…ls and process shutdown (grpc#36903)

Fixes the CBF of `src/ruby/end2end/killed_client_thread_test.rb` (failure mode is a hang of the child process that receives the SIGTERM) that has been happening since grpc#36724

So far grpc-ruby CQ pluck operations have so far used a 20ms-interval busy poll to check interrupts in case we've received a signal, handle process shutdown, etc. This means ongoing RPCs will not terminate their CQ operations if we need to terminate the process (the loop simply exits without waiting for the CQ op to finish), causing a leak. Those RPCs can leave refs over their corresponding channels preventing [this](https://github.com/grpc/grpc/blob/8564f72e8e0334c25c480e0aec1df75bdc1fce14/src/ruby/ext/grpc/rb_channel.c#L653) from terminating (the channels don't reach state SHUTDOWN after being destroyed).

Fix is to unblock CQ pluck operations by cancelling calls, and thus allowing the CQ pluck to actually complete its operation. For server listening CQ operations, we unblock them by shutting down the server.

A side win here is to remove the [20ms-interval busy poll](https://github.com/grpc/grpc/blob/8564f72e8e0334c25c480e0aec1df75bdc1fce14/src/ruby/ext/grpc/rb_completion_queue.c#L44) on CQ operations, which was only needed to handle shutdown.

Closes grpc#36903

COPYBARA_INTEGRATE_REVIEW=grpc#36903 from apolcyn:fix_ruby_interrupt bed1ee2
PiperOrigin-RevId: 643046465

…ions handle signals and process shutdown #36903" (#36916)

backport #36903

…k operations handle signals and process shutdown grpc#36903" (grpc#36916)"

This reverts commit 3469d7b.

Fixes #37234

Following up on the problem described in #36903, there are a number of paths in `client_server_spec.rb` and a few other tests where client call objects can leak due to RPC lifecycles not being properly completed, leading to a thread not terminating.

Some of the tests, which don't use the surface-level APIs, are changed to manually close calls (and not rely on GC which might not happen before shutdown of ruby threads). `client_server_spec.rb` is updated to use surface level APIs, which manages call lifecycles correctly (this also improves the test's fidelity).

While we're here: expose `cancel_with_status` on call operations. This was only accidentally private so far. The test refactoring caught it.

Closes #37410

COPYBARA_INTEGRATE_REVIEW=#37410 from apolcyn:fix_call_leak b230472
PiperOrigin-RevId: 660430463

…ls and process shutdown (grpc#36903)

Fixes the CBF of `src/ruby/end2end/killed_client_thread_test.rb` (failure mode is a hang of the child process that receives the SIGTERM) that has been happening since grpc#36724

So far grpc-ruby CQ pluck operations have so far used a 20ms-interval busy poll to check interrupts in case we've received a signal, handle process shutdown, etc. This means ongoing RPCs will not terminate their CQ operations if we need to terminate the process (the loop simply exits without waiting for the CQ op to finish), causing a leak. Those RPCs can leave refs over their corresponding channels preventing [this](https://github.com/grpc/grpc/blob/8564f72e8e0334c25c480e0aec1df75bdc1fce14/src/ruby/ext/grpc/rb_channel.c#L653) from terminating (the channels don't reach state SHUTDOWN after being destroyed).

Fix is to unblock CQ pluck operations by cancelling calls, and thus allowing the CQ pluck to actually complete its operation. For server listening CQ operations, we unblock them by shutting down the server.

A side win here is to remove the [20ms-interval busy poll](https://github.com/grpc/grpc/blob/8564f72e8e0334c25c480e0aec1df75bdc1fce14/src/ruby/ext/grpc/rb_completion_queue.c#L44) on CQ operations, which was only needed to handle shutdown.

Closes grpc#36903

COPYBARA_INTEGRATE_REVIEW=grpc#36903 from apolcyn:fix_ruby_interrupt bed1ee2
PiperOrigin-RevId: 643046465

Fixes grpc#37234

Following up on the problem described in grpc#36903, there are a number of paths in `client_server_spec.rb` and a few other tests where client call objects can leak due to RPC lifecycles not being properly completed, leading to a thread not terminating.

Some of the tests, which don't use the surface-level APIs, are changed to manually close calls (and not rely on GC which might not happen before shutdown of ruby threads). `client_server_spec.rb` is updated to use surface level APIs, which manages call lifecycles correctly (this also improves the test's fidelity).

While we're here: expose `cancel_with_status` on call operations. This was only accidentally private so far. The test refactoring caught it.

Closes grpc#37410

COPYBARA_INTEGRATE_REVIEW=grpc#37410 from apolcyn:fix_call_leak b230472
PiperOrigin-RevId: 660430463

…le signals and process shutdown (grpc#36903)"

This reverts commit d4b5e12.