The JVM has something called "safe point" and "safe zone". When GC occurs, all threads will execute to the "safe point" and stop. When GC is required, the JVM will set a flag. When the thread reaches a safe point, it will poll and detect this flag. If it is found that GC is needed, the thread will hang on its own and will not resume running until the GC is completed.
Another strategy is to directly suspend all threads when GC occurs, and then detect whether all threads are at a safe point. If they are not at a safe point, resume the execution of the thread, and wait until the execution reaches a safe point before suspending.
But for some threads that have not obtained or cannot obtain CPU time, there is no way to wait until it reaches a safe point. Therefore, as long as the thread is in the safe area at this time, GC can also be performed. The safe area is a section of code. The reference relationship of the objects in this code segment will not change, so it is safe to perform GC at this time.
GC will not be very frequent, it will only occur when there is insufficient memory (of course it is not absolute), so stop the world will not be very frequent, which ensures that it will not affect user threads, and user threads have a safe zone , to stop the world, you need to wait for all threads to be in the safe zone. It is possible that some long-term operations will wait a long time to reach the safe zone. This means that stop the world may wait longer than the gc itself. The way to solve this problem is to set the safety zone reasonably. For example, there is a big difference between whether you can pause inside the loop and when the loop is completed. The specific setting of the security zone depends on the decision of the jvm manufacturer
![[Web front-end] Node.js quick start](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
The JVM has something called "safe point" and "safe zone". When GC occurs, all threads will execute to the "safe point" and stop.
When GC is required, the JVM will set a flag. When the thread reaches a safe point, it will poll and detect this flag. If it is found that GC is needed, the thread will hang on its own and will not resume running until the GC is completed.
Another strategy is to directly suspend all threads when GC occurs, and then detect whether all threads are at a safe point. If they are not at a safe point, resume the execution of the thread, and wait until the execution reaches a safe point before suspending.
But for some threads that have not obtained or cannot obtain CPU time, there is no way to wait until it reaches a safe point. Therefore, as long as the thread is in the safe area at this time, GC can also be performed. The safe area is a section of code. The reference relationship of the objects in this code segment will not change, so it is safe to perform GC at this time.
The Java method of safely pausing threads is not stable. I think the native method should be used during Full GC.
Get some attention and learn. .
Try to avoid full gc and reduce the use of large arrays.
GC will not be very frequent, it will only occur when there is insufficient memory (of course it is not absolute), so stop the world will not be very frequent, which ensures that it will not affect user threads, and user threads have a safe zone , to stop the world, you need to wait for all threads to be in the safe zone. It is possible that some long-term operations will wait a long time to reach the safe zone. This means that stop the world may wait longer than the gc itself. The way to solve this problem is to set the safety zone reasonably. For example, there is a big difference between whether you can pause inside the loop and when the loop is completed. The specific setting of the security zone depends on the decision of the jvm manufacturer