Why don't companies use Python (programming language) for Apache Hadoop programming?

Most of Hadoop written in Java.  Most extensions and native APIs are simply extended using Java which is why you see so much code for Hadoop in Java.  However, it is simply not true that Python is scarcely used with Hadoop.  When I was doing consulting work for Facebook, we used a couple of Python frameworks almost exclusively for our ETL jobs within the Hadoop clusters.  I wrote some rather complex custom reducers using python through Hive.  There's also plenty of map-reduce libraries available for Python.

I think it's  because companies who have their own Apache stack already have competent Java data scientists/engineers. If you're already good with Java, learning python might not be the most productive thing for the company. On top of that, once a majority of your work is in Java, having someone come in and write streaming python map-reduce just makes the whole less consistent and maintainable for future developers.