shenandoahArguments adjust the safepoint timeouts for debug builds:

  // Make sure safepoint deadlocks are failing predictably. This sets up VM to report
  // fatal error after 10 seconds of wait for safepoint syncronization (not the VM
  // operation itself). There is no good reason why Shenandoah would spend that
  // much time synchronizing.
#ifdef ASSERT
  FLAG_SET_DEFAULT(SafepointTimeout, true);
  FLAG_SET_DEFAULT(SafepointTimeoutDelay, 10000);
  FLAG_SET_DEFAULT(AbortVMOnSafepointTimeout, true);
#endif

It was introduced before integration:
 https://mail.openjdk.java.net/pipermail/shenandoah-dev/2018-October/007733.html

On some of the busy servers with parallel tests, some threads are actually that late for safepoint synchronization! So this block introduces false failures, and should be removed.