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

hpc新增字段 

Former-commit-id: 5958b28a8285b807d4cf693d64e9cf313be41235
This commit is contained in:
zhangwei 2024-09-04 16:56:22 +08:00
parent 37afd2a284
commit c9a754e79b
5 changed files with 123 additions and 127 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

225
desc/hpc/pcm-hpc.api
View File

@ -1,137 +1,138 @@
syntax = "v1"
info(
title: "type title here"
desc: "type desc here"
author: "type author here"
email: "type email here"
version: "type version here"
title: "type title here"
desc: "type desc here"
author: "type author here"
email: "type email here"
version: "type version here"
)
type (
commitHpcTaskReq {
Name string `json:"name"` // paratera:jobName
Description string `json:"description,optional"`
TenantId int64 `json:"tenantId,optional"`
TaskId int64 `json:"taskId,optional"`
AdapterIds []string `json:"adapterIds"`
MatchLabels map[string]string `json:"matchLabels,optional"`
CardCount int64 `json:"cardCount,optional"`
WorkDir string `json:"workDir,optional"` //paratera:workingDir
WallTime string `json:"wallTime,optional"`
CmdScript string `json:"cmdScript,optional"` // paratera:bootScript
AppType string `json:"appType,optional"`
AppName string `json:"appName,optional"` // paratera:jobGroupName ac:appname
Queue string `json:"queue,optional"`
NNode string `json:"nNode,optional"`
SubmitType string `json:"submitType,optional"`
StdOutFile string `json:"stdOutFile,optional"`
StdErrFile string `json:"stdErrFile,optional"`
StdInput string `json:"stdInput,optional"`
Environment map[string]string `json:"environment,optional"`
ClusterType string `json:"clusterType,optional"`
}
commitHpcTaskReq {
Name string `json:"name"` // paratera:jobName
Description string `json:"description,optional"`
TenantId int64 `json:"tenantId,optional"`
TaskId int64 `json:"taskId,optional"`
AdapterIds []string `json:"adapterIds"`
MatchLabels map[string]string `json:"matchLabels,optional"`
CardCount int64 `json:"cardCount,optional"`
WorkDir string `json:"workDir,optional"` //paratera:workingDir
WallTime string `json:"wallTime,optional"`
CmdScript string `json:"cmdScript,optional"` // paratera:bootScript
AppType string `json:"appType,optional"`
AppName string `json:"appName,optional"` // paratera:jobGroupName ac:appname
Queue string `json:"queue,optional"`
NNode string `json:"nNode,optional"`
SubmitType string `json:"submitType,optional"`
StdOutFile string `json:"stdOutFile,optional"`
StdErrFile string `json:"stdErrFile,optional"`
StdInput string `json:"stdInput,optional"`
Environment map[string]string `json:"environment,optional"`
ClusterType string `json:"clusterType,optional"`
Partition string `json:"partition"`
}
commitHpcTaskResp {
TaskId int64 `json:"taskId"`
Code int32 `json:"code"`
Msg string `json:"msg"`
}
commitHpcTaskResp {
TaskId int64 `json:"taskId"`
Code int32 `json:"code"`
Msg string `json:"msg"`
}
)
type (
hpcOverViewReq {
}
hpcOverViewResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data HPCOverView `json:"data"`
}
HPCOverView {
AdapterCount int32 `json:"adapterCount"`
StackCount int32 `json:"stackCount"`
ClusterCount int32 `json:"clusterCount"`
TaskCount int32 `json:"taskCount"`
}
hpcOverViewReq {
}
hpcOverViewResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data HPCOverView `json:"data"`
}
HPCOverView {
AdapterCount int32 `json:"adapterCount"`
StackCount int32 `json:"stackCount"`
ClusterCount int32 `json:"clusterCount"`
TaskCount int32 `json:"taskCount"`
}
)
type (
hpcAdapterSummaryReq {
}
hpcAdapterSummaryResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data []HPCAdapterSummary `json:"data"`
}
HPCAdapterSummary {
AdapterName string `json:"adapterName"`
StackCount int32 `json:"stackCount"`
ClusterCount int32 `json:"clusterCount"`
TaskCount int32 `json:"taskCount"`
}
hpcAdapterSummaryReq {
}
hpcAdapterSummaryResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data []HPCAdapterSummary `json:"data"`
}
HPCAdapterSummary {
AdapterName string `json:"adapterName"`
StackCount int32 `json:"stackCount"`
ClusterCount int32 `json:"clusterCount"`
TaskCount int32 `json:"taskCount"`
}
)
type (
hpcJobReq {
}
hpcJobResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data []Job `json:"data"`
}
Job {
JobName string `json:"jobName"`
JobDesc string `json:"jobDesc"`
SubmitTime string `json:"submitTime"`
JobStatus string `json:"jobStatus"`
AdapterName string `json:"adapterName"`
ClusterName string `json:"clusterName"`
ClusterType string `json:"clusterType"`
}
hpcJobReq {
}
hpcJobResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data []Job `json:"data"`
}
Job {
JobName string `json:"jobName"`
JobDesc string `json:"jobDesc"`
SubmitTime string `json:"submitTime"`
JobStatus string `json:"jobStatus"`
AdapterName string `json:"adapterName"`
ClusterName string `json:"clusterName"`
ClusterType string `json:"clusterType"`
}
)
type (
hpcResourceReq {
}
hpcResourceResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data HPCResource `json:"data"`
}
HPCResource {
GPUCardsTotal float64 `json:"gpuCoresTotal"`
CPUCoresTotal float64 `json:"cpuCoresTotal"`
RAMTotal float64 `json:"ramTotal"`
GPUCardsUsed float64 `json:"gpuCoresUsed"`
CPUCoresUsed float64 `json:"cpuCoresUsed"`
RAMUsed float64 `json:"ramUsed"`
GPURate float64 `json:"gpuRate"`
CPURate float64 `json:"cpuRate"`
RAMRate float64 `json:"ramRate"`
}
hpcResourceReq {
}
hpcResourceResp {
Code int32 `json:"code"`
Msg string `json:"msg"`
Data HPCResource `json:"data"`
}
HPCResource {
GPUCardsTotal float64 `json:"gpuCoresTotal"`
CPUCoresTotal float64 `json:"cpuCoresTotal"`
RAMTotal float64 `json:"ramTotal"`
GPUCardsUsed float64 `json:"gpuCoresUsed"`
CPUCoresUsed float64 `json:"cpuCoresUsed"`
RAMUsed float64 `json:"ramUsed"`
GPURate float64 `json:"gpuRate"`
CPURate float64 `json:"cpuRate"`
RAMRate float64 `json:"ramRate"`
}
)
type QueueAssetsResp {
QueueAssets []QueueAsset `json:"queueAsset"`
QueueAssets []QueueAsset `json:"queueAsset"`
}
type QueueAsset {
TenantName string `json:"tenantName"` //租户名称
ParticipantId int64 `json:"participantId"`
AclHosts string `json:"aclHosts"` // 可用节点，多个节点用逗号隔开
QueNodes string `json:"queNodes"` //队列节点总数
QueMinNodect string `json:"queMinNodect,omitempty"` //队列最小节点数
QueMaxNgpus string `json:"queMaxNgpus,omitempty"` //队列最大GPU卡数
QueMaxPPN string `json:"queMaxPPN,omitempty"` //使用该队列作业最大CPU核心数
QueChargeRate string `json:"queChargeRate,omitempty"` //费率
QueMaxNcpus string `json:"queMaxNcpus,omitempty"` //用户最大可用核心数
QueMaxNdcus string `json:"queMaxNdcus,omitempty"` //队列总DCU卡数
QueueName string `json:"queueName,omitempty"` //队列名称
QueMinNcpus string `json:"queMinNcpus,omitempty"` //队列最小CPU核数
QueFreeNodes string `json:"queFreeNodes,omitempty"` //队列空闲节点数
QueMaxNodect string `json:"queMaxNodect,omitempty"` //队列作业最大节点数
QueMaxGpuPN string `json:"queMaxGpuPN,omitempty"` //队列单作业最大GPU卡数
QueMaxWalltime string `json:"queMaxWalltime,omitempty"` //队列最大运行时间
QueMaxDcuPN string `json:"queMaxDcuPN,omitempty"` //队列单作业最大DCU卡数
QueFreeNcpus string `json:"queFreeNcpus"` //队列空闲cpu数
QueNcpus string `json:"queNcpus"` //队列cpu数
TenantName string `json:"tenantName"` //租户名称
ParticipantId int64 `json:"participantId"`
AclHosts string `json:"aclHosts"` // 可用节点，多个节点用逗号隔开
QueNodes string `json:"queNodes"` //队列节点总数
QueMinNodect string `json:"queMinNodect,omitempty"` //队列最小节点数
QueMaxNgpus string `json:"queMaxNgpus,omitempty"` //队列最大GPU卡数
QueMaxPPN string `json:"queMaxPPN,omitempty"` //使用该队列作业最大CPU核心数
QueChargeRate string `json:"queChargeRate,omitempty"` //费率
QueMaxNcpus string `json:"queMaxNcpus,omitempty"` //用户最大可用核心数
QueMaxNdcus string `json:"queMaxNdcus,omitempty"` //队列总DCU卡数
QueueName string `json:"queueName,omitempty"` //队列名称
QueMinNcpus string `json:"queMinNcpus,omitempty"` //队列最小CPU核数
QueFreeNodes string `json:"queFreeNodes,omitempty"` //队列空闲节点数
QueMaxNodect string `json:"queMaxNodect,omitempty"` //队列作业最大节点数
QueMaxGpuPN string `json:"queMaxGpuPN,omitempty"` //队列单作业最大GPU卡数
QueMaxWalltime string `json:"queMaxWalltime,omitempty"` //队列最大运行时间
QueMaxDcuPN string `json:"queMaxDcuPN,omitempty"` //队列单作业最大DCU卡数
QueFreeNcpus string `json:"queFreeNcpus"` //队列空闲cpu数
QueNcpus string `json:"queNcpus"` //队列cpu数
}

22
etc/pcm.yaml
View File

@ -18,14 +18,12 @@ Monitoring:
#rpc
THRpcConf:
target: nacos://10.206.0.12:8848/pcm.th.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
# Endpoints:
# - 127.0.0.1:8888
NonBlock: true
#rpc
ModelArtsRpcConf:
# target: nacos://10.206.0.12:8848/pcm.modelarts.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
Endpoints:
- 127.0.0.1:2002
NonBlock: true
@ -33,36 +31,31 @@ ModelArtsRpcConf:
#rpc
ModelArtsImgRpcConf:
target: nacos://10.206.0.12:8848/pcm.modelarts.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
# Endpoints:
# - 127.0.0.1:2002
NonBlock: true
Endpoints:
- 127.0.0.1:2007
NonBlock: true
#rpc
ACRpcConf:
target: nacos://10.206.0.12:8848/pcm.ac.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
# Endpoints:
# - 127.0.0.1:8888
Endpoints:
- 127.0.0.1:8888
NonBlock: true
Timeout: 50000
#rpc
CephRpcConf:
# target: nacos://10.206.0.12:8848/pcm.ceph.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
Endpoints:
- pcm-participant-ceph-service:2008
NonBlock: true
Timeout: 50000
OctopusRpcConf:
target: nacos://10.206.0.12:8848/pcm.octopus.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
# Endpoints:
# - 127.0.0.1:8888
Endpoints:
- 127.0.0.1:8888
NonBlock: true
Timeout: 20000
OpenstackRpcConf:
# target: nacos://10.206.0.12:8848/pcm.openstack.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
Endpoints:
- 127.0.0.1:2010
NonBlock: true
@ -70,7 +63,6 @@ OpenstackRpcConf:
# core rpc
PcmCoreRpcConf:
# target: nacos://10.206.0.12:8848/pcm.core.rpc?timeout=30s&namespaceid=test&groupname=DEFAULT_GROUP&appName=pcm.core.api
Endpoints:
- pcm-core-rpc:2004
NonBlock: true

1
internal/logic/hpc/commithpctasklogic.go
View File

@ -90,6 +90,7 @@ func (l *CommitHpcTaskLogic) CommitHpcTask(req *types.CommitHpcTaskReq) (resp *t
StdOutFile: req.StdOutFile,
StdErrFile: req.StdErrFile,
StdInput: req.StdInput,
Partition: req.Partition,
DeletedFlag: 0,
CreatedBy: 0,
CreatedTime: time.Now(),

1
internal/types/types.go
View File

@ -1242,6 +1242,7 @@ type CommitHpcTaskReq struct {
StdInput string `json:"stdInput,optional"`
Environment map[string]string `json:"environment,optional"`
ClusterType string `json:"clusterType,optional"`
Partition string `json:"partition"`
}
type CommitHpcTaskResp struct {

1
pkg/models/taskhpcmodel_gen.go
View File

@ -71,6 +71,7 @@ type (
StdOutFile string `db:"std_out_file"` // 工作路径/std.err.%j
StdErrFile string `db:"std_err_file"` // 工作路径/std.err.%j
StdInput string `db:"std_input"`
Partition string `db:"partition"`
Environment string `db:"environment"`
DeletedFlag int64 `db:"deleted_flag"` // 是否删除（0-否，1-是）
CreatedBy int64 `db:"created_by"` // 创建人