[Fix][Core/Dashboard] component_cpu_percentage metric always zero for dashboard subprocess modules

[MortalHappiness]

Description

The component_cpu_percentage metric is currently always zero for dashboard subprocess modules. This PR fixes this.

Related issues

Similar symptom with #29848

Additional information

Before

After

[gemini-code-assist]

Code Review

This pull request introduces caching of psutil.Process handles for dashboard subprocess modules to ensure accurate CPU usage reporting, as psutil.Process.cpu_percent() requires a persistent handle to measure relative CPU usage. It also adds comprehensive unit tests to verify this behavior. The reviewer's feedback recommends handling potential exceptions like psutil.NoSuchProcess or psutil.AccessDenied when instantiating psutil.Process for a PID, and checking for None before recording metrics to prevent the metric recording loop from crashing.

[gemini-code-assist]

Since _get_subprocess_module_proc can return None if the process is missing or inaccessible, we should check if the returned process is not None before attempting to record its metrics.

Suggested change

	self._record_cpu_mem_metrics_for_proc(
	proc, subprocess_module_handle.module_cls.__name__
	self._get_subprocess_module_proc(pid),
	subprocess_module_handle.module_cls.__name__,
	)
	proc = self._get_subprocess_module_proc(pid)
	if proc is not None:
	self._record_cpu_mem_metrics_for_proc(
	proc,
	subprocess_module_handle.module_cls.__name__,
	)

[rueian]

Looks like a valid suggestion.

[gemini-code-assist]

If a subprocess module exits or restarts, or if there are permission restrictions, psutil.Process(pid) can raise psutil.NoSuchProcess or psutil.AccessDenied. Wrapping the instantiation in a try-except block and returning None prevents these exceptions from crashing the periodic metric recording loop.

    def _get_subprocess_module_proc(self, pid: int) -> Optional[psutil.Process]:
        """Return the cached psutil.Process for pid, creating it if needed.

        Reusing the handle is what gives cpu_percent() a baseline to measure
        against. See the comment on self._subprocess_module_procs.
        """
        proc = self._subprocess_module_procs.get(pid)
        # is_running() is False once the process is gone, and also once the pid
        # has been reused by an unrelated process, in which case the cached
        # handle's baseline no longer describes the process being measured.
        if proc is None or not proc.is_running():
            try:
                proc = psutil.Process(pid)
                self._subprocess_module_procs[pid] = proc
            except (psutil.NoSuchProcess, psutil.AccessDenied):
                return None
        return proc