Skip to content

Fix Xid healthcheck for replicated GPUs - #1992

Open
abraaoteixeira wants to merge 1 commit into
NVIDIA:mainfrom
abraaoteixeira:fix-xid-health-check
Open

Fix Xid healthcheck for replicated GPUs#1992
abraaoteixeira wants to merge 1 commit into
NVIDIA:mainfrom
abraaoteixeira:fix-xid-health-check

Conversation

@abraaoteixeira

Copy link
Copy Markdown

Fixes #1929

This PR ensures that all logical devices (such as Time-Slicing or MPS replicas) sharing a physical GPU are properly marked as unhealthy when a critical Xid error occurs on the underlying hardware, instead of just the last registered replica.

(The Signed-off-by is already included in the commit message).

When a physical GPU experiences a critical Xid error, the device plugin
needs to mark all associated logical devices (Time-Slicing, MPS) as unhealthy.
Previously, the parentToDeviceMap was overwritten by replicated logical devices,
causing only one of them to be marked unhealthy on an Xid event.

This patch changes the mapping to map[string][]*Device so all logical
devices bound to the failing physical UUID are properly marked.

Fixes NVIDIA#1929

Signed-off-by: Abraão Teixeira <abraaoteixeira0101@gmail.com>
@copy-pr-bot

copy-pr-bot Bot commented Aug 30, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug]: Health check misses devices that share a physical GPU

1 participant