pyodbc doesn't have any async capbiltities. the only database among those supported that itself has async capabilitiy as all is postgresql. there's some discussion about postgresql with pyodbc and async at mkleehammer/pyodbc#348 and essentially there's no real rationale for it to support this.

there's no actual reason to use gevent with relational database access however and it will only make your code less reliable and more difficult to debug. I speak from considerable experience in this area.

@zzzeek I agree the code can become less reliable and difficult to debug. However, I believe there will be performance improvements for the application in general due to the async approach if the application is I/O bound.

For example, here http://blog.devsda.fr/python-gevent.html they reported significant performance improvements when using gevent with pymssql.

So I can't get you gevent support here, and also my studied opinion is that it's very hard to make Python IO bound with a high performance database on a local network, as it is not Go, Erlang, Node, or C, it's interpreted Python and has very high CPU overhead relative to compiled languages , see https://techspot.zzzeek.org/2015/02/15/asynchronous-python-and-databases/ , http://calpaterson.com/async-python-is-not-faster.html etc. I have yet to see a real benchmark showing async Python to be faster.

Also with the monkeypatching approach you put yourself in great danger of CPU bound processes hanging up your database, I've dealt with an actual outage of that nature, and additionally I've dealt with several across multiple applications which were directly caused by the usage of eventlet or gevent which would then pile up requests that were blocked due to mis-configuration of the number of greenlets relative to what the database was tailored for.

@zzzeek thx for the explanation

in case it helps SQLAlchemy 1.4 will have native asyncio support. no path for pyodbc that I'm familiar with though :(. closing this for now.

ok thx