Segfaults in CI unit tests

IBM / depiction

interpret deep learning models in a framework-independent fashion

Apache License 2.0

41 stars 21 forks source link

Segfaults in CI unit tests #70

Open drugilsberg opened 4 years ago

drugilsberg commented 4 years ago

Updgrade to tensorflow==1.15.0 causes all the interpreters tests to raise a segfault. Something might have changed in the docker image we start from or in the dependencies we don't pin.
Even with tensorflow==1.14.0, the imports for PaccMann example module causes segfault.

Linked to #8

drugilsberg commented 4 years ago

Lines in the Dockerfile that might be problematic:

FROM continuumio/miniconda3
# labels
LABEL maintainer="Matteo Manica <drugilsberg@gmail.com>, An-phi Nguyen <nguyen.phineas@gmail.com>, Joris Cadow <joriscadow@gmail.com>"
# needed settings
ENV PATH /opt/conda/bin:$PATH
ENV LANG C
# install system dependencies
RUN apt-get update \
    && apt-get install -y --no-install-recommends gcc g++ python3-dev libxrender-dev\
    && rm -rf /var/lib/apt/lists/*

drugilsberg commented 4 years ago

Reproduce the error. Just build docke image and run the tests as follows:

docker-compose -f .travis/docker-compose.yml up -d
# paccmann
docker exec -it depiction-test python3 -m unittest discover -v -t /build -p "*_test.py" /build/depiction/models/examples/paccmann/
# interpreters when bumping 1.15.0
docker exec -it depiction-test python3 -m unittest discover -v -t /build -p "*_test.py" /build/depiction/interpreters/