Skip to content

change: update image_uri_configs 09-21-2026 13:30:13 PST - #6297

Closed
Yadan-Wei wants to merge 2 commits into
masterfrom
master-rba-v3
Closed

Yadan-Wei wants to merge 2 commits into
masterfrom
master-rba-v3

Conversation

@Yadan-Wei

Copy link
Copy Markdown
Contributor

Issue #, if available:

Description of changes:

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

@github-actions

github-actions Bot commented Sep 21, 2026

Copy link
Copy Markdown

Claude finished @Yadan-Wei's task in 1m 46s —— View job


PR Review — image_uri_config updates

  • Read the full PR diff
  • Analyze JSON config changes for correctness
  • Check version ordering / aliases / tag structure consistency
  • Post review feedback

This PR is a data-only update to the image_uri_config/*.json files: adding the new me-west-1 region across many configs, backfilling ADC/partition registries (eu-isoe-west-1, eusc-de-east-1, us-iso-east-1, us-isob-east-1, us-isof-east-1, us-isof-south-1), introducing new framework versions (vllm 0.23–0.29, sglang 0.5.13–0.5.20, llama-cpp 1.0.0, ray-serve 1.2/1.3, etc.), and reordering keys within version blocks. No Python/public-API changes, and the account IDs added are the standard public AWS DLC registry accounts, so there are no security concerns.

The changes look mechanically sound (no duplicate JSON keys introduced by the reorderings, and the moved container_version/tag_prefix blocks net out correctly). A few things worth confirming before merge:

1. Missing trailing newline on 10 files (nit, but likely CI-relevant)

The diff shows \ No newline at end of file for:
llama-cpp-arm64.json, llama-cpp.json, pytorch-amzn2023.json, ray-serve.json, sglang-server.json, sglang.json, vllm-omni.json, vllm-server.json, vllm.json, whisperx.json.

These files previously ended with a newline. If the repo runs an end-of-file-fixer / pre-commit / JSON-format check, this will fail. Worth restoring the trailing newline for consistency with the untouched configs (pytorch.json, tensorflow.json, huggingface-llm*.json all keep theirs).

2. Region availability regresses on the newest versions — please confirm intentional

In several files the newest version ships with fewer partitions than the version right before it:

  • vllm.json: 0.27.1 and 0.28.0 include the China/GovCloud/ISO partitions (cn-*, us-gov-*, us-iso*, eusc-de-east-1, eu-isoe-west-1), but the newest 0.29.0 (and 0.23.00.26.0) omit all of them.
  • sglang.json: 0.5.18 includes the full partition set, but the newer 0.5.19/0.5.20 and the 0.5.130.5.17 entries omit cn-*/gov/iso.
  • ray-serve.json: 1.2 and 1.3 omit cn-*/gov/iso while 1 and 1.4 include them.

This is plausibly a staged rollout (image not yet published to those partitions), which is fine — but since 0.29.0 is the newest tag and the 0.29 alias points to it, a retrieve(...) call for these in GovCloud/ISO/China will fail even though an older version works there. Please double-check the omissions are deliberate and not a generation artifact.

3. Version aliases — verified consistent

vllm (0.29 → 0.29.0, 0.25 → 0.25.1, etc.), sglang (0.5 → 0.5.20), and huggingface-llm aliases all resolve to existing version entries. No dangling aliases spotted.

Overall this looks like a routine, low-risk config bump. Items (1) and (2) are the only things I'd want addressed/confirmed.

@codecov

codecov Bot commented Sep 21, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.42%. Comparing base (5918e0b) to head (07e8b14).
⚠️ Report is 3 commits behind head on master.

Additional details and impacted files
@@           Coverage Diff           @@
##           master    #6297   +/-   ##
=======================================
  Coverage   71.42%   71.42%           
=======================================
  Files         556      556           
  Lines       68598    68598           
=======================================
  Hits        48996    48996           
  Misses      19602    19602           
🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@github-actions

github-actions Bot commented Sep 22, 2026

Copy link
Copy Markdown

Claude finished @mufaddal-rohawala's task in 1m 13s —— View job


PR Review — image_uri_config updates

  • Read the full PR diff (/tmp/pr.diff)
  • Validate JSON structure, key reorderings, aliases
  • Check region/partition consistency across versions
  • Trace how retrieve() handles missing regions
  • Post review

This is a data-only update to image_uri_config/*.json — adding me-west-1, backfilling ADC/partition registries (eu-isoe-west-1, eusc-de-east-1, us-iso-east-1, us-isob-east-1, us-isof-east-1, us-isof-south-1), introducing new framework versions (vllm 0.23–0.29, sglang 0.5.13–0.5.20, ray-serve 1.2/1.3, llama-cpp 1.0.0, whisperx 3.8.6, vllm-omni/server variants), and reordering keys within version blocks. No Python/public-API changes and the added account IDs are the standard public AWS DLC registry accounts — no security concerns.

The structural edits are mechanically sound: the reorderings (moving container_version to the top, repository/tag_prefix to the bottom) net out with no duplicate keys, and all aliases resolve to existing entries (vllm 0.25 → 0.25.1, 0.29 → 0.29.0, etc.; sglang 0.5 → 0.5.20; huggingface-llm unchanged). Two things to address/confirm before merge:

1. Missing trailing newline on 10 files

The diff shows \ No newline at end of file for:
llama-cpp-arm64.json, llama-cpp.json, pytorch-amzn2023.json, ray-serve.json, sglang-server.json, sglang.json, vllm-omni.json, vllm-server.json, vllm.json, whisperx.json.

These files previously ended with a newline (untouched configs like pytorch.json, tensorflow.json, huggingface-llm.json still do). If any end-of-file-fixer / JSON-format check runs in CI or pre-commit this will fail, and it's an unnecessary diff otherwise. Please restore the trailing newline on these 10 files.

2. Region availability regresses on the newest (aliased) versions — please confirm intentional

In several configs the newest version — which the floating alias resolves to — ships with fewer partitions than an older version, dropping the China / GovCloud / ISO registries (cn-*, us-gov-*, us-iso*, eusc-de-east-1, eu-isoe-west-1):

  • vllm.json: 0.27.1 and 0.28.0 include those partitions, but the newest 0.29.0 (target of the 0.29 alias) and 0.23.00.26.x omit them.
  • sglang.json: 0.5.18 includes the full set, but the newer 0.5.19/0.5.20 (target of the 0.5 alias) and 0.5.130.5.17 omit cn/gov/iso.
  • ray-serve.json: 1 and 1.4 include them; 1.2 and 1.3 omit them.
  • vllm-omni.json: 1.6 includes them; the newest 1.7 omits them (also 1.3/1.4/1.5 are mixed).
  • vllm-server.json: 2.1 omits them while 2.2/2.3/2.4 include them.

This is plausibly a staged rollout (image not yet published to those partitions), which is fine as data. The consequence worth noting: because these are the alias targets, a call like image_uris.retrieve(framework="vllm", version="0.29", region="cn-north-1") will hit _registry_from_region_validate_arg (sagemaker-core/src/sagemaker/core/image_uris.py:537, :632) and raise Unsupported region: ... You may need to upgrade your SDK version — a misleading error, since the region isn't stale-SDK, the image just isn't published there, and an older version does work in that region. Please confirm the omissions are deliberate (staged publish) rather than a generation artifact; if deliberate, no code change is needed here.

Verified OK

  • Alias targets all exist — no dangling aliases.
  • Key reorderings introduce no duplicate JSON keys; moved container_version/tag_prefix/repository blocks net out correctly.
  • Added registry account IDs match the standard public DLC accounts already used elsewhere in these files.

Overall a routine, low-risk config bump. Only item (1) needs a fix; item (2) just needs a quick confirmation.

Note: the repo's inline-comment tool wasn't available in this run, so findings are consolidated here rather than posted inline.

This branch was successfully deployed

1 active deployment
auto-approve 07e8b149 Deployed Sep 22, 2026 by Yadan-Wei via wait-for-approval #409
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.

3 participants