Nested parallelism is now enabled where appropriate. This is a minor performance hit when not using nested parallelism (the default) but allows for very simple control and reduces the number of compilation variations. Compilation is beginning to get quite slow, but the next numba release may help. This nested parallelism should help in cases where chunks are very large and running many dask threads is problematic (KGB as an example).
Nested parallelism is now enabled where appropriate. This is a minor performance hit when not using nested parallelism (the default) but allows for very simple control and reduces the number of compilation variations. Compilation is beginning to get quite slow, but the next numba release may help. This nested parallelism should help in cases where chunks are very large and running many dask threads is problematic (KGB as an example).