Skip to content

[plugin][dashboard] use nightly date tagged docker#503

Merged
valarLip merged 12 commits intomainfrom
zejun/refine_oot_benchmark_0405
Apr 13, 2026
Merged

[plugin][dashboard] use nightly date tagged docker#503
valarLip merged 12 commits intomainfrom
zejun/refine_oot_benchmark_0405

Conversation

@zejunchen-zejun
Copy link
Copy Markdown
Contributor

@zejunchen-zejun zejunchen-zejun commented Apr 7, 2026

when running vllm-atom benchmark, only use docker image with the date instead of latest docker
CC: @wuhuikx

Copilot AI review requested due to automatic review settings April 7, 2026 08:13
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR improves reproducibility and presentation of ATOM vLLM OOT benchmark runs by resolving the floating vllm-latest Docker tag to a stable, date-tagged (or digest-pinned) image reference and by surfacing the resolved image digest in the benchmark dashboard while suppressing commit metadata in OOT detail views.

Changes:

  • Update the OOT benchmark workflow to resolve rocm/atom-dev:vllm-latest to a same-digest nightly tag (or digest-pin) and record the pulled image digest into the result payload.
  • Add a new .github/scripts/resolve_oot_image.py helper to perform the Docker Hub registry resolution from a floating tag to a stable reference.
  • Extend the OOT dashboard pipeline/UI to ingest and display the Docker image digest, and hide commit/message/author for ATOM-vLLM detail views.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File Description
.github/workflows/atom-vllm-oot-benchmark.yaml Resolves prebuilt OOT image to a stable nightly/digest reference and records the pulled digest into the benchmark payload.
.github/scripts/resolve_oot_image.py New resolver script that maps a floating tag to a same-digest nightly tag (or digest-pinned fallback).
.github/scripts/oot_benchmark_to_dashboard.py Adds oot_image_digest into the extra metadata string for dashboard ingestion.
.github/dashboard/index.html Parses digest from extra and updates detail/popover rendering to show digest and hide commit metadata for ATOM-vLLM.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot AI review requested due to automatic review settings April 8, 2026 03:32
@zejunchen-zejun zejunchen-zejun changed the title [plugin][OOT dashboard] use nightly date tagged docker and neglect commit/message/author info [plugin][OOT dashboard] use nightly date tagged docker Apr 8, 2026
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@wuhuikx
Copy link
Copy Markdown
Contributor

wuhuikx commented Apr 9, 2026

Low — Reverse lookup in .github/scripts/resolve_oot_image.py performs a linear, per-tag digest fetch for nightly tags. As the number of tags grows, request volume increases and may hit Docker Hub rate limits, which can force a fallback to floating vllm-latest and weaken reproducibility stability.

resolve_oot_image.py
Lines 162–167

candidates = nightly_candidates(list_tags(repository, token), preferred_version)
matched_tag = None
for candidate in candidates:
if get_manifest_digest(repository, candidate.tag, token) == reference_digest:
matched_tag = candidate.tag
Suggestion: Limit the scan window (for example, only the most recent N nightly tags), or use a tag-list API/flow that includes digests when possible to reduce N manifest requests.

@zejunchen-zejun
Copy link
Copy Markdown
Contributor Author

Low — Reverse lookup in .github/scripts/resolve_oot_image.py performs a linear, per-tag digest fetch for nightly tags. As the number of tags grows, request volume increases and may hit Docker Hub rate limits, which can force a fallback to floating vllm-latest and weaken reproducibility stability.

resolve_oot_image.py Lines 162–167

candidates = nightly_candidates(list_tags(repository, token), preferred_version) matched_tag = None for candidate in candidates: if get_manifest_digest(repository, candidate.tag, token) == reference_digest: matched_tag = candidate.tag Suggestion: Limit the scan window (for example, only the most recent N nightly tags), or use a tag-list API/flow that includes digests when possible to reduce N manifest requests.

The docker image is searched from the latest to the oldest according to the digest, so the search overhead is very small.
If the docker cannot be found, we will fallback to the docker with the digest tag instead of vLLM:latest

Copilot AI review requested due to automatic review settings April 9, 2026 07:04
and neglect commit/message/author info

Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
@zejunchen-zejun zejunchen-zejun force-pushed the zejun/refine_oot_benchmark_0405 branch from 512283e to 3d69252 Compare April 9, 2026 07:08
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot AI review requested due to automatic review settings April 10, 2026 14:18
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot AI review requested due to automatic review settings April 11, 2026 01:56
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.


💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

@wuhuikx
Copy link
Copy Markdown
Contributor

wuhuikx commented Apr 11, 2026

Can we remove the OOT or oot naming?

@zejunchen-zejun zejunchen-zejun changed the title [plugin][OOT dashboard] use nightly date tagged docker [plugin][dashboard] use nightly date tagged docker Apr 12, 2026
@zejunchen-zejun
Copy link
Copy Markdown
Contributor Author

zejunchen-zejun commented Apr 12, 2026

Can we remove the OOT or oot naming?

Sure. I will remove the OOT terminology in this PR's code change. For other remove work, let's make in PR #541

Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
wuhuikx
wuhuikx previously approved these changes Apr 13, 2026
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

.github/workflows/atom-vllm-benchmark.yaml:1

  • The workflow name was changed to "ATOM vLLM Benchmark", but later steps still query the old workflow name ("ATOM vLLM OOT Benchmark") when downloading baseline runs via gh run list --workflow=.... That lookup will return no runs, so regression/baseline comparison will silently stop working. Update the gh run list filter to match the new workflow name (or switch to referencing the workflow file path / workflow ID to avoid breakage on future renames).
    .github/workflows/atom-vllm-benchmark.yaml:186
  • This fallback uses the floating tag rocm/atom-dev:vllm-latest when resolution fails, which can undermine the PR goal of avoiding non-date-tagged (and non-stable) images for main-branch benchmarks. Consider failing the run when digest resolution cannot be performed, or add a more robust fallback that still produces a digest-pinned reference (e.g., retry/backoff, optional Docker Hub auth to avoid rate limits, or a second resolver using docker manifest inspect).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

wuhuikx
wuhuikx previously approved these changes Apr 13, 2026
valarLip
valarLip previously approved these changes Apr 13, 2026
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
@zejunchen-zejun zejunchen-zejun dismissed stale reviews from valarLip and wuhuikx via ef7dc87 April 13, 2026 10:12
@valarLip valarLip merged commit a6fe785 into main Apr 13, 2026
25 of 29 checks passed
@valarLip valarLip deleted the zejun/refine_oot_benchmark_0405 branch April 13, 2026 11:45
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.

4 participants