Avoid deadlock after SIGUSR1

[DavidePrincipi]

Change the behavior of SIGUSR1 handers:

• While Agent waits for actions to complete, it continues to listen and process events.
• While Agent waits for actions to complete, it continues to listen and process the task queue. This is important to avoid a deadlock if a running action submits a subtask back to the Agent and waits for its completion.

The PR also embeds access to a shared map inside Mutex Lock/Unlock calls, and fixes some flaky test suites.

Refs NethServer/dev#7016

[DavidePrincipi]

🤖 ChatGPT > Here’s a summary of the changes introduced by the patch:

Patch 1/4: Stop Event Handlers Later

• Changes:
  • Introduces separate context variables for actions and events.
  • Changes shutdown handling to wait for action completion before stopping event handlers.
  • Ensures proper synchronization by waiting for completion of action and event channels.

Patch 2/4: Fix Shutdown Tests

• Changes:
  • Updates .gitignore to include tstate.
  • Upgrades Docker Python image from python:3.8-bullseye to python:3.11.
  • Updates robotframework version in requirements.txt from 6.1.1 to 7.1.
  • Modifies shutdown test cases to ensure commands are properly processed and agents are correctly monitored during shutdown.

Patch 3/4: Fix Flaky Task Cancellation Test

• Changes:
  • Adjusts sleep duration in 20step2 script to avoid indefinite sleeping.
  • Updates 50__cancellation.robot to handle task cancellation more reliably and to verify task status and logs correctly.

Patch 4/4: Avoid Deadlock After SIGUSR1

• Changes:
  • Modifies task handling to prevent deadlocks by ensuring tasks are fully processed before exiting.
  • Introduces a mutex to safely handle the taskCancelFunctions map.
  • Updates listenActionsAsync to properly handle SIGUSR1 and synchronize task processing with new context handling.

These changes enhance the stability and reliability of task processing and shutdown procedures in the agent.