Investigate removing pandas dependency from Core Airflow

[ashb]

I want to take a look at removing numpy/pyarrow from core airflow. We don't really need it, and it would make installing easier.

After 2.0

Originally posted by @ashb in #11950 (comment)

[turbaszek]

+1 for that, I faced some dependency issues with pyarrow too

[potiuk]

Oh yeah! That will help with #11950 as well!

[turbaszek]

Searching for pandas import it seems that only DbApiHook requires it explicitly for one method: get_pandas_df.
Apart from that:

root@1572924b0714:/opt/airflow# pipdeptree -r -p pandas
pandas==1.1.0
  - apache-airflow==2.0.0b3 [requires: pandas>=0.17.1,<2.0]
  - nteract-scrapbook==0.4.1 [requires: pandas]
  - pandas-gbq==0.13.2 [requires: pandas>=0.19.0]

Where nteract-scrapbook is required for papermill provider and pandas-gbq is for google. I removed pandas from setup.cfg and using clean venv installed airflow. No pandas was installed and running few airflow commands showed no errors.

[ashb]

It probably used to be used more - in the old chart view I think, but we've removed those uses now.

[ashb]

And no pandas means no numpy right?

[turbaszek]

And no pandas means no numpy right?

Yes, removing pandas will remove also numpy 👌

[jchacks]

airflow/airflow/utils/json.py

Line 21 in 7d38645

import numpy as np

That line in utils imports numpy, not sure if this is relevant.
It was causing my webserver to crash since my airflow version did not install numpy by default, for some reason.