Use Vagrant to automatically configure a Ubuntu VM from scratch and bootstrap it with SaltStack to install the SciPy stack with a mix of Ubuntu packages and virtualenvs.
To get started install:
- Virtualbox (Virtual Machine runner)
- Vagrant (Command-line tool to control VMs)
It can be helpful to tell vagrant to automatically install the virtualbox guest additions for the matching version to solve shared folder issues:
[host]$ vagrant plugin install vagrant-vbguest
Then you can start the VM with:
[host]$ vagrant up
The first start takes a couple of minutes to download the VM image, boot it,
and install most of the Scipy stack packages with Salt. Subsequent calls to
vagrant up
will be much faster.
Once started you can ssh into the VM with:
[host]$ vagrant ssh
SSH agent forwarding is enabled by default to be able to use git from inside the VM if needed.
Once there you can activate one of the virtualenvs, for instance for the Python 3.4 virtualenv:
[guest]$ workon 3.4
(3.4)[guest]$ ipython
You can suspend / resume the VM with to temporarily free the memory on the host
with vagrant suspend
/ vagrant resume
. Or shutfown and destroy the VM
completely to also free the disk with:
[host]$ vagrant destroy
Edit the Vagrantfile
to customize:
- VM image (Ubuntu 13.04 64 bits by default)
- VM RAM size
- shared folders
- port forwarding
The configuration is a bunch of text files in:
salt/roots/salt/
In particular the Ubuntu packages and virtualenv configuration is in:
salt/roots/salt/python-venvs.sls
Read the SaltStack documentation to get started with custom state files.
If you edit the salt configuration you can ask salt to quickly take all your changes into accout from inside the VM:
[guest]$ sudo salt-call state.highstate
Alternatively you can reboot the whole VM with from outside of the VM:
[host]$ vagrant reload
You can even shutdown and delete the VM to restart the provisioning from scratch with:
[host]$ vagrant destroy
[host]$ vagrant up
To speedup the dev setup, this repo comes with a wheelhouse folder with pre-built
numpy and scipy whl
packages.
To build wheel packages for newer versions of numpy and scipy you can use (for instance with Python 3.4):
[guest]$ workon 3.4
(3.4)[guest]$ pip wheel --download-cache=~/pip-cache \
--wheel-dir=/vagrant/wheelhouse numpy
(3.4)[guest]$ pip install /vagrant/wheelhouse/numpy-*-cp34-cp34m-linux_x86_64.whl
(3.4)[guest]$ pip wheel --download-cache=~/pip-cache \
--wheel-dir=/vagrant/wheelhouse scipy
(3.4)[guest]$ pip install /vagrant/wheelhouse/scipy-*-cp34-cp34m-linux_x86_64.whl