I want to update you about Motor, my asynchronous Python driver for MongoDB and Tornado. Motor development has been on hiatus since January while I concentrated on my tasks for PyMongo 3. After PyMongo 3, I took over as the lead for libmongoc, the MongoDB C Driver, and most of my Python work ceased.
Spring was a good time for me to take a break from Motor. It is actually possible for Motor to be finished, at least for periods. It does one thing well: it integrates MongoDB with Tornado, probably the most popular Python async framework. Motor's limited scope lets it rest peacefully for months, feature-complete and free of major bugs. While Motor quiesced like this, I devoted myself to writing the Server Discovery And Monitoring Spec, then to implementing that spec in PyMongo 3, and then to relearning C fast enough to keep up with libmongoc's demands.
At the time I set Motor aside, I had begun integrating it with Python 3.4's new asyncio module, so you could choose between asyncio and Tornado. I adapted a portion of the Motor-on-Tornado test suite into a Motor-on-asyncio suite and got it passing, with good confidence that the rest of the suite would pass once ported. The remaining tasks to make the asyncio integration production-ready were certainly tractable. Meanwhile, I had developed substantial features and bugfixes on the Motor 0.4 release branch, so the next step was to merge the two branches. But with asyncio not yet widely used, and more urgent work looming, I put down the asyncio integration. That was where I expected it to sit until autumn, if not longer.
But I recently received two surprise gifts. First, in April, Rémi Jolin stepped in and did the hard work of merging 97 commits from the Motor 0.4 release branch into the asyncio branch, and fixing up the result: now Motor's master branch has both the asyncio integration and the fixes and features needed to become Motor 0.5.
My second gift arrived in June when Andrew Svetlov, one of asyncio's main authors, wrote to say he wants to finish the integration so he can use asyncio with MongoDB in production. Andrew and his colleague Nikolay Novik at DataRobot in Ukraine are porting the remainder of Motor's tests to asyncio, and they'll refactor how Motor uses asyncio's streams to benefit from the framework's latest features.
"async" and "await"
So Motor speeds forward, with my hand only lightly on the wheel. I hope for an asyncio-compatible Motor 0.5 this fall.
While I'm at it, it should be an easy win, and a huge one, to add support for Python 3.5's new "async" and "await" keywords defined in PEP 492. So I plan to add support for them in Motor 0.5 as well—the enhancements will make Motor cursors faster and more convenient with asyncio and Tornado!
The new Motor will still wrap the outdated PyMongo 2.8, however. So my next priority, after Motor 0.5 is released, is to port Motor to PyMongo 3 to take advantage of PyMongo's new features: its implementation of the Server Discovery and Monitoring Spec, the Server Selection Spec, and the new CRUD API. This will make Motor much more scalable when connected to large replica sets, and makes its API consistent with our other drivers. That may merit the name Motor 1.0.
And after that? In the next year, Motor will need new features to fully support MongoDB 3.2.
Farther out, my work on asyncio integration has convinced me that it's feasible for Motor to someday work with Twisted, too. PyMongo already supports Gevent, so if Motor does Twisted then all major async frameworks will be supported by official drivers. There will be no more excuse for async Python applications not to use MongoDB.