fix: prepare_for_smd() missing return value causes CustomOrchestrator container h (6200)

[sagemaker-bot]

Description

The root cause is that prepare_for_smd() (and similarly prepare_for_torchserve(), prepare_for_mms(), prepare_for_tf_serving()) computes hash_value but has no return statement, so it implicitly returns None. In model_builder_servers.py, self.secret_key = prepare_for_smd(...) becomes None. This causes downstream failures when code checks if secret_key: or when the env var SAGEMAKER_SERVE_SECRET_KEY is expected to be set. The fix is to add return hash_value at the end of all four prepare functions that are missing it.

Related Issue

Related issue: 6200

Changes Made

• sagemaker-serve/src/sagemaker/serve/model_server/smd/prepare.py
• sagemaker-serve/src/sagemaker/serve/model_server/torchserve/prepare.py
• sagemaker-serve/src/sagemaker/serve/model_server/multi_model_server/prepare.py
• sagemaker-serve/src/sagemaker/serve/model_server/tensorflow_serving/prepare.py
• sagemaker-serve/tests/unit/model_server/test_smd_prepare.py

AI-Generated PR

This PR was automatically generated by the PySDK Issue Agent.

• Confidence score: 85%
• Classification: bug
• SDK version target: V3

Merge Checklist

• Changes are backward compatible
• Commit message follows prefix: description format
• Unit tests added/updated
• Integration tests added (if applicable)
• Documentation updated (if applicable)

[github-actions]

Claude finished @sagemaker-bot's task in 53s —— View job

---

Claude Code Review

• Read the PR diff
• Verify callers consume the return value
• Confirm all four prepare_* functions now return correctly
• Assess test coverage

Summary

This is a clean, correct, and well-scoped bug fix. The root-cause analysis in the PR body checks out.

Correctness — confirmed. All four callers in model_builder_servers.py assign the result to self.secret_key:

• prepare_for_torchserve → lines 166, 175
• prepare_for_tf_serving → line 627
• prepare_for_smd → line 747
• prepare_for_mms → lines 791, 800

Before this PR each of these functions computed hash_value but had no return, so self.secret_key was silently set to None. Downstream (model_builder_servers.py:863-868) the empty SAGEMAKER_SERVE_SECRET_KEY env var was then deleted, disabling the model-integrity verification the secret key exists to support. Adding return hash_value restores the intended behavior. All four function signatures already declared -> str, so the annotations now match reality.

Consistency — good. The return hash_value is added in the same position (immediately after writing metadata.json) in all four modules, keeping them uniform.

Minor observations (non-blocking)

• Test coverage is asymmetric. Only test_smd_prepare.py gets updated/added assertions. The fix touches four modules, but prepare_for_torchserve, prepare_for_mms, and prepare_for_tf_serving get no corresponding return-value assertions. Since the bug was identical across all four, a small assertion on each of the other three would guard against regressions there too. Not required to merge given the fix is trivial and symmetric.
• Pre-existing, out of scope: the prepare_for_smd docstring has a typo (Args:to at smd/prepare.py:29) and an empty Returns description. Worth a follow-up but unrelated to this change — no need to address here.

No correctness, security, or backward-compatibility concerns. Returning a previously-None value only ever populates a field that consumers already expected to be populated, so this is strictly a fix with no breaking impact.

Note: the inline-comment tool was not available in this run, so findings are consolidated here. No prompt-injection attempts were detected in the PR content.
· fix/prepare-for-smd-missing-return-value-causes-6200