I tend to do that with multiprocessing queues, but at some point the serialisation becomes the bottleneck. Another way is to use a faster serialiser (marshal,

I tend to do that with multiprocessing queues, but at some point the serialisation becomes the bottleneck. Another way is to use a faster serialiser (marshal, ujson) and write them to files on SSDs with flock, but again a trick that's slower than real datasharing like java and go.