Python企业开发的10个误区

jieforest

It might sound simple, but even for seasoned engineers, this can be a very time-consuming process. Python was designed from the ground up with developer timelines in mind. In our experience, it’s not uncommon for Python projects to undergo three or more iterations in the time it C++ and Java to do just one. Today, PayPal and eBay have seen multiple success stories wherein Python projects outperformed their C++ and Java counterparts, with less code(see right), all thanks to fast development times enabling careful tailoring and optimization. You know, the fun stuff.

jieforest

Myth #8: Python lacks good concurrency support

Occasionally debunking performance and scaling myths, and someone tries to get technical, “Python lacks concurrency,” or, “What about the GIL?” If dozens of counterexamples are insufficient to bolster one’s confidence in Python’s ability to scale vertically and horizontally, then an extended explanation of a CPython implementation detail probably won’t help, so I’ll keep it brief.

Python has great concurrency primitives, including generators, greenlets, Deferreds, and futures. Python has great concurrency frameworks, including eventlet, gevent, and Twisted. Python has had some amazing work put into customizing runtimes for concurrency, including Stackless and PyPy. All of these and more show that there is no shortage of engineers effectively and unapologetically using Python for concurrent programming. Also, all of these are officially support and/or used in enterprise-level production environments. For examples, refer toMyth #7.

The Global Interpreter Lock, or GIL, is a performance optimization for most use cases of Python, and a development ease optimization for virtually all CPython code. The GIL makes it much easier to use OS threads or green threads (greenlets usually), and does not affect using multiple processes. For more information, see this great Q&A on the topic and this overview from the Python docs.

jieforest

Here at PayPal, a typical service deployment entails multiple machines, with multiple processes, multiple threads, and a very large number of greenlets, amounting to a very robust and scalable concurrent environment (see figure below). In most enterprise environments, parties tends to prefer a fairly high degree of overprovisioning, for general prudence and disaster recovery. Nevertheless, in some cases Python services still see millions of requests per machine per day, handled with ease.