RIP Python 2

[Citron]

Hello

Ubuntu 20.04 will come in one month, killing Python2 support, and I'll probably upgrade all our servers.

I see that Dataiku depends heavily on Python 2 and still does not provide Python 3.7 integration.

As all my deep learning code base is now Python 3.7 compatible, I won't use Dataiku before it has evolved to the now default Python 3 standard.

May you please tell me what are your plans and timeline ? 

[Clément_Stenac · Accepted solution]

Hi,

Dataiku fully supports Python 3 and does not depend on Python 2 anymore (optionally).

We have settled on Python 3.6 as the baseline and the Python 3 for which we provide prebuilt installation support, as it is the golden Python version for Redhat / CentOS, our main devleopment target.

We are looking into Python 3.7 support for code environments, but the base environment will still require Python 3.6 for the foreseeable future, as we do not plan to ship binary packages for all Python versions.

In order to install DSS with Python 3 as the base environment, add the -P python3 or -P python3.6 flag to the installer

[Clément_Stenac · Accepted solution]

Hi,

Dataiku fully supports Python 3 and does not depend on Python 2 anymore (optionally).

We have settled on Python 3.6 as the baseline and the Python 3 for which we provide prebuilt installation support, as it is the golden Python version for Redhat / CentOS, our main devleopment target.

We are looking into Python 3.7 support for code environments, but the base environment will still require Python 3.6 for the foreseeable future, as we do not plan to ship binary packages for all Python versions.

In order to install DSS with Python 3 as the base environment, add the -P python3 or -P python3.6 flag to the installer

[MRvLuijpen]

Hi, just a follow-up question. Can this option -P python or -P python3.6 also be used in an updating DSS environment, or only with a new installation of DSS?

Thanks in advance 

Marc Robert

[Clément_Stenac · Accepted solution]

Hi,

It is possible to use it on an existing instance (you'll need to remove "pyenv" first for DSS to recreate it).

We would NOT recommend doing that, however, since:

• You will likely break existing user code written in the base env
• Python pickles are not cross-Python-version compatible, so you'll break usage of trained models with the Python scoring engine
• For the same reason, you may break artifacts saved by users as pickles in managed folders