Continuations: The magic behind virtual threads in Java by Balkrishna Rawool

Virtual threads are lightweight user threads introduced in Java 21 that provide high scalability by requiring significantly less memory than platform threads

Continuations are the underlying mechanism enabling virtual thread functionality - they represent the current state of a program and allow pausing/resuming execution

Virtual threads use a many-to-many relationship with platform threads (carrier threads) - multiple virtual threads can share a small pool of platform threads

When a virtual thread is blocked (e.g. waiting for I/O), it gets unmounted from the platform thread, allowing other virtual threads to use that platform thread efficiently

Virtual threads should not be pooled since they are cheap to create and destroy, unlike platform threads which are expensive OS resources

The mounting/unmounting behavior of virtual threads is what enables their high scalability - platform threads aren’t wasted while virtual threads are blocked

Virtual threads require a platform thread to execute but can be unmounted when waiting, making them ideal for I/O-bound applications

Continuations work by manipulating the call stack - saving stack frames to heap memory when yielding and restoring them when resuming

You can create thousands or millions of virtual threads since they have low memory overhead compared to platform threads

The default number of carrier (platform) threads is set to match CPU cores, but can be configured via JVM arguments