fix: bump vllm-router image to nightly-20260708-b93cbcb / 升级 vllm-router 镜像至 nightly-20260708-b93cbcb

[hongxiayang]

Summary

• Bump the disagg VLLM_ROUTER_IMAGE default in benchmarks/multi_node/amd_utils/job.slurm from nightly-20260629-e667ebb to nightly-20260708-b93cbcb.
• Why: vllm/vllm-router retains only ~15 dated nightlies on Docker Hub (one added daily). nightly-20260629-e667ebb is currently the second-oldest retained tag, so it will be garbage-collected within ~1-2 days — after which every multi-node vllm-disagg job would fail to pull the router container (a 404, the same class of failure previously seen with the expired nightly-20260617 tag).
• nightly-20260708-b93cbcb was already exercised in a green disagg run, so it's a known-good router build.

This is a small, config-independent infra fix affecting all AMD vllm-disagg configs; no perf-changelog.yaml change (no benchmark-config change).

Test plan

• Confirm nightly-20260708-b93cbcb resolves on Docker Hub (HTTP 200).
• A subsequent multi-node vllm-disagg run pulls the router container successfully.

中文说明

• 将 benchmarks/multi_node/amd_utils/job.slurm 中分离式(disagg)的 VLLM_ROUTER_IMAGE 默认值从 nightly-20260629-e667ebb 升级为 nightly-20260708-b93cbcb。
• 原因： vllm/vllm-router 在 Docker Hub 上仅保留约 15 个每日 nightly 标签（每天新增一个）。nightly-20260629-e667ebb 目前是倒数第二个仍保留的标签，一两天内即会被回收；届时所有多节点 vllm-disagg 任务都将无法拉取路由器容器（404，与此前 nightly-20260617 标签过期时同类的故障）。
• nightly-20260708-b93cbcb 已在一次通过的 disagg 运行中验证，是已知可用的路由器构建。

这是一个与具体配置无关的小型基础设施修复，影响所有 AMD vllm-disagg 配置；不涉及 perf-changelog.yaml（未改动基准配置）。

Made with Cursor

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

[claude]

LGTM — one-line vllm-router image tag rotation to avoid Docker Hub GC of the old nightly.

Extended reasoning...

Overview

Single-line change to benchmarks/multi_node/amd_utils/job.slurm bumping the default VLLM_ROUTER_IMAGE from nightly-20260629-e667ebb to nightly-20260708-b93cbcb. Only the default env-var value in a parameter-expansion (${VLLM_ROUTER_IMAGE:-...}) is changed; no logic, no plumbing, no benchmark config touched.

Security risks

None. The tag lives in the same vllm/vllm-router Docker Hub namespace and follows the same nightly-YYYYMMDD-<sha> naming pattern already in use; no new registry, credential, or attack surface introduced.

Level of scrutiny

Very low. This is exactly the class of infra-hygiene fix the codebase has done before (the comment on line 316 already documents the ~16-nightly retention window). The PR description states the new tag was exercised in a green disagg run, and the change is trivially revertable.

Other factors

• No CODEOWNER-guarded logic touched.
• No perf-changelog.yaml change (correct — no benchmark-config semantics change).
• Bug hunting system found nothing.
• Only affects the vllm-disagg router path; SGLang and Atom disagg paths are unaffected.