hardware: GPU page shows frozen samples during an nvidia-smi outage #70

Open
opened 2026-09-22 11:40:22 +02:00 by julian · 0 comments
Owner

Raised in the review of #62.

While the nvidia-smi probe is failing, the last written GPU samples (writeGpu) stay in place and nothing marks them stale. Before #62 that meant frozen values until the container restarted; with the backoff it means frozen values for the length of the outage (a driver upgrade, a module reload, a card falling off the bus), with retries up to a minute apart.

Question to decide: what should the GPU page show during an outage?

  • clear the samples on the first backed-off failure (card stays listed, metrics empty)
  • keep the samples but attach a staleSince / updatedAt so the UI can grey them out
  • leave as is

Deliberately left out of #62; the fix there is about the metrics dying, not about how an outage is displayed.

Raised in the review of #62. While the nvidia-smi probe is failing, the last written GPU samples (`writeGpu`) stay in place and nothing marks them stale. Before #62 that meant frozen values until the container restarted; with the backoff it means frozen values for the length of the outage (a driver upgrade, a module reload, a card falling off the bus), with retries up to a minute apart. Question to decide: what should the GPU page show during an outage? - clear the samples on the first backed-off failure (card stays listed, metrics empty) - keep the samples but attach a `staleSince` / `updatedAt` so the UI can grey them out - leave as is Deliberately left out of #62; the fix there is about the metrics dying, not about how an outage is displayed.
Sign in to join this conversation.
No labels
No milestone
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
OpsDeck/core#70
No description provided.