Page History

CHTC has instructions for making a tarball of python packages (e.g. numpy, pyuvdata, etc).  This tarball is then either transferred to the execution host if it is under 100MB or downloaded from their web proxy server (with prior authorization) if 100MB or over.  This, coupled with a python installation via tarball, may be enough for execution hosts to run HERA calibration jobs.

Here is a sample set of instructions to build a python-packages tarball

cat > build.sub <<EOM

# Python build file

universe = vanilla

log = interactive.log

# Choose a version of Python from the table above

transfer_input_files = http://proxy.chtc.wisc.edu/SQUID/chtc/python37.tar.gz

+IsBuildJob = true

requirements = (OpSysMajorVer =?= 7)

request_cpus = 1

request_memory = 2GB

request_disk = 2GB

queue

EOM

condor_submit -i build.sub

tar xfz python37.tar.gz

export PATH=${PWD}/python/bin:${PATH}

export PYTHONPATH=${PWD}/hera_calibration_packages

mkdir -p ${PYTHONPATH}

python3 -m pip install --target ${PYTHONPATH} numpy scipy astropy h5py toml six matplotlib scikit-learn cached-property

python3 -m pip install --target ${PYTHONPATH} healpy

python3 -m pip install --target ${PYTHONPATH} pyuvdata

python3 -m pip install --target ${PYTHONPATH} aipy

# trying more packages.  This seems to work

python3 -m pip install --target ${PYTHONPATH} git+https://github.com/HERA-Team/pyuvdata.git

python3 -m pip install --target ${PYTHONPATH} git+https://github.com/HERA-Team/linsolve.git

python3 -m pip install --target ${PYTHONPATH} git+https://github.com/HERA-Team/hera_qm.git

python3 -m pip install --target ${PYTHONPATH} git+https://github.com/HERA-Team/uvtools.git

python3 -m pip install --target ${PYTHONPATH} git+https://github.com/HERA-Team/hera_cal.git

# plaplant said "hera_opm is only used for setting up makeflow. so the

# execution host itself does not need it

tar cfz hera_calibration_packages.tar.gz hera_calibration_packages

exit