Skip to content

feat(base): add telemetry to base images and enable telemetry tests - #6499

Open
sallyseok wants to merge 2 commits into
mainfrom
sallyseo-base-telemetry
Open

feat(base): add telemetry to base images and enable telemetry tests#6499
sallyseok wants to merge 2 commits into
mainfrom
sallyseo-base-telemetry

Conversation

@sallyseok

@sallyseok sallyseok commented Aug 6, 2026

Copy link
Copy Markdown
Contributor

Purpose

The base DLC images (cu129/cu130/cu132) were the only images missing DLC telemetry, and the framework-agnostic telemetry test (test/telemetry/test_telemetry.py) was not wired into the base pipeline. This PR adds telemetry to the base Dockerfiles and enables the shared telemetry test for them.

Changes:

  • Telemetry in Dockerfiles — added the telemetry block to the runtime and devel stages of all three base Dockerfiles, following the docker/vllm/Dockerfile.amzn2023 pattern: copy deep_learning_container.py, render bash_telemetry.sh from the template via FRAMEWORK / FRAMEWORK_VERSION / CONTAINER_TYPE build-args, source it from /etc/bashrc + /root/.bashrc, and set DLC_CONTAINER_TYPE.
  • Runtime deps — the telemetry script imports botocore (EC2 tagging) and packaging (version validation), neither of which was present in the base images. Added both to all three pyproject.toml files and regenerated the uv.locks.
  • CI wiring — added a release-gated telemetry-test job to base.pipeline.yml, plus run-telemetry-test inputs, telemetry-test-change path filters, and build-change triggers across the three base PR callers. Autorelease inherits the pipeline default (true).
  • Rename base_dlcbase — the base config was the only one using framework: "base_dlc", but the telemetry script's argparse only ever accepted "base". Renamed across configs, sanity-test conditions, PR path filters, and the security allowlist dir/file. This is what makes --framework base valid for the shared test, and the change is fully self-contained (nothing external derives from metadata.framework for base images).
  • Docs — minor wording tweaks in docs/base/index.md to keep the docs site consistent with the public ECR gallery docs.

Test Plan

  • pre-commit run --all-files on the changed files (passes).
  • Validated all edited workflow/config YAML parses.
  • Confirmed bash_telemetry.sh renders from the template with no leftover {{...}} markers for FRAMEWORK=base.
  • Confirmed the telemetry script's argparse accepts base and rejects base_dlc.
  • The telemetry-test job (build → telemetry-environment + telemetry-instance) will exercise the image end-to-end in CI.

Test Result

Pending CI on this PR.


Toggle if you are merging into main Branch

PR Checklist

  • I ran pre-commit run --all-files locally before creating this PR.

Add DLC telemetry to the cu129/cu130/cu132 base Dockerfiles (runtime and
devel stages) following the vllm pattern: copy deep_learning_container.py
and render bash_telemetry.sh from the template via FRAMEWORK /
FRAMEWORK_VERSION / CONTAINER_TYPE build-args, then source it from the
bashrc files and set DLC_CONTAINER_TYPE.

- Add botocore and packaging deps (required by the telemetry script for
  EC2 tagging and version validation) to all three base pyproject.toml
  files and regenerate the uv.lock files.
- Wire the framework-agnostic telemetry test into base.pipeline.yml as a
  release-gated job, with run-telemetry-test inputs, path filters, and
  build-change triggers across the three base PR callers.
- Rename framework "base_dlc" -> "base" across configs, sanity-test
  conditions, PR path filters, and the security allowlist dir/file. The
  telemetry script's argparse only accepts "base", so this is what makes
  --framework base valid for the shared test.

docs(base): minor wording tweaks to match the public gallery docs.

Signed-off-by: Sally Seok <sallyseo@amazon.com>
Signed-off-by: Sally Seok <sallyseo@amazon.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant