test: enforce pytest markers on components/src tests

[furionw]

Why

We aren't running all unit tests in components/src/ because sometimes we forget to add pytest markers. This PR enforces that to have better test coverage.

CI was silently skipping pytest marker enforcement on components/src/ tests because the pre-commit hook's lean venv could not import test modules that pull in torch, vllm, sglang, PIL, etc.

What Change

• Extend the marker validator to enforce coverage across components/src/ test trees.
• Replace MagicMock-based dependency stubs with real-class stubs that satisfy pydantic schema generation, ForwardRef resolution, and class inheritance.
• Stub ~95 leaf modules transitively imported by tests (torch, PIL, fsspec, sglang., vllm., msgspec., mistral_common., nixl., transformers.models.).
• Add the missing gpu_0 hardware marker to two flagged test files surfaced by the validator.

Test Plan

• Lean pre-commit venv (matches CI ubuntu-latest): pre-commit clean && pre-commit run pytest-marker-report --all-files — Passed, 1871 tests checked, 0 missing.
• Verified the same hook now flags any test missing one of the three required marker categories.

[coderabbitai]

Walkthrough

This PR adds module-level pytest markers to standardize test categorization across multiple test suites in memory, multimodal, common utilities, frontend, and TRT-LLM modules. Additionally, it enhances the pytest marker reporting utility with comprehensive module stubbing, multiple test directory support, and improved collection behavior.

Changes

Cohort / File(s)	Summary
Multimodal and Memory Test Markers components/src/dynamo/common/tests/memory/test_multimodal_embedding_cache_manager.py, components/src/dynamo/common/tests/multimodal/test_async_encoder_cache.py, components/src/dynamo/common/tests/multimodal/test_embedding_transfer.py	Added module-level pytestmark lists applying unit, gpu_0, pre_merge markers (and asyncio, multimodal for embedding transfer test). Test logic unchanged.
Common Utils Test Markers components/src/dynamo/common/utils/tests/test_inject_reasoning_content.py, components/src/dynamo/common/utils/tests/test_label_injecting_collector.py	Added module-level pytestmark configuration with unit, gpu_0, pre_merge markers and required pytest import. No test logic modifications.
Frontend Test Markers components/src/dynamo/frontend/tests/test_sglang_processor_api.py, components/src/dynamo/frontend/tests/test_sglang_processor_unit.py, components/src/dynamo/frontend/tests/test_sglang_tool_calls.py, components/src/dynamo/frontend/tests/test_vllm_processor_unit.py	Added module-level pytestmark lists with framework-specific markers (sglang or vllm), unit, gpu_0, pre_merge. Test assertions and logic unchanged.
TRT-LLM Test Markers components/src/dynamo/trtllm/tests/test_dynamic_flags.py, components/src/dynamo/trtllm/tests/test_trtllm_additional_metrics.py	Added module-level pytestmark configuration applying unit, trtllm, gpu_0, pre_merge markers across test suites. No test logic altered.
Pytest Marker Reporting Utility tests/report_pytest_markers.py	Extended STUB_MODULES to cover torch, sglang, vllm, pydantic, fsspec, msgspec and related dependencies. Updated CLI --tests argument to support multiple directory paths with default ["tests", "components/src"]. Changed run_collection() signature to accept list of test paths. Enhanced stubbing logic for pydantic BaseModel replacements and vllm.__version__. Modified exit-code behavior to return 0 when no missing markers exist, preventing collection errors from blocking validation.
Mock GPU Allocation Test Utility tests/utils/test_mock_gpu_alloc.py	Updated module-level pytestmark from single skipif marker to list format, preserving CI/environment-based skipping logic while adding unit and nightly markers.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 55.56% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: enforcing pytest markers on component test files.
Description check	✅ Passed	The PR description covers the Why (enforcing markers across components/src), What (extending validator, adding stubs), and Test Plan with verification steps, aligning well with the template structure.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (2)

components/src/dynamo/frontend/tests/test_sglang_tool_calls.py (1)

23-28: Declare the SGLang test model in pytestmark as well.

This module pulls Qwen/Qwen3-0.6B through get_tokenizer(MODEL), so it should advertise that dependency with pytest.mark.model(MODEL) too. Otherwise predownload_models still has to discover it at runtime.

Suggested change

+MODEL = "Qwen/Qwen3-0.6B"
+
 pytestmark = [
     pytest.mark.unit,
     pytest.mark.sglang,
     pytest.mark.gpu_0,
     pytest.mark.pre_merge,
+    pytest.mark.model(MODEL),
 ]
-
-MODEL = "Qwen/Qwen3-0.6B"

As per coding guidelines, use pytest.mark.model("org/model-name") to declare HF models used; the predownload_models fixture reads these markers to download only what's needed.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@components/src/dynamo/frontend/tests/test_sglang_tool_calls.py` around lines
23 - 28, The pytest module-level marker list `pytestmark` is missing the HF
model marker for the SGLang test; add pytest.mark.model(MODEL) to the pytestmark
array so the predownload_models fixture knows to download the tokenizer/model
used by get_tokenizer(MODEL). Update the pytestmark declaration (the symbol
pytestmark and the constant MODEL referenced by get_tokenizer) to include
pytest.mark.model("Qwen/Qwen3-0.6B") or pytest.mark.model(MODEL) so the test
advertises its HF dependency.

components/src/dynamo/frontend/tests/test_vllm_processor_unit.py (1)

15-20: Add the HF model marker while you're centralizing pytestmark.

The tokenizer fixture loads Qwen/Qwen3-0.6B via AutoTokenizer.from_pretrained(MODEL), so this module should also declare pytest.mark.model(MODEL). Otherwise predownload_models cannot prefetch it for this pre-merge unit path.

Suggested change

+MODEL = "Qwen/Qwen3-0.6B"
+
 pytestmark = [
     pytest.mark.unit,
     pytest.mark.vllm,
     pytest.mark.gpu_0,
     pytest.mark.pre_merge,
+    pytest.mark.model(MODEL),
 ]
-
-MODEL = "Qwen/Qwen3-0.6B"

As per coding guidelines, use pytest.mark.model("org/model-name") to declare HF models used; the predownload_models fixture reads these markers to download only what's needed.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@components/src/dynamo/frontend/tests/test_vllm_processor_unit.py` around
lines 15 - 20, The pytest module-level markers list pytestmark needs to include
the HF model marker so predownload_models can fetch the tokenizer model; update
the pytestmark list (the pytestmark variable) to include
pytest.mark.model(MODEL) (or pytest.mark.model("Qwen/Qwen3-0.6B") if MODEL is
not available) so the tokenizer fixture that calls
AutoTokenizer.from_pretrained(MODEL) will be predownloaded.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@components/src/dynamo/common/tests/multimodal/test_embedding_transfer.py`:
- Around line 24-30: The module-level pytest marker list (pytestmark)
incorrectly includes pytest.mark.gpu_1; remove pytest.mark.gpu_1 from the
pytestmark list and instead add pytest.mark.gpu_1 only to the specific test
functions that require GPU (e.g., annotate the NIXL/CUDA tests with
`@pytest.mark.gpu_1` above their def), leaving CPU-only tests with their existing
markers; update the pytestmark variable and the individual test decorators
accordingly so gpu_1 is not applied module-wide.

In `@tests/report_pytest_markers.py`:
- Around line 560-570: The shim _FakeBaseModel defines model_config as a shared
class dict which leaks mutations across subclasses; modify
_FakeBaseModel.__init_subclass__ to assign a fresh dict to cls.model_config (or
replace the class attribute with an immutable default) so each subclass of
_FakeBaseModel gets its own model_config; ensure the test replaces
_pydantic.BaseModel with this fixed _FakeBaseModel so subclass mutations won’t
affect other tests.

In `@tests/utils/test_mock_gpu_alloc.py`:
- Around line 17-26: Remove the pytest.mark.nightly marker from the pytestmark
list in this module so the local-only profiler helper is not accidentally
included in scheduled/nightly runs; keep the existing pytest.mark.skipif that
prevents CI execution and, if needed, replace nightly with a dedicated
registered marker (e.g., pytest.mark.local or pytest.mark.local_only) and add
that marker to pytest.ini instead of reusing pytest.mark.nightly — update the
pytestmark variable to only include skipif and unit (or the new registered
local-only marker) and ensure the marker name referenced in pytest.ini matches
the one you add.

---

Nitpick comments:
In `@components/src/dynamo/frontend/tests/test_sglang_tool_calls.py`:
- Around line 23-28: The pytest module-level marker list `pytestmark` is missing
the HF model marker for the SGLang test; add pytest.mark.model(MODEL) to the
pytestmark array so the predownload_models fixture knows to download the
tokenizer/model used by get_tokenizer(MODEL). Update the pytestmark declaration
(the symbol pytestmark and the constant MODEL referenced by get_tokenizer) to
include pytest.mark.model("Qwen/Qwen3-0.6B") or pytest.mark.model(MODEL) so the
test advertises its HF dependency.

In `@components/src/dynamo/frontend/tests/test_vllm_processor_unit.py`:
- Around line 15-20: The pytest module-level markers list pytestmark needs to
include the HF model marker so predownload_models can fetch the tokenizer model;
update the pytestmark list (the pytestmark variable) to include
pytest.mark.model(MODEL) (or pytest.mark.model("Qwen/Qwen3-0.6B") if MODEL is
not available) so the tokenizer fixture that calls
AutoTokenizer.from_pretrained(MODEL) will be predownloaded.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 3cd4a7be-fdd1-4890-ae46-21feec73fbd1

📥 Commits

Reviewing files that changed from the base of the PR and between 91a8c6a and 09de537.

📒 Files selected for processing (13)

• components/src/dynamo/common/tests/memory/test_multimodal_embedding_cache_manager.py
• components/src/dynamo/common/tests/multimodal/test_async_encoder_cache.py
• components/src/dynamo/common/tests/multimodal/test_embedding_transfer.py
• components/src/dynamo/common/utils/tests/test_inject_reasoning_content.py
• components/src/dynamo/common/utils/tests/test_label_injecting_collector.py
• components/src/dynamo/frontend/tests/test_sglang_processor_api.py
• components/src/dynamo/frontend/tests/test_sglang_processor_unit.py
• components/src/dynamo/frontend/tests/test_sglang_tool_calls.py
• components/src/dynamo/frontend/tests/test_vllm_processor_unit.py
• components/src/dynamo/trtllm/tests/test_dynamic_flags.py
• components/src/dynamo/trtllm/tests/test_trtllm_additional_metrics.py
• tests/report_pytest_markers.py
• tests/utils/test_mock_gpu_alloc.py

[dmitry-tokarev-nv]

lgtm. Thank you!