SubscribeJobSummarySignal

ClientInfo

message ClientInfo{

string role = 1;

string clientId = 3;

}

ClientInfo.role: client role

ClientInfo.clientId: client ID

stream

gRPC stream

This API returns a gRPC stream. (The data structure of the return value is based on the programming language selected by the client.)

The client can call the stream's Receive method (the actual name is determined by the client's programming language) to receive data pushed by the server.

JobSummarySignal

message JobSummarySignal{

string uuid = 1;

string jobId = 2;

string jobName = 3;

string namespace =4;

string frameWork = 5;

string jobStatus = 6;

string time = 7;

string cmIndex = 8;

string total = 9;

string HcclJson = 10;

string deleteTime = 11;

string sharedTorIp = 12;

string masterAddr = 13;

string operator = 14;

}

uuid: message ID

jobId: Kubernetes ID of a job

jobName: job name

namespace: job namespace

frameWork: framework

jobStatus: job status, including:

• pending
• running
• complete
• failed

time: time when a job starts

cmIndex: index

total: total number of jobsummary ConfigMaps

HcclJson: communication information about the processor used by a job

The value can be escaped to the JSON format. The fields are described as follows:

• status: whether RankTable has been generated.
• initializing: Devices are being allocated to a job, and RankTable is not generated.
• complete: After RankTable is generated, the status changes to complete immediately, and fields such as server_list are displayed synchronously.
• server_list: device allocation information
• device: NPU allocation information, including the NPU IP address, and rank ID.
• server_id: AI server ID, which is globally unique.
• server_name: server name
• server_sn: server SN Ensure that the server SN exists. If it does not exist, contact Huawei technical support.
• server_count: number of servers used by a job
• version: version information

deleteTime: time when a job is deleted

sharedTorIp: information about the shared switch used by a job

masterAddr: MASTER_ADDR value specified during PyTorch training

operator: After receiving the command for adding a job, the status is updated to add. After receiving the command for deleting a job, the status is updated to delete.