Redirect log message to stderr in nvidia runtime wrapper script#1400
Merged
cdesiniotis merged 1 commit intoNVIDIA:mainfrom Nov 3, 2025
Merged
Conversation
This change is required to make our nvidia runtime wrapper compliant with the OCI runtime spec. All OCI-compliant runtimes must support the operations documented at https://github.com/opencontainers/runtime-spec/blob/v1.2.1/runtime.md#operations. Before this change, our nvidia runtime wrapper was not producing the expected output when the query state operation (`state <container-id>`) was invoked AND the nvidia kernel modules happened to not be loaded. In this case, we were emitting an extra log message which caused the stdout of this command to not adhere to the schema defined in the OCI runtime spec. Redirecting the log message to stderr makes us compliant. This issue was discovered when deploying GPU Operator 25.10.0 on nodes using cri-o. GPU Operator 25.10.0 is the first release that installs nvidia runtime handlers with cri-o by default, as opposed to installing an OCI hook file. When performing a GPU driver upgrade, pods in the gpu-operator namespace would be in the `Init:RunContainerError` state for several minutes until the new driver finished installing -- note that no nvidia driver modules are loaded during this span of several minutes. When inspecting the cri-o logs, we observed the following error message: ``` level=warning msg="Error updating the container status \"16779f4cd2414a164aae56856b491f86fe0c6b803a3b4474ada2cc0864c8e028\": failed to decode container status for 16779f4cd2414a164aae56856b491f86fe0c6b803a3b4474ada2cc0864c8e028: skipThreeBytes: expect ull, error found in NVIDIA#2 byte of ...|nvidia drive|..., bigger context ...|nvidia driver modules are not yet loaded, invoking /|..." id=a4b48041-edc4-48c2-8d75-4ad03cb3d8e1 name=/runtime.v1.RuntimeService/CreateContainer ``` This error message indicates cri-o failed to get the status of the container because it could not decode the JSON returned by the runtime handler. Signed-off-by: Christopher Desiniotis <cdesiniotis@nvidia.com>
6f7fb59 to
61f9bde
Compare
Member
|
/cherry-pick release-1.18 |
tariq1890
approved these changes
Nov 1, 2025
|
🤖 Backport PR created for |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This change is required to make our nvidia runtime wrapper compliant with the OCI runtime spec. All OCI-compliant runtimes must support the operations documented at https://github.com/opencontainers/runtime-spec/blob/v1.2.1/runtime.md#operations. Before this change, our nvidia runtime wrapper was not producing the expected output when the query state operation (
state <container-id>) was invoked AND the nvidia kernel modules happened to not be loaded. In this case, we were emitting an extra log message which caused the stdout of this command to not adhere to the schema defined in the OCI runtime spec. Redirecting the log message to stderr makes us compliant.This issue was discovered when deploying GPU Operator 25.10.0 on nodes using cri-o. GPU Operator 25.10.0 is the first release that installs nvidia runtime handlers with cri-o by default, as opposed to installing an OCI hook file. When performing a GPU driver upgrade, pods in the gpu-operator namespace would be in the
Init:RunContainerErrorstate for several minutes until the new driver finished installing -- note that no nvidia driver modules are loaded during this span of several minutes. When inspecting the cri-o logs, we observed the following error message:This error message indicates cri-o failed to get the status of the container because it could not decode the JSON returned by the runtime handler.