Merge pull request 'alert' (#102) from zhangweiii/pcm-coordinator:master into master

Former-commit-id: e884d1e9647d53e0e2ec288a8c3fdb7d3ab3eb45
This commit is contained in:
zhangweiii 2024-04-08 09:20:26 +08:00
commit 801c1b9cb5
6 changed files with 137 additions and 9 deletions

View File

@ -12,7 +12,11 @@ type CreateAlertRuleReq {
}
type (
alertRulesResp {
AlertRulesResp {
alertRules []AlertRule `json:"alertRules"`
}
AlertRule {
Id int64 `json:"id"`
ClusterName string `json:"clusterName"`
Name string `json:"name"`
@ -29,6 +33,8 @@ type (
Metrics string `form:"metrics"`
}
nodesLoadTopResp {
code int `json:"code"`
data interface{} `json:"data"`
msg string `json:"msg"`
}
)

View File

@ -952,7 +952,7 @@ service pcm {
@doc "alert rules"
@handler alertRulesHandler
get /monitoring/alert/rule returns (alertRulesResp)
get /monitoring/alert/rule returns (AlertRulesResp)
@doc "cluster resource load"
@handler clustersLoadHandler

View File

@ -1,6 +1,7 @@
package monitoring
import (
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
"net/http"
"github.com/zeromicro/go-zero/rest/httpx"
@ -19,10 +20,6 @@ func NodesLoadTopHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
l := monitoring.NewNodesLoadTopLogic(r.Context(), svcCtx)
resp, err := l.NodesLoadTop(&req)
if err != nil {
httpx.ErrorCtx(r.Context(), w, err)
} else {
httpx.OkJsonCtx(r.Context(), w, resp)
}
result.HttpResult(r, w, resp, err)
}
}

View File

@ -25,6 +25,8 @@ func NewAlertRulesLogic(ctx context.Context, svcCtx *svc.ServiceContext) *AlertR
func (l *AlertRulesLogic) AlertRules() (resp *types.AlertRulesResp, err error) {
resp = &types.AlertRulesResp{}
l.svcCtx.DbEngin.Raw("SELECT ar.id,ar.*,GROUP_CONCAT(tc.`name` ORDER BY tc.`name` ASC SEPARATOR ',') as cluster_name FROM alert_rule ar JOIN t_cluster tc ON ar.cluster_id = tc.id WHERE ar.deleted_at IS NULL AND tc.deleted_at IS NULL GROUP BY ar.id;").Scan(&resp)
var alertRules []types.AlertRule
l.svcCtx.DbEngin.Raw("SELECT ar.id,ar.*,GROUP_CONCAT(tc.`name` ORDER BY tc.`name` ASC SEPARATOR ',') as cluster_name FROM alert_rule ar JOIN t_cluster tc ON ar.cluster_id = tc.id WHERE ar.deleted_at IS NULL AND tc.deleted_at IS NULL GROUP BY ar.id").Scan(&alertRules)
resp.AlertRules = alertRules
return resp, nil
}

View File

@ -38,7 +38,7 @@ func (l *NodesLoadTopLogic) NodesLoadTop(req *types.NodesLoadTopReq) (resp *type
ForceContentType("application/json").
Get(server + "/api/v1/monitoring/node")
if err != nil || response.IsError() {
return nil, err
}
return resp, nil
}

View File

@ -5316,6 +5316,123 @@ type AiAlgorithmsResp struct {
Algorithms []string `json:"algorithms"`
}
type PullTaskInfoReq struct {
AdapterId int64 `form:"adapterId"`
}
type PullTaskInfoResp struct {
HpcInfoList []*HpcInfo `json:"HpcInfoList,omitempty"`
CloudInfoList []*CloudInfo `json:"CloudInfoList,omitempty"`
AiInfoList []*AiInfo `json:"AiInfoList,omitempty"`
VmInfoList []*VmInfo `json:"VmInfoList,omitempty"`
}
type HpcInfo struct {
Id int64 `json:"id"` // id
TaskId int64 `json:"task_id"` // 任务id
JobId string `json:"job_id"` // 作业id(在第三方系统中的作业id)
AdapterId int64 `json:"adapter_id"` // 执行任务的适配器id
ClusterId int64 `json:"cluster_id"` // 执行任务的集群id
ClusterType string `json:"cluster_type"` // 执行任务的集群类型
Name string `json:"name"` // 名称
Status string `json:"status"` // 状态
CmdScript string `json:"cmd_script"`
StartTime string `json:"start_time"` // 开始时间
RunningTime int64 `json:"running_time"` // 运行时间
DerivedEs string `json:"derived_es"`
Cluster string `json:"cluster"`
BlockId int64 `json:"block_id"`
AllocNodes int64 `json:"alloc_nodes"`
AllocCpu int64 `json:"alloc_cpu"`
CardCount int64 `json:"card_count"` // 卡数
Version string `json:"version"`
Account string `json:"account"`
WorkDir string `json:"work_dir"` // 工作路径
AssocId int64 `json:"assoc_id"`
ExitCode int64 `json:"exit_code"`
WallTime string `json:"wall_time"` // 最大运行时间
Result string `json:"result"` // 运行结果
DeletedAt string `json:"deleted_at"` // 删除时间
YamlString string `json:"yaml_string"`
AppType string `json:"app_type"` // 应用类型
AppName string `json:"app_name"` // 应用名称
Queue string `json:"queue"` // 队列名称
SubmitType string `json:"submit_type"` // cmd命令行模式
NNode string `json:"n_node"` // 节点个数当指定该参数时GAP_NODE_STRING必须为""
StdOutFile string `json:"std_out_file"` // 工作路径/std.err.%j
StdErrFile string `json:"std_err_file"` // 工作路径/std.err.%j
StdInput string `json:"std_input"`
Environment string `json:"environment"`
DeletedFlag int64 `json:"deleted_flag"` // 是否删除0-否1-是)
CreatedBy int64 `json:"created_by"` // 创建人
CreatedTime string `json:"created_time"` // 创建时间
UpdatedBy int64 `json:"updated_by"` // 更新人
UpdatedTime string `json:"updated_time"` // 更新时间
}
type CloudInfo struct {
Participant int64 `json:"participant,omitempty"`
Id int64 `json:"id,omitempty"`
TaskId int64 `json:"taskId,omitempty"`
ApiVersion string `json:"apiVersion,omitempty"`
Kind string `json:"kind,omitempty"`
Namespace string `json:"namespace,omitempty"`
Name string `json:"name,omitempty"`
Status string `json:"status,omitempty"`
StartTime string `json:"startTime,omitempty"`
RunningTime int64 `json:"runningTime,omitempty"`
Result string `json:"result,omitempty"`
YamlString string `json:"yamlString,omitempty"`
}
type AiInfo struct {
ParticipantId int64 `json:"participantId,omitempty"`
TaskId int64 `json:"taskId,omitempty"`
ProjectId string `json:"project_id,omitempty"`
Name string `json:"name,omitempty"`
Status string `json:"status,omitempty"`
StartTime string `json:"startTime,omitempty"`
RunningTime int64 `json:"runningTime,omitempty"`
Result string `json:"result,omitempty"`
JobId string `json:"jobId,omitempty"`
CreateTime string `json:"createTime,omitempty"`
ImageUrl string `json:"imageUrl,omitempty"`
Command string `json:"command,omitempty"`
FlavorId string `json:"flavorId,omitempty"`
SubscriptionId string `json:"subscriptionId,omitempty"`
ItemVersionId string `json:"itemVersionId,omitempty"`
}
type VmInfo struct {
ParticipantId int64 `json:"participantId,omitempty"`
TaskId int64 `json:"taskId,omitempty"`
Name string `json:"name,omitempty"`
FlavorRef string `json:"flavor_ref,omitempty"`
ImageRef string `json:"image_ref,omitempty"`
NetworkUuid string `json:"network_uuid,omitempty"`
BlockUuid string `json:"block_uuid,omitempty"`
SourceType string `json:"source_type,omitempty"`
DeleteOnTermination bool `json:"delete_on_termination,omitempty"`
State string `json:"state,omitempty"`
}
type PushTaskInfoReq struct {
AdapterId int64 `json:"adapterId"`
HpcInfoList []*HpcInfo `json:"hpcInfoList"`
CloudInfoList []*CloudInfo `json:"cloudInfoList"`
AiInfoList []*AiInfo `json:"aiInfoList"`
VmInfoList []*VmInfo `json:"vmInfoList"`
}
type PushTaskInfoResp struct {
Code int64 `json:"code"`
Msg string `json:"msg"`
}
type PushResourceInfoReq struct {
AdapterId int64 `json:"adapterId"`
}
type CreateAlertRuleReq struct {
CLusterId int64 `json:"clusterId"`
ClusterName string `json:"clusterName"`
@ -5328,6 +5445,10 @@ type CreateAlertRuleReq struct {
}
type AlertRulesResp struct {
AlertRules []AlertRule `json:"alertRules"`
}
type AlertRule struct {
Id int64 `json:"id"`
ClusterName string `json:"clusterName"`
Name string `json:"name"`
@ -5343,5 +5464,7 @@ type NodesLoadTopReq struct {
}
type NodesLoadTopResp struct {
Code int `json:"code"`
Data interface{} `json:"data"`
Msg string `json:"msg"`
}