An unopinionated asynchronous DAG DSL and executor built into Python.
Oftentimes, Python is not the core infrastructure powering large systems. At a certain point, Python buckles under its own weight; its runtime was simply not designed to allow for the control necessary for certain levels of load. Instead, if Python is used, it is used as a glue that binds together other components of the system. This shift signifies a switch from a CPU-bound workload to an IO-bound workload.
There are myriad existing approaches to extracting more performance out of IO-bound workloads in Python, like the built in asyncio, gevent, and others. In general, these approaches have one of two problems:
-
They require that you aggressively buy in to their specific async system
-
They provide insufficient structure, allowing programmers to create an inefficient program
graft attempts to address these issues for request/response-style programs by:
-
Structuring the API to allow for gradual rewrites; graphs are recursive structures where subgraphs are first class objects. Parts of the program can be rewritten as a self-contained graph and subsequently added to a larger graph as a migration continues.
-
Aggressively decoupling code from its execution order; many of the inefficiencies from IO-bound systems come from overhead encouraged by a procedural language. In practice programmers are often trying to describe a system of dependent IO operations. By offloading thework of scheduling an executing these operations, programmers can increase the parallelism of their IO waits.
This package is not on pip, sorry :(. I'll get around to it eventually.
$ git clone https://github.com/crockeo/graft
$ cd graft
$ pip install .We'll have more here soon, but:
-
probably use
virtualenv! this is written on Python 3.8 and you might not want to install that on your system directly -
please install pre-commit, and do
pre-commit installin the repo directory
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent