Trouble installing scipy in virtualenv on a amazon ec2 linux micro instance

Question

I have successfully installed scipy in the default python compiler on an amazon ec2 micro instance (Ubuntu 13.04). However i am not able to install scipy in a virtualenv.

pip install scipy ends with this error

scipy/sparse/sparsetools/csr_wrap.cxx: In function ‘void init_csr()’:

scipy/sparse/sparsetools/csr_wrap.cxx:73303:21: warning: variable ‘md’ set but not used [-Wunused-but-set-variable]

c++: internal compiler error: Killed (program cc1plus)

Please submit a full bug report,

with preprocessed source if appropriate.

See <file:///usr/share/doc/gcc-4.7/README.Bugs> for instructions.

----------------------------------------
Cleaning up...
Command /home/ubuntu/pnr/bin/python -c "import setuptools;__file__='/home/ubuntu/pnr/build/scipy/setup.py';exec(compile(open(__file__).read().replace('\r\n', '\n'), __file__, 'exec'))" install --record /tmp/pip-t8Drvd-record/install-record.txt --single-version-externally-managed --install-headers /home/ubuntu/pnr/include/site/python2.7 failed with error code -9 in /home/ubuntu/pnr/build/scipy

and

Traceback (most recent call last):
  File "/home/ubuntu/pnr/bin/pip", line 9, in <module>
    load_entry_point('pip==1.4.1', 'console_scripts', 'pip')()
  File "/home/ubuntu/pnr/local/lib/python2.7/site-packages/pip/__init__.py", line 148, in main
    return command.main(args[1:], options)
  File "/home/ubuntu/pnr/local/lib/python2.7/site-packages/pip/basecommand.py", line 169, in main
    text = '\n'.join(complete_log)
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 53: ordinal not in range(128)

Before anyone asks. pip freeze for default compiler returns

Cheetah==2.4.4
Landscape-Client==12.12
M2Crypto==0.21.1
PAM==0.4.2
Pillow==2.0.0
PyYAML==3.10
Twisted-Core==12.3.0
Twisted-Names==12.3.0
Twisted-Web==12.3.0
apt-xapian-index==0.45
argparse==1.2.1
boto==2.3.0
chardet==2.0.1
cloud-init==0.7.2
configobj==4.7.2
distribute==0.6.34
distro-info==0.10
euca2ools==2.1.1
numpy==1.7.1
oauth==1.0.1
paramiko==1.7.7.1
prettytable==0.6.1
pyOpenSSL==0.13
pycrypto==2.6
pycurl==7.19.0
pygobject==3.8.0
pyserial==2.6
python-apt==0.8.8ubuntu6
python-debian==0.1.21-nmu2ubuntu1
requests==1.1.0
scipy==0.11.0
six==1.2.0
ssh-import-id==3.14
urllib3==1.5
virtualenv==1.10.1
wsgiref==0.1.2
zope.interface==4.0.5

pip freeze command for virtualenv returns

Cython==0.19.2
Flask==0.10.1
Flask-Bootstrap==3.0.0.1
Flask-WTF==0.9.3
Jinja2==2.7.1
MarkupSafe==0.18
WTForms==1.0.5
Werkzeug==0.9.4
argparse==1.2.1
beautifulsoup4==4.3.2
itsdangerous==0.23
numpy==1.7.1
pymongo==2.6.2
requests==2.0.0
wsgiref==0.1.2

Answer 1

One solution is to temporarily enable swap on your micro instance. As described at this SO post, enable 1gb swap via:

sudo /bin/dd if=/dev/zero of=/var/swap.1 bs=1M count=1024
sudo /sbin/mkswap /var/swap.1
sudo /sbin/swapon /var/swap.1

Once swap is on, install scipy via pip:

sudo apt-get install -y libatlas-base-dev gfortran python-dev build-essential g++
sudo pip install numpy
sudo pip install scipy

Once scipy successfully installs, you can disable it via:

sudo swapoff /var/swap.1
sudo rm /var/swap.1

Answer 2

This worked for me:

pip --no-cache-dir install scipy

See:

• Memory error while using pip install Matplotlib

  https://github.com/pypa/pip/blob/9a23d4ed119327d3b823ec223aaead90964bac58/pip/basecommand.py#L56-L63

• https://github.com/pypa/pip/blob/28cca11e284b37cc2c7977fd25be6f494adda9d3/src/pip/_internal/download.py#L359-L367

note:

• works for other service providers, hardware, VMs, and containers.
• if RAM allocation size of 1GB
• just calculate the diff between the cached directory memory usage and available ram

Answer 3

Yes, 512MB is not enough for compiling that C++ file.

Your best option is to build Scipy as a binary package (bdist, or eggs, or, more modern wheels) e.g. via python setupegg.py bdist_egg on a different machine with compatible environment. For instance, use a similar Linux version to the EC2 instance in a virtual machine.

In general, it's good to remember that when pip installs packages, it compiles source files. If the package is not tiny, this is inefficient and it's better to use binary packages. The wheel package format is supposed to play well together with pip.