Rewrite PessimisticLockingTest#test_no_locks_no_wait to avoid costly sleeping

[fatkodima]

Follow up to #47499.

PessimisticLockingTest#test_no_locks_no_wait test uses costly sleeping for no reason and relies on the correct order of threads activating via sleeping. It wastes 5 seconds just for this test case.

Made a profile of this change.

Before

     TOTAL    (pct)     SAMPLES    (pct)     FRAME
     45395   (8.3%)       45395   (8.3%)     PG::Connection#exec_params
===> 34741   (6.4%)       34741   (6.4%)     Thread#join
     20520   (3.8%)       20520   (3.8%)     Thread::Mutex#sleep
     10580   (1.9%)       10580   (1.9%)     (marking)
      8755   (1.6%)        8755   (1.6%)     (sweeping)
      7380   (1.4%)        7378   (1.4%)     PG::Connection#exec_prepared
      6252   (1.1%)        6252   (1.1%)     IO#wait
      5662   (1.0%)        5662   (1.0%)     Kernel#sleep
      3989   (0.7%)        3989   (0.7%)     Process.waitpid
    164585  (30.2%)        3066   (0.6%)     Thread.handle_interrupt
      3036   (0.6%)        3035   (0.6%)     String#sub!

After

     TOTAL    (pct)     SAMPLES    (pct)     FRAME
     49190   (9.2%)       49185   (9.2%)     PG::Connection#exec_params
     20507   (3.8%)       20507   (3.8%)     Thread::Mutex#sleep
     10680   (2.0%)       10680   (2.0%)     (marking)
===> 10007   (1.9%)       10007   (1.9%)     Thread#join
      8952   (1.7%)        8952   (1.7%)     (sweeping)
      7134   (1.3%)        7134   (1.3%)     PG::Connection#exec_prepared
      6785   (1.3%)        6785   (1.3%)     IO#wait
      5665   (1.1%)        5665   (1.1%)     Kernel#sleep
      3992   (0.7%)        3992   (0.7%)     Process.waitpid
      3153   (0.6%)        3153   (0.6%)     String#sub!

[p8]

Ah, nice. I wasn't familiar with Thread::Queue#pop

If the queue is empty, the calling thread is suspended until data is pushed onto the queue.
https://rubyapi.org/3.2/o/thread/queue#method-i-pop

So this waits until Thread a pushes true into b_wakeup.

[matthewd]

Nice find!

We prefer Concurrent::Event, Concurrent::CyclicBarrier, and friends where appropriate, just for the extra clarity that we're not actually passing values through queues.

[fatkodima]

@matthewd Thank you for the suggestion! I rewrote it to use Concurrent::Event.