())
//// 生成pID对应副本数 数组
//arrReplica := make(map[int64]int, len(s.participantIds))
//for i := 0; i < len(s.participantIds)-1; i++ {
// arrReplica[s.participantIds[i]] = rand.Intn(replicas)
// replicas -= arrReplica[s.participantIds[i]] // 更新剩余的和
//}
//arrReplica[s.participantIds[len(s.participantIds)-1]] = replicas
//
////将副本数依次写入新的yaml中并生成result数据
//yamlArray := make(map[int64]string, len(s.participantIds))
//
//for i := 0; i < len(s.participantIds)-1; i++ {
// //调整yaml
// yamlArray[s.participantIds[i]] = "sds"
//}
}
s.result[s.participantIds[0]] = ""
}
//todo 屏蔽调度算法目前直接在matchlabel阶段随机分配到不同集群
replicas := s.task.Metadata.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64)
//var crd = s.task.Metadata
//for i := 0; i < len(s.task.Metadata); i++ {
//
//}
array := make([]int, len(s.participantIds))
for i := range array {
array[i] = int(replicas) / len(s.participantIds)
s.result[ParticipantId(s.participantIds[i])] = Replicas(array[i])
}
array[0] += int(replicas) % len(s.participantIds)
s.result[ParticipantId(s.participantIds[0])] = Replicas(array[0])
return
return nil
}
func (s *scheduler) AssignAndSchedule() error {
@ -125,8 +153,8 @@ func (s *scheduler) AssignAndSchedule() error {
if len(s.participantIds) == 1 {
s.task.ParticipantId = s.participantIds[0]
replicas := s.task.Metadata.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64)
result := make(map[ParticipantId]Replicas)
result[ParticipantId(s.participantIds[0])] = Replicas(replicas)
result := make(map[int64]string)
result[s.participantIds[0]] = strconv.FormatFloat(replicas, 'f', 2, 64)
s.result = result
return nil
@ -162,7 +190,12 @@ func (s *scheduler) AssignAndSchedule() error {
func (s *scheduler) SaveToDb() error {
for key, value := range s.result {
structForDb, err := s.scheduleService.getNewStructForDb(s.task, int64(key), int32(value))
num, err := strconv.Atoi(value)
if err != nil {
fmt.Println("转换失败:", err)
}
structForDb, err := s.scheduleService.getNewStructForDb(s.task, int64(key), int32(num))
if err != nil {
return err
}
@ -196,7 +229,7 @@ func (s *scheduler) assignReplicasToResult(strategy *algo.Strategy, providerList
if e == 0 {
continue
}
s.result[ParticipantId(providerList[i].Pid)] = Replicas(int64(e))
s.result[providerList[i].Pid] = string(e)
}
if len(s.result) == 0 {