[AMD][CI] Temporarily bypass local-registry image pulls - #36171
Merged
Merged
Conversation
bingxche
marked this pull request as ready for review
August 24, 2026 09:41
yctseng0211
approved these changes
Aug 24, 2026
4 tasks
4 tasks
cursor Bot
pushed a commit
that referenced
this pull request
Aug 25, 2026
The cache probe measured hits at up to 13-way concurrency, but the pull it is compared against was only ever measured at low concurrency, so the speedup mixed two regimes. This adds the same 24-way, no-max-parallel fan-out with AMD_CI_IMAGE_TARBALL_CACHE=0, so every job pulls from Docker Hub and the comparison is like-for-like. The in-network registry is deliberately left unmeasured: #36171 removed that path from main because concurrent CI pulls saturate it, so no code remains to exercise, and aiming 24 simultaneous 64 GB pulls at a host already known to be overloaded would disrupt unrelated jobs. Co-authored-by: quitenode <quitenode@users.noreply.github.com>
saturn-acc
pushed a commit
to saturn-acc/sglang
that referenced
this pull request
Aug 31, 2026
nzr-niu
pushed a commit
to nzr-niu/sglang
that referenced
this pull request
Sep 1, 2026
jakki-amd
pushed a commit
to jakki-amd/sglang
that referenced
this pull request
Sep 9, 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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
AMD CI currently tries the shared local registry before Docker Hub. The nightly jobs that explicitly push newly published images into the local registry are disabled, so a cold image can still trigger a large upstream transfer through that shared path.
During the current incident, job 97284747841 spent about 4 hours 16 minutes pulling
rocm/sgl-dev:v0.5.18-rocm724-mi30x-20260823through the local registry while many AMD runners were active.Changes
rocm/sgl-devimage with the existing authenticated, six-attempt exponential-backoff path.--custom-image; this only changes automatic prebuilt-image selection.This is an intentionally narrow incident-mitigation change. Image selection and container launch behavior are otherwise unchanged.
Risk and rollback
This shifts CI image traffic to Docker Hub and may expose jobs to rate limits or public-network instability. The existing optional Docker Hub login and retry path reduce that risk.
Rollback is straightforward: restore the local-first blocks after local-registry publishing or pre-warming is enabled and the path is verified under expected concurrency.
Testing
CI States
Latest PR Test (Base): ✅ Run #32711860839
Latest PR Test (Extra): ❌ Run #32711860836
Latest PR Test (AMD ROCm 7.2): ❌ Run #32711860962