Feat/gpu-ci

[specture724]

Refactor test_update.py to pytest style and create a workflow for main branch

[Copilot]

Pull Request Overview

This PR refactors the test_update.py to use pytest framework and adds CI workflow capabilities. The main change is replacing the direct test execution with a pytest-based test structure that uses subprocess to run distributed training via torchrun.

• Converted test_update.py from standalone script to pytest-based test structure
• Added pytest mark for GPU-specific tests with multi-GPU requirements
• Enhanced logging control in ParameterServer to support flexible rank-based logging

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
tests/test_update.py	Refactored from standalone script to pytest framework with subprocess-based distributed test execution
checkpoint_engine/ps.py	Enhanced logging system to support configurable logger rank for different update scenarios
.pre-commit-config.yaml	Added ruff ignore rule for subprocess security warning (S603)

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Using random port selection for distributed training can lead to port conflicts in CI environments. Consider using a more robust port selection mechanism or environment variable override.

[Copilot]

The _logger_rank initialization should be documented to explain when and how this value changes during execution.

Suggested change

	self._logger_rank = 0
	# _logger_rank determines which rank is responsible for logging output.
	# By default, it is set to 0, meaning only rank 0 will perform logging.
	# If logging from other ranks is required, this value can be changed accordingly.
	self._logger_rank = 0