Instead of creating an mmap per worker process, we should create a single mmap in the main process, and then just slice a memoryview of this mmap in each worker.
Should be rather close performance wise (benchmarked locally) but would make the code cleaner.
Instead of creating an mmap per worker process, we should create a single mmap in the main process, and then just slice a
memoryviewof thismmapin each worker.Should be rather close performance wise (benchmarked locally) but would make the code cleaner.