Skip to content

Profiling metrics stop getting collected after some time #635

@madiyar-wayve

Description

@madiyar-wayve

What is the version?

4.4.1-4.6.0

What happened?

DCGM exporter pods stop collecting profiling metrics, e.g. DCGM_FI_PROF_SM_ACTIVE, DCGM_FI_PROF_DRAM_ACTIVE after running for a while (usually around after 5-6 days). Restarting the pods mitigates the problem.

Error logs from nv-hostengine.log:

2026-02-13 19:48:54.762 ERROR [1:38] Got nvmlSt 3 for field 139 from nvmlDeviceGetFieldValues [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:13381] [DcgmCacheManager::ReadAndCacheNvLinkBandwidth]
2026-02-13 19:48:54.762 ERROR [1:38] Got nvmlSt 3 for field 138 from nvmlDeviceGetFieldValues [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:13381] [DcgmCacheManager::ReadAndCacheNvLinkBandwidth]
2026-02-13 19:48:55.221 ERROR [1:54] GetLatestSample returned Feature not supported for entityId 0 groupId 1 fieldId 449 [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:4709] [DcgmCacheManager::GetMultipleLatestSamples]
2026-02-13 19:49:02.002 ERROR [1:37] [[NvSwitch]] Not attached to NvSwitches. Aborting [/builds/dcgm/dcgm/modules/nvswitch/DcgmNscqManager.cpp:680] [DcgmNs::DcgmNscqManager::ReadNvSwitchStatusAllSwitches]
2026-02-13 19:49:10.219 ERROR [1:1] GetLatestSample returned Feature not supported for entityId 0 groupId 1 fieldId 449 [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:4709] [DcgmCacheManager::GetMultipleLatestSamples]
2026-02-13 19:49:24.762 ERROR [1:38] Got nvmlSt 3 for field 139 from nvmlDeviceGetFieldValues [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:13381] [DcgmCacheManager::ReadAndCacheNvLinkBandwidth]
2026-02-13 19:49:24.762 ERROR [1:38] Got nvmlSt 3 for field 138 from nvmlDeviceGetFieldValues [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:13381] [DcgmCacheManager::ReadAndCacheNvLinkBandwidth]
2026-02-13 19:49:25.220 ERROR [1:53] GetLatestSample returned Feature not supported for entityId 0 groupId 1 fieldId 449 [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:4709] [DcgmCacheManager::GetMultipleLatestSamples]
2026-02-13 19:49:32.002 ERROR [1:37] [[NvSwitch]] Not attached to NvSwitches. Aborting [/builds/dcgm/dcgm/modules/nvswitch/DcgmNscqManager.cpp:680] [DcgmNs::DcgmNscqManager::ReadNvSwitchStatusAllSwitches]
2026-02-13 19:49:40.221 ERROR [1:61] GetLatestSample returned Feature not supported for entityId 0 groupId 1 fieldId 449 [/builds/dcgm/dcgm/dcgmlib/src/DcgmCacheManager.cpp:4709] [DcgmCacheManager::GetMultipleLatestSamples]

What did you expect to happen?

Long running DCGM exporters should keep collecting profiling metrics.

What is the GPU model?

No response

What is the environment?

No response

How did you deploy the dcgm-exporter and what is the configuration?

Deployed via helm-chart

How to reproduce the issue?

No response

Anything else we need to know?

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions