The Limits of Parallelization of PG Dump

It's just how parallelization of PG dump is implemented. At database level, not, you cannot not realize single table unless it's partitioned. And a couple of years ago, a special option for PG bench was added. So now you can create when you create PG bench database dash I initialize it, you can say I wanted to have partitioned, did you bench accounts? And then you can speed up your dump. This is interesting example showing that like, there are many places where parallelization can be tricky. To go full speed, especially if you have very curious, you definitely need multiple threads or multiple processes and the communication between processes.