engine: context event loop.

[pwhelan]

Summary

Add a new event loop for contexts and for signaling context events.

Description

This event loop has a channel that is used internally for signals, ie: shutdown, reload, etc... The main loop now uses this event loop to receive shutdown and reload signals.

This should solve the issue addressed in #9908 in a cross platform and much more stable and clean manner.

---

Enter [N/A] in the box, if an item is not applicable to your change.

Testing
Before we can approve your change; please submit the following in a comment:

• Example configuration file for the change
• Debug log output from testing the change

• Attached Valgrind output that shows no leaks or memory corruption was found

If this is a change to packaging of containers or native binaries then please confirm it works for all targets.

• Run local packaging test showing all targets (including any new ones) build.
• Set ok-package-test label to test for all targets (requires maintainer to do).

Documentation

• Documentation required for this feature

Backporting

• Backport to latest stable release.

---

Fluent Bit is licensed under Apache 2.0, by submitting this pull request I understand that this code will be released under the terms of that license.

[leonardo-albertovich]

I can't conduct a complete review now but please check this out.

[leonardo-albertovich]

You can't call flb_pipe_w from a signal handler. That's why I said you were working on something similar in PR #10275.

[pwhelan]

That was an experiment to see what could be done to avoid using atomic variables as signal flags. I obvioulsy crashed head first into the fact that EINTR will interrupt and basically drop the events from the queue before they can be emitted.

I think the only real way around using atomic variables will be to add support for signal events to monkey. I was hoping to avoid it but I at least cannot see another way through.

[leonardo-albertovich]

That's why I tagged you in that other PR.

I think what he did with the signal queue is correct but we need to do it using the platform agnostic atomic operation wrapper which we need to move from cmetrics to cfl (I thought that was already done but it seems I was wrong).

I think with that, the signal mask tweak and a bit of elbow grease to ensure that the mechanism is tidy we should be good.

[leonardo-albertovich]

Were you able to make any progress on this? Do you need anything from me?