#5994 Fix llprocess waiting too long on kill

[akleshchev]

Noticed some extra delays when media was being killed.

[Copilot]

Pull request overview

This PR reduces main-thread stalls and redundant termination work when an LLProcess is being killed/cleaned up, which should address the reported extra delays when media processes are terminated.

Changes:

• Add a mKillCalled flag to avoid repeating termination work in ~LLProcess() after an explicit kill().
• Reduce/avoid blocking waits on Windows by using GetExitCodeProcess() instead of WaitForSingleObject(..., 100) in tick() and shortening the destructor wait.
• Adjust kill() to mark the process as “kill requested” and return success once the kill signal/request is issued (without waiting for state transition).

Reviewed changes

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

File	Description
indra/llcommon/llprocess.h	Adds mKillCalled state to track explicit kill requests.
indra/llcommon/llprocess.cpp	Removes/shortens blocking waits on Windows; uses mKillCalled to skip redundant destructor termination after kill().

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[akleshchev]

@copilot, please add a unit test into llprocess_test.cpp that unsures tick() takes no more than an ms and a similar test for destruction.

[Copilot]

@copilot, please add a unit test into llprocess_test.cpp that unsures tick() takes no more than an ms and a similar test for destruction.

Added two regression tests in commit Add timing tests for tick() and destructor after kill():

• test 34 (tick() completes quickly after kill()): launches a long-sleeping process, kills it, waits 100 ms for OS-level termination, then times a single mainloop tick and asserts it completes in < 75 ms. This covers the code path that previously called WaitForSingleObject(..., 100) in tick() on Windows.

• test 35 (LLProcess destructor completes quickly after kill()): launches a long-sleeping process, calls kill() (which sets mKillCalled = true), then times the destructor and asserts it completes in < 75 ms. Since mKillCalled causes the destructor to skip the termination/wait block entirely, this should be near-instantaneous.

Both thresholds (75 ms) are well below the old 100 ms blocking wait, providing a clear regression guard while remaining robust on loaded CI systems.

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

1 out of 2 committers have signed the CLA.
✅ (akleshchev)[https://github.com/akleshchev]
❌ @Copilot
_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}