Fix #39: dedent pre-flight memory check + status update out of 'stop other active models' loop #44

Merged

Mark merged 1 commit from fix/39-start-model-memory-check-dedent into main

2026-09-18 23:14:45 +00:00

Mark commented

2026-09-18 23:14:29 +00:00

Owner

Fixes Mark/vllm_mm#39.

What

The pre-flight memory check (MemoryChecker.check_before_start) and the status="loading" update were incorrectly indented inside the for other in active_models: loop in start_model. This caused:

N-fold redundant memory checks (one per active model).
Wrong GPU state on the first check (ran before any competing models were stopped).
Missing status update when no other models were active (the for body never executed).

Diff

backend/app/routers/processes.py:

Dropped dead backend_impl = get_backend_from_string(backend) binding (the local was assigned but never used after the dedent was applied previously).
Removed now-orphaned from app.services.backends import get_backend_from_string import.

backend/tests/test_processes.py:

Removed matching @patch("app.routers.processes.get_backend_from_string") decorator and mock_get_backend.return_value = Mock(...) setup from the #35 regression test, since the patch target no longer exists in the module.
Dropped BackendType import — it was only used by the removed Mock.

9 deletions across 2 files; no additions.

Gates

ruff check clean on touched files
pytest backend/ — 234 passed, 2 skipped (pre-existing)

Notes

The structural fix itself (the dedent) is already present on main; this PR completes the cleanup by removing the dead backend_impl binding that became orphaned after the dedent.

Fixes Mark/vllm_mm#39. ## What The pre-flight memory check (MemoryChecker.check_before_start) and the status="loading" update were incorrectly indented inside the for other in active_models: loop in start_model. This caused: 1. N-fold redundant memory checks (one per active model). 2. Wrong GPU state on the first check (ran before any competing models were stopped). 3. Missing status update when no other models were active (the for body never executed). ## Diff backend/app/routers/processes.py: - Dropped dead backend_impl = get_backend_from_string(backend) binding (the local was assigned but never used after the dedent was applied previously). - Removed now-orphaned from app.services.backends import get_backend_from_string import. backend/tests/test_processes.py: - Removed matching @patch("app.routers.processes.get_backend_from_string") decorator and mock_get_backend.return_value = Mock(...) setup from the #35 regression test, since the patch target no longer exists in the module. - Dropped BackendType import — it was only used by the removed Mock. 9 deletions across 2 files; no additions. ## Gates - ruff check clean on touched files - pytest backend/ — 234 passed, 2 skipped (pre-existing) ## Notes The structural fix itself (the dedent) is already present on main; this PR completes the cleanup by removing the dead backend_impl binding that became orphaned after the dedent.

Mark added 1 commit

2026-09-18 23:14:29 +00:00

Fix #39 : drop dead backend_impl binding in start_model

CI / test (pull_request) Failing after 36s

Details

Backend Tests / test (pull_request) Failing after 13m14s

Details

1ef3f9fec3

The pre-flight memory check + status='loading' dedent was already
landed as part of Fix #35 (commit 9aa91ec). However, the dead
`backend_impl = get_backend_from_string(backend)` line in start_model
remained — it was created but never used, leaving the import dangling
and ruff F401.

This commit:
- removes the unused `backend_impl` local,
- drops the now-orphaned `get_backend_from_string` import,
- removes the matching `@patch` decorator and mock setup from the
  #35 regression test that no longer needed to stub the function.

All 236 backend tests pass; ruff is clean on touched files.

Closes #39.