Skip to content

[BUG] agent: symbol lookup error: /opt/datadog-agent/bin/agent/agent: undefined symbol: nvmlVgpuTypeGetCapabilities #32419

Open
@rawlingsj

Description

@rawlingsj

Agent Environment
datadog-agent version 7.60.1
Linux

Describe what happened:

After compiling from source I've not been able to run the agent:

/opt/datadog-agent/bin/agent/agent: symbol lookup error: /opt/datadog-agent/bin/agent/agent: undefined symbol: nvmlVgpuTypeGetCapabilities

It looks like there's a relatively recent go mod dependency on go-nvml, then used here. AFAICT this is wired in at compile time so adding any runtime config to disable the gpu monitor doesn't seem to work.

Describe what you expected:

For the gpu monitor to be optional if nvidia libraries are not available.

Steps to reproduce the issue:

We're building the agent using

invoke -e agent.build \
        --bundle process-agent \
        --bundle trace-agent \
        --bundle system-probe \
        --bundle security-agent \
        --exclude-rtloader \
        --no-development \
        --bundle-ebpf \
        --embedded-path /usr/lib

Additional environment details (Operating System, Cloud provider, etc):

This is a package build built on Wolfi OS using Melange, the build pipeline is a little complex but if more context is wanted this may help... https://github.com/wolfi-dev/os/blob/f4727bc/datadog-agent.yaml

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions