Repository navigation
feat(observability): record which GPUs each Workload holds - #393
Merged
Merged
Conversation
droot
approved these changes
Oct 29, 2026
This was referenced Oct 29, 2026
The scheduler now records which GPUs a Workload holds and when. It resolves the claim's allocated devices to the UUIDs their ResourceSlices publish, sets placedAt when it sets nodeName, and sets releasedAt when the node clears, the claim changes or teardown frees the seat. The API server can then tell which run held which GPU from Workload status alone.
The inventory now has placements: each Workload that holds GPUs, with its node, pod, GPU UUIDs, placement time and the runs it serves, read live from Workload status. When the API server releases a run's workers, a background thread saves each Workload's last placement to open_rl:placements:<runtime> so it outlives the Workload. Nothing reads that history yet; the activity view will.
ShubyM
force-pushed
the
observability/placements
branch
from
October 30, 2026 02:33
1404076 to
2aa061a
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
Phase 4 of #390. Nothing recorded which GPUs a run's workers held, or when. The scheduler knew, but the record disappeared when the Workload was deleted.
What changed
gpuUUIDs,placedAtandreleasedAton each Workload's status./api/v1/inspect/inventorygainsplacements: one entry per Workload holding GPUs now, with its runs, role, node, pod and GPU UUIDs. It reads Workload status, so a new placement shows within seconds.open_rl:placements:<runtime>, newest 200, 7 days). The write runs in the background and cannot delay the release. Nothing reads it yet;/activitywill in phase 5.Scope
openrl_gpu_heldmetric.runtime_of'sGET, which could hang forever before.Verification
make testandmake lintpass. A Go test checks a placed Workload gets its UUIDs andplacedAt, and a released one getsreleasedAt.fmt-check,vetandmanifests-checkpassed in a local Linux container. The regenerated deepcopy matched the committed file.placementscode parses the live cluster's Workloads. The new status fields appear once a scheduler built from this PR runs.