Installing pylibmc on Heroku
That which is not documented will cause you grief
pylibmc is a memcached client
for Python. It’s a requirement for adding caching with memcached to a
Python application, like a Django project. However you can find
where people have had difficulty installing it on Heroku.
The solution is actually mentioned in that first link, but I missed the significance of it until checking out the buildback source code.
The problem I was having, and all of these others folks, too, is that I
was using multiple pip requirements files. As you can see on line 2, the
buildback checks for pylibmc in one place,
requirements.txt, the base
file. My root
requirements.txt was used only to inlclude
pylibmc was never detected, so
libmemcached was never bootstrapped, and install failed. The solution
is just moving this requirement to the base file.
requirements.txt file now looks like this:
pylibmc==1.3.0 -r requirements/production.txt
About multiple requirements files
One response to this problem was that multiple pip requirements files add more ‘overhead’ and breaks dev/production parity.
I fail to see what kind of overhead a few well organized requirements
add to a project. If nothing else, it’s extremely convenient to be able
to break out app requirements from development requirements. For a
Django project, the core requirements would go into a
file and requirements like
Sphinx (you do document, right?) and
flake8 go into a
requirements/test.txt file. This lets you tell
other developers to install from one pip requirements file, instead of
checking off each that they need. Same with a CI server.
As for dev/production parity, this shouldn’t be a concern for the aforementioned requirements. Especially when slug size is a potential concern on Heroku, don’t install stuff in production you don’t need to run the application. It’s as simple as that. And on some teams, some projects, breaking that perfect dev/production parity might be acceptable. This is a tradeoff between environment parity and environment overhead. Mac to Linux already breaks this parity, so you need to assume you’re working with VM’s, e.g. Vagrant, using the same OS at your target production system. That’s a brilliant idea, but for simpler, short-run projects not always necessary.
Update: the excellent folks at Memcachier have since updated their documentation to make this explicit.