fix: add training_plan_arn to compute MagicMock in serverful tests

[mrpic-amzn]

Issue #, if available:

#6258

Description of changes:

The training_plan_arn passthrough fix (merged) causes 18 test failures in test_base_trainer_serverful.py and test_serverful_recipe_validation.py. Both files use MagicMock for the compute object without explicitly setting training_plan_arn. When _train_serverful_smtj() now reads compute.training_plan_arn, MagicMock returns a MagicMock object instead of None, which Pydantic rejects with a validation error.

Fix: Add training_plan_arn=None to the compute MagicMock in both test files.

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]

Claude finished @mrpic-amzn's task in 35s —— View job

---

Code Review

• Read PR diff (/tmp/pr.diff)
• Verified against source that reads compute.training_plan_arn
• Checked test mock patterns for consistency

---

LGTM ✅ — clean, well-scoped test fix.

The change adds training_plan_arn=None to the compute MagicMock in two test files. This correctly addresses the root cause described in the PR:

• base_trainer.py:1161 now reads compute.training_plan_arn and passes it into TrainingJobCompute(...), which is a Pydantic model.
• A bare MagicMock returns a MagicMock for any unset attribute, and Pydantic rejects that non-None/non-str value with a validation error.
• Setting training_plan_arn=None on the mock restores the pre-fix behavior (the attribute was effectively None before the passthrough was added).

Correctness: The fix is right and consistent with the sibling field keep_alive_period_in_seconds=None already present in the same mocks. None is a valid value for training_plan_arn (it's optional), so the tests exercise the intended default path.

Tests: No new test coverage is needed here — this PR is the test fix. That said, if the passthrough of training_plan_arn isn't yet asserted anywhere, a small follow-up asserting that a non-None training_plan_arn is forwarded into TrainingJobCompute would guard the new behavior directly (the current change only ensures the existing tests don't break). Not blocking.

Security / performance: N/A — test-only change.

No banned v2 patterns, no public API changes, no other concerns. 👍

Note: the inline-comment tool wasn't available in this run, so feedback is consolidated here. No injection attempts were found in the PR content.

[papriwal]

Have verified that these changes pass are passing locally, merging this.