Running the same docker config a few months later results in "ValueError: numpy.ndarray size changed"

Question

We've built a docker image in Summer '21 and it ran fine until December. Now, after having deleted the docker cache, it doesn't run anymore.

The error is:

ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

There are quite a lot of related SO questions. The answers tend to be similar:

• reinstall numpy
• use numpy version X
• use --no-cache-dir option on pip install
• use a new pip version

We've tried all of it (upgrading pip by adding the latest version to requirements.txt). A specific numpy version was not in our requirements.txt, I thought that could be the cause here, but after adding a few very different versions without success, I am out of ideas.

We're not even sure whether purging the docker cache played a role.

This is the minimal repro:

main.py:

from string_grouper import match_strings, match_most_similar

requirements.txt:

string_grouper==0.1.1

Dockerfile:

FROM amancevice/pandas:1.3.1-slim

# for string_grouper
RUN DEBIAN_FRONTEND=noninteractive apt-get update \
    && apt-get install g++ -y \
    && apt-get clean

ENV PYTHONDONTWRITEBYTECODE=1
ENV PYTHONUNBUFFERED=1

COPY requirements.txt .
RUN python -m pip install --no-cache-dir -r requirements.txt

WORKDIR /app
COPY . /app
RUN useradd appuser && chown -R appuser /app
USER appuser

CMD ["python", "main.py"]

What else can we try to troubleshoot this?

EDIT: I believe the likeliest solution is "explicitly install numpy version X". In that case, the question would just be how to find the right one.