Repository navigation
feat(observability): show GPU activity from placements and DCGM - #395
Merged
Merged
Conversation
ShubyM
marked this pull request as ready for review
October 30, 2026 18:16
/api/v1/inspect/activity lists one lane per Workload that held GPUs in a time window. Each lane has its role, node, pod, GPU UUIDs, the runs on its runtime that lived then, and the intervals it held the GPUs, clipped to the window. Live placements come from Workload status. Released ones come from open_rl:placements:<runtime>. When two lanes held one GPU at overlapping times, both are marked shared_gpu, because that GPU's metrics are not one lane's own. If Redis fails or times out, it returns the live lanes and an error naming Redis. /api/v1/inspect/query runs a named DCGM query (gpu_busy, gpu_util, gpu_memory_mib) over the GPUs a run's or a node's lanes held. It reads the Prometheus-compatible endpoint set in OPEN_RL_PROMETHEUS_URL and is off when that is unset. gpu_busy reads DCGM_FI_PROF_GR_ENGINE_ACTIVE and falls back to DCGM_FI_DEV_GPU_UTIL where that has no data; the response's metric field names the one it read.
ShubyM
force-pushed
the
observability/gpu-activity
branch
from
October 30, 2026 18:24
2d39ab6 to
130a4d9
Compare
droot
approved these changes
Oct 30, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 5 of #390. Show which GPUs each run held over time, and how busy they were.
What changed
/api/v1/inspect/activityreturns one lane per Workload with its role, node, pod, GPUs, runs and held interval. It reads live placements from Workload status and history from Redis. Lanes that share a GPU at the same time are markedshared_gpu./api/v1/inspect/query?name=runs named PromQL queries (gpu_busy,gpu_util,gpu_memory_mib) scoped to a run's or node's GPUs.gpu_busyreports which DCGM metric it used.What you configure
OPEN_RL_PROMETHEUS_URLto a Prometheus-compatible endpoint, such as the in-cluster GMP frontend. Unset turns/queryoff./activityneeds no setup.Verification
make testandmake lintpass at each commit./query gpu_utilreturned a node's GPU series andgpu_busyreportedDCGM_FI_PROF_GR_ENGINE_ACTIVE./activitystill returns live lanes with an error.