floraachy
/
pcm-coordinator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: 域信息查询数算集群监控 

Signed-off-by: devad <cossjie@foxmail.com>

Former-commit-id: 15f598faac392865f3e668136339e741811b7613
This commit is contained in:
devad 2023-06-17 11:29:05 +08:00
parent c5df5b1292
commit 4549ff388b
8 changed files with 309 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
adaptor/PCM-CORE/api/desc/core/pcm-core.api
View File

@ -329,11 +329,21 @@ type (
DomainSource int64 `json:"domainSource"` // 资源域数据来源：0-nudt，1-鹏城
Stack string `json:"stack"` // 技术栈
ResourceType string `json:"resourceType"` // 资源类型
Cpu string `json:"cpu"` // cpu
Memory string `json:"memory"` // 内存
Disk string `json:"disk"` // 存储
NodeCount string `json:"nodeCount"` //节点数量
Cpu float64 `json:"cpu"` // cpu使用率
Memory float64 `json:"memory"` // 内存使用率
Disk float64 `json:"disk"` // 存储使用率
NodeCount float64 `json:"nodeCount"` //节点使用率
// DeleteFlag int64 `json:"delete_flag"` // 是否删除 0:未删除，1:已经删除
Description string `json:"description"` //集群描述
ClusterName string `json:"clusterName"` //集群名称
CpuTotal float64 `json:"cpuTotal"` //cpu总核数
MemoryTotal float64 `json:"memoryTotal"` //内存总量Gi
DiskTotal float64 `json:"diskTotal"` //存储总量GB
NodeTotal float64 `json:"nodeTotal"` //容器节点数
CpuUsage float64 `json:"cpuUsage"` //cpu已使用核数
MemoryUsage float64 `json:"memoryUsage"` //内存已使用Gi
DiskUsage float64 `json:"diskUsage"` //存储已使用GB
NodeUsage float64 `json:"nodeUsage"` //容器节点已使用
}
)
@ -379,4 +389,28 @@ type (
CreateTime string `json:"createTime"` //创建时间,
UpdateTime string `json:"updateTime"` //更新时间
}
)
)
//jccSchedule容器集群资源监控 > start
type ClusterMetrics struct {
Code int `json:"code,omitempty"`
Msg string `json:"msg,omitempty"`
Data []MetricResult `json:"data,omitempty"`
}
type MetricResult struct {
MetricName string `json:"metric_name, omitempty"`
MetricData MetricData `json:"data, omitempty"`
}
type MetricData struct {
Result []map[string]interface{} `json:"result"`
ResultType string `json:"resultType"`
}
type ResultData struct {
Status string `json:"status"`
Data MetricData `json:"data"`
}
//jccSchedule容器集群资源监控 >end

1
adaptor/PCM-CORE/api/internal/config/config.go
View File

@ -33,4 +33,5 @@ type Config struct {
CephRpcConf zrpc.RpcClientConf
OctopusRpcConf zrpc.RpcClientConf
NexusUrl string
JccScheduleUrl string
}

228
adaptor/PCM-CORE/api/internal/logic/core/listdomainresourcelogic.go
View File

@ -6,11 +6,12 @@ import (
"PCM/adaptor/PCM-CORE/model"
"PCM/common/tool"
"context"
"math/rand"
"strconv"
"time"
"fmt"
"github.com/shopspring/decimal"
"github.com/zeromicro/go-zero/core/logx"
"k8s.io/apimachinery/pkg/util/json"
"math"
"strconv"
)
type ListDomainResourceLogic struct {
@ -28,23 +29,218 @@ func NewListDomainResourceLogic(ctx context.Context, svcCtx *svc.ServiceContext)
}
func (l *ListDomainResourceLogic) ListDomainResource() (resp *types.DomainResourceResp, err error) {
//TODO 域信息查询数算集群监控
resp = &types.DomainResourceResp{}
var domainResourceModel []model.DomainResource
sqlStr := `select * from domain_resource where delete_flag=0 order by id`
sqlStr := "select * from `joint_domain`.domain_cluster"
l.svcCtx.DbEngin.Raw(sqlStr).Scan(&domainResourceModel)
tool.Convert(domainResourceModel, &resp.DomainResourceList)
rand.Seed(time.Now().UnixNano())
//TODO rand
metrics := &types.ClusterMetrics{}
//获取指定集群监控数据
for i := 0; i < len(resp.DomainResourceList); i++ {
c := rand.Intn(65) + 10
m := rand.Intn(65) + 10
d := rand.Intn(65) + 10
n := rand.Intn(65) + 10
resp.DomainResourceList[i].Cpu = strconv.Itoa(c)
resp.DomainResourceList[i].Memory = strconv.Itoa(m)
resp.DomainResourceList[i].Disk = strconv.Itoa(d)
resp.DomainResourceList[i].NodeCount = strconv.Itoa(n)
url := fmt.Sprintf("%s/api/v1/resource/getClusterMetrics?clusterName=%s", l.svcCtx.Config.JccScheduleUrl, resp.DomainResourceList[i].ClusterName)
bytes, err := tool.HttpGet("GET", url)
if err != nil {
logx.WithContext(l.ctx).Errorf("获取云算指定集群监控数据 err: %s", err.Error())
return nil, err
}
json.Unmarshal(bytes, metrics)
if metrics.Code == 200 {
//fmt.Print(metrics)
for _, metricResult := range metrics.Data {
//集群内存已使用
if metricResult.MetricName == "cluster_memory_usage_wo_cache" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1 / (1024 * 1024 * 1024)).Round(2).Float64()
resp.DomainResourceList[i].MemoryUsage = v2
} else {
continue
}
}
} else {
continue
}
}
//集群内存总量
if metricResult.MetricName == "cluster_memory_total" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1 / (1024 * 1024 * 1024)).Round(2).Float64()
resp.DomainResourceList[i].MemoryTotal = v2
} else {
continue
}
}
} else {
continue
}
}
//集群cpu已使用
if metricResult.MetricName == "cluster_cpu_usage" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1).Round(2).Float64()
resp.DomainResourceList[i].CpuUsage = v2
} else {
continue
}
}
} else {
continue
}
}
//集群cpu总核数
if metricResult.MetricName == "cluster_cpu_total" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1).Round(2).Float64()
resp.DomainResourceList[i].CpuTotal = v2
} else {
continue
}
}
} else {
continue
}
}
//集群容器节点已使用数量
if metricResult.MetricName == "cluster_pod_running_count" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1).Round(2).Float64()
resp.DomainResourceList[i].NodeUsage = v2
} else {
continue
}
}
} else {
continue
}
}
//集群容器节点总量
if metricResult.MetricName == "cluster_pod_quota" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1).Round(2).Float64()
resp.DomainResourceList[i].NodeTotal = v2
} else {
continue
}
}
} else {
continue
}
}
//集群磁盘已使用
if metricResult.MetricName == "cluster_disk_size_usage" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1 / (1000 * 1000 * 1000)).Round(2).Float64()
resp.DomainResourceList[i].DiskUsage = v2
} else {
continue
}
}
} else {
continue
}
}
//集群磁盘总量
if metricResult.MetricName == "cluster_disk_size_capacity" {
format, ok := metricResult.MetricData.Result[0]["values"].([]interface{})
if ok {
for _, val := range format {
f, ok := val.([]interface{})
if ok {
value, _ := f[1].(string)
v1, err := strconv.ParseFloat(value, 32)
if err != nil {
continue
}
v2, _ := decimal.NewFromFloat(v1 / (1000 * 1000 * 1000)).Round(2).Float64()
resp.DomainResourceList[i].DiskTotal = v2
} else {
continue
}
}
} else {
continue
}
}
if resp.DomainResourceList[i].CpuUsage != 0 && resp.DomainResourceList[i].CpuTotal != 0 {
resp.DomainResourceList[i].Cpu = math.Ceil((resp.DomainResourceList[i].CpuUsage / resp.DomainResourceList[i].CpuTotal) * 10000 / 100.00)
}
if resp.DomainResourceList[i].NodeUsage != 0 && resp.DomainResourceList[i].NodeTotal != 0 {
resp.DomainResourceList[i].NodeCount = math.Ceil((resp.DomainResourceList[i].NodeUsage / resp.DomainResourceList[i].NodeTotal) * 10000 / 100.00)
}
if resp.DomainResourceList[i].DiskUsage != 0 && resp.DomainResourceList[i].DiskTotal != 0 {
resp.DomainResourceList[i].Disk = math.Ceil((resp.DomainResourceList[i].DiskUsage / resp.DomainResourceList[i].DiskTotal) * 10000 / 100.00)
}
if resp.DomainResourceList[i].MemoryUsage != 0 && resp.DomainResourceList[i].MemoryTotal != 0 {
resp.DomainResourceList[i].Memory = math.Ceil((resp.DomainResourceList[i].MemoryUsage / resp.DomainResourceList[i].MemoryTotal) * 10000 / 100.00)
}
resp.DomainResourceList[i].DomainName = resp.DomainResourceList[i].Description
resp.DomainResourceList[i].ResourceType = "云算"
resp.DomainResourceList[i].Stack = "kubernetes"
}
}
}
return resp, nil
}

53
adaptor/PCM-CORE/api/internal/types/types.go
View File

@ -297,17 +297,27 @@ type DomainResourceResp struct {
}
type DomainResource struct {
Id int64 `json:"id"` // id
DomainId string `json:"domainId"` // 资源域id
DomainName string `json:"domainName"` // 资源域名称
JobCount int64 `json:"jobCount"` // 资源域任务数量
DomainSource int64 `json:"domainSource"` // 资源域数据来源：0-nudt，1-鹏城
Stack string `json:"stack"` // 技术栈
ResourceType string `json:"resourceType"` // 资源类型
Cpu string `json:"cpu"` // cpu
Memory string `json:"memory"` // 内存
Disk string `json:"disk"` // 存储
NodeCount string `json:"nodeCount"` //节点数量
Id int64 `json:"id"` // id
DomainId string `json:"domainId"` // 资源域id
DomainName string `json:"domainName"` // 资源域名称
JobCount int64 `json:"jobCount"` // 资源域任务数量
DomainSource int64 `json:"domainSource"` // 资源域数据来源：0-nudt，1-鹏城
Stack string `json:"stack"` // 技术栈
ResourceType string `json:"resourceType"` // 资源类型
Cpu float64 `json:"cpu"` // cpu使用率
Memory float64 `json:"memory"` // 内存使用率
Disk float64 `json:"disk"` // 存储使用率
NodeCount float64 `json:"nodeCount"` //节点使用率
Description string `json:"description"` //集群描述
ClusterName string `json:"clusterName"` //集群名称
CpuTotal float64 `json:"cpuTotal"` //cpu总核数
MemoryTotal float64 `json:"memoryTotal"` //内存总量Gi
DiskTotal float64 `json:"diskTotal"` //存储总量GB
NodeTotal float64 `json:"nodeTotal"` //容器节点数
CpuUsage float64 `json:"cpuUsage"` //cpu已使用核数
MemoryUsage float64 `json:"memoryUsage"` //内存已使用Gi
DiskUsage float64 `json:"diskUsage"` //存储已使用GB
NodeUsage float64 `json:"nodeUsage"` //容器节点已使用
}
type ResourcePanelConfigReq struct {
@ -352,6 +362,27 @@ type ResourcePanelConfigResp struct {
UpdateTime string `json:"updateTime"` //更新时间
}
type ClusterMetrics struct {
Code int `json:"code,omitempty"`
Msg string `json:"msg,omitempty"`
Data []MetricResult `json:"data,omitempty"`
}
type MetricResult struct {
MetricName string `json:"metric_name, omitempty"`
MetricData MetricData `json:"data, omitempty"`
}
type MetricData struct {
Result []map[string]interface{} `json:"result"`
ResultType string `json:"resultType"`
}
type ResultData struct {
Status string `json:"status"`
Data MetricData `json:"data"`
}
type Job struct {
SlurmVersion string `json:"slurmVersion"`
Name string `json:"name"`

2
adaptor/PCM-CORE/model/domainresourcemodel_gen.go
View File

@ -53,6 +53,8 @@ type (
CreateTime time.Time `db:"create_time"` // 数据创建时间
UpdateTime time.Time `db:"update_time"` // 数据更新时间
DeleteFlag int64 `db:"delete_flag"` // 是否删除 0:未删除，1:已经删除
Description string `db:"description"` //集群描述
ClusterName string `db:"cluster_name"` //集群名称
}
)

11
common/tool/decimal.go Normal file
View File

@ -0,0 +1,11 @@
package tool
import (
"fmt"
"strconv"
)
func FloatConv(num float64) float64 {
num, _ = strconv.ParseFloat(fmt.Sprintf("%.2f", num), 64)
return num
}

1
go.mod
View File

@ -21,6 +21,7 @@ require (
github.com/nacos-group/nacos-sdk-go/v2 v2.2.1
github.com/pkg/errors v0.9.1
github.com/robfig/cron/v3 v3.0.1
github.com/shopspring/decimal v1.3.1
github.com/sirupsen/logrus v1.9.0
github.com/zeromicro/go-queue v1.1.8
github.com/zeromicro/go-zero v1.5.1

1
go.sum
View File

@ -991,6 +991,7 @@ github.com/segmentio/kafka-go v0.4.38/go.mod h1:ikyuGon/60MN/vXFgykf7Zm8P5Be49gJ
github.com/shirou/gopsutil v2.19.11+incompatible/go.mod h1:5b4v6he4MtMOwMlS0TUMTu2PcXUg8+E1lC7eC3UO/RA=
github.com/shirou/gopsutil v3.21.11+incompatible/go.mod h1:5b4v6he4MtMOwMlS0TUMTu2PcXUg8+E1lC7eC3UO/RA=
github.com/shirou/w32 v0.0.0-20160930032740-bb4de0191aa4/go.mod h1:qsXQc7+bwAM3Q1u/4XEfrquwF8Lw7D7y5cD8CuHnfIc=
github.com/shopspring/decimal v1.3.1 h1:2Usl1nmF/WZucqkFZhnfFYxxxu8LG21F6nPQBE5gKV8=
github.com/shopspring/decimal v1.3.1/go.mod h1:DKyhrW/HYNuLGql+MJL6WCR6knT2jwCFRcu2hWCYk4o=
github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeVRXNmyL/1OwPU0+IJeTBvfc=
github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=