Wrong return of TestName in DCGM Diag Run

[g-rayban]

When results, err := dcgm.RunDiag(dcgm.DiagQuick, dcgm.GroupAllGPUs()) is called and we print the results, the following is the output

{Software:[{Status:pass TestName:presence of drivers on the denylist (e.g. nouveau) TestOutput:Allocated 83618558100 bytes (98.4%) ErrorCode:0 ErrorMessage:} {Status:pass TestName:presence of drivers on the denylist (e.g. nouveau) TestOutput:Allocated 83618558100 bytes (98.4%) ErrorCode:0 ErrorMessage:} {Status:pass TestName:presence of drivers on the denylist (e.g. nouveau) TestOutput:Allocated 83618558100 bytes (98.4%) ErrorCode:0 ErrorMessage:}]}

Should not the TestName be "software", "memory" and "pcie" the way it's displayed in dcgmi command. I also see an used function gpuTestName in diag.go which should be the ideal testname.

dcgmi diag -r 2
Successfully ran diagnostic for group. +---------------------------+------------------------------------------------+ | Diagnostic | Result | +===========================+================================================+ |----- Metadata ----------+------------------------------------------------| | DCGM Version | 4.4.1 | | Driver Version Detected | 580.76.05 | | GPU Device IDs Detected | 2330 | |----- Deployment --------+------------------------------------------------| | software | Pass | | | GPU0: Pass | +----- Hardware ----------+------------------------------------------------+ | memory | Pass | | | GPU0: Pass | +----- Integration -------+------------------------------------------------+ | pcie | Pass | | | GPU0: Pass | +---------------------------+------------------------------------------------+

[g-rayban]

A kind reminder to look into this
@nvvfedorov may have an idea on this as gpuTestName was introduced in commit

[g-rayban]

The output of diagRun and the command dcgmi diag -r should match ideally

[g-rayban]

Even with changes in commit the info is still coming partial, and component name is missing

[nvvfedorov]

Thank you for bringing the issue to my attention. We are working on the fix.

[g-rayban]

Can we also look to pass complete info instead of truncating it and just returning the first info.

E.g. Expected output using my suggested changes for diag r -2: {Status:pass TestName: memory/memory TestOutput: | Allocated 83618558100 bytes (98.4%) ErrorCode:0 ErrorMessage: SerialNumber:1652923033635 EntityID:0} {Status:pass TestName:Integration/pcie/pcie TestOutput: | GPU to Host bandwidth: 28.27 GB/s | Host to GPU bandwidth: 27.65 GB/s | bidirectional bandwidth: 50.59 GB/s | GPU to Host latency: 1.305 us | Host to GPU latency: 2.097 us | bidirectional latency: 2.666 us ErrorCode:0 ErrorMessage: SerialNumber:1652923033635 EntityID:0}]}

Current output: {Status:pass TestName:memory/memory TestOutput:Allocated 83618558100 bytes (98.4%) ErrorCode:0 ErrorMessage: SerialNumber:1652923033635 EntityID:0} {Status:pass TestName:pcie/pcie TestOutput:GPU to Host bandwidth: 28.60 GB/s ErrorCode:0 ErrorMessage: SerialNumber:1652923033635 EntityID:0}]}

Suggest something similar to below to fix info if possible:

func getInfoMsg(entityId uint, testId uint, response C.dcgmDiagResponse_v12) string {
        msg := ""
        for i := 0; i < int(response.numInfo); i++ {
                if uint(response.info[i].entity.entityId) != entityId || uint(response.info[i].testId) != testId {
                        continue
                }

                msg += " | " + C.GoString((*C.char)(unsafe.Pointer(&response.info[i].msg)))
                // return msg
        }

        return msg
}

[nvvfedorov]

@g-rayban, please create a separate issue for the request with output change.

[nvvfedorov]

@g-rayban, The change was small. Here is the MR: #101.

To make it easy to track requests and changes, please try to create one issue per request to change.

[g-rayban]

Ack on the request for a separate issue. Will do that from now onwards. Thanks for taking up the previous one in the same. We can close this issue with the PR merged.