ModifyTrainingDataTraceSwitch

Description

Modifies the dynamic dotting switch of various data, which is called externally.

If lightweight profiling is enabled to obtain flushed data through the gRPC interface provided by ClusterD, the lifecycle of the created data-trace-<Job name> ConfigMap is deleted with the job. If the job does not exist, this API fails to be called.

Prototype

rpc ModifyTrainingDataTraceSwitch (DataTypeReq) returns (DataTypeRes)

Input Parameters

Parameter

Type (Defined by Protobuf)

Description

DataTypeReq

message DataTypeReq{

string jobNsName = 1;

ProfilingSwitch profilingSwitch = 2;

}

message ProfilingSwitch{

string CommunicationOperator = 1;

string Step = 2;

string SaveCheckpoint = 3;

string FP =4;

string DataLoader =5;

}

jobNsName: job namespace and name to be modified, separated by a slash (/), for example, default/test-pytorch.

profilingSwitch: details of each switch
  • CommunicationOperator: communication operator
  • Step: step latency
  • SaveCheckpoint: time taken by saveCheckpoint
  • FP: forward propagation data
  • DataLoader: time taken by DataLoader

Return Value

Parameter

Type (Defined by Protobuf)

Description

DataTypeRes

message DataTypeRes{

string message = 1;

int32 code = 2;

}

message: API calling result

code: API calling return code

  • 1 (300): invalid input parameter
  • 2 (404): ConfigMap not queried
  • 3 (500): server error
  • 4 (200): normal response