Vertica is a very powerful analytics database, and we can easily extend functionality now by building in Python functions. This is great and all, so here I'll focus on setting up a development environment for building a simple UDx.

The documentation from Vertica is not super specific, so this may help!

For the local setup, we'll be using https://github.com/jbfavre/docker-vertica/. I'm going to focus on the CentOS version, since it is most similar to Amazon's RHEL images. To get started (on OSX, you need the Docker client first):

docker pull jbfavre/vertica:9.2.0-7_centos-7

Also take a second to get Python 3.5.1 here. Those are the two steps that require the most bandwidth!

Now, run that thing:

docker run -p 5433:5433 jbfavre/vertica:9.2.0-7_centos-7

Grab the name of the running container from docker ps command, and set it to $DOCKER_NAME, and copy in our Python source:

DOCKER_NAME=suspicious_chatelet
docker cp Downloads/Python-3.5.1.tgz $DOCKER_NAME:/home/dbadmin/

Phew! We're done with the parts that are specific to a local setup. If you're working on a server somewhere, copy up the Python-3.5.1.tgz and put it at /home/dbadmin/Python-3.5.1.tgz. For good measure (i.e., if you copied with the root user) make sure it's owned by dbadmin:

chown dbadmin /home/dbadmin/Python-3.5.1.tgz

As root, install all of the dependencies that you need for Python. WARNING: these may not be everything that YOU need (see alternative dependencies in an install like this) or rely on the python3 that is available with the yum installer directly.

yum install openssl-devel bzip2-devel expat-devel gdbm-devel readline-devel sqlite-devel

Grab a coffee.

Switch to dbadmin user and do:

cd /home/dbadmin/Python-3.5.1
./configure
make
make install

Probably another coffee for that one!

If you run the above as dbadmin, which you should, skip the make install line. Or only do that last line as root.

Maybe still as root, though again this should be fine (and preferable) as dbadmin user in the /home/dbadmin directory. If you used root for anything above other than make install, you might need to keep root user active. Here goes:

Python-3.5.1/python -m venv pyenv
pyenv/bin/pip install -U pip
pyenv/bin/pip install requirements.txt

If you're within a firewall, you might need --index-url=[your company's artifactory] for the pip install lines. What is the requirements.txt line? All of the packages that your function needs, just like the requirements file from any python package.

Finally you should be able to execute the SQL needed to build your function! Put your code in myfunction.py and then you can run vsql as dbadmin, and do:

\set libfile '\''`pwd`'/myfunction.py\''
DROP LIBRARY mylib CASCADE;

CREATE LIBRARY mylib AS :libfile DEPENDS '/home/dbadmin/pyenv/lib/python3.5/site-packages/' LANGUAGE 'Python';
CREATE FUNCTION myfunction AS NAME 'myfunction_factory' LIBRARY mylib;

If you run into permissions issues having done all of the Python stuff as root, you can get the function to build by opening up the virtual environment permissions using these commands (as root in /home/dbadmin):

find pyenv -type f -exec chmod 666 {} \;
find . -type d -exec chmod 777 {} \;

Cheers! I could build the python3 and package that into a Docker based off the original, but that would be less fun for you.