如何在 Python 3 中的 Databricks 中使用 python 雪花连接器连接到雪花？

Question

When I try to attach the snowflake-sqlalchemy library to a Python 3 cluster in Databricks it breaks my python build and it gives me the following error when I install subsequent libraries:

AttributeError: cffi library '_openssl' has no function, constant or global variable named 'Cryptography_HAS_ED25519'

I have tried attaching the latest version of the Cryptography library to the cluster separately however this gave me the same issue. I think it might be related to the following links:

connecting-to-snowflake-from-azure-databricks-notebook-message-openssl-has-no-function-constant-or-global-variable-named-cryptography

https://github.com/snowflakedb/snowflake-connector-python/issues/32

In the second link it mentions a workaround:

The workaround is:
Uninstall cryptography by running pip uninstall cryptography
Delete the directory .../site-packages/cryptography/ manually
Reinstall snowflake-connector-python

Looks like the directory structure of cryptography changed since 1.7.2.*

Is there any way to uninstall the pre-installed cryptography 1.5 python library within Databricks so that I can reinstall the latest version of cryptography (2.5) with the new directory structure?

Answer 1

Out of date libraries:

%sh sudo apt-get install python3-pip -y

Followed by:

%sh pip3 install --upgrade snowflake-connector-python

See https://datathirst.net/blog/2019/1/11/databricks-amp-snowflake-python-errors for more detail.

Answer 2

I have found an answer to my problem.

The issue is caused by the version of openssl in Databricks being too out of date for snowflake-sqlalchemy to work with it.

The solution is as follows:

1. Upgrade PIP

   %sh /databricks/python/bin/pip install --upgrade pip

2. Uninstall pyopenssl

   %sh /databricks/python/bin/pip uninstall pyopenssl -y

3. Install pyopenssl

   %sh /databricks/python/bin/pip install --upgrade pyopenssl

4. Install snowflake-sqlalchemy

   %sh /databricks/python/bin/pip install --upgrade snowflake-sqlalchemy

The answer to this question was helpful: Python AttributeError: 'module' object has no attribute 'SSL_ST_INIT'

I have created an init file using the following code:

dbutils.fs.mkdirs("dbfs:/databricks/init/")

dbutils.fs.put("dbfs:/databricks/init/sf-initiation.sh" ,"""
#!/bin/bash
/databricks/python/bin/pip install --upgrade pip
/databricks/python/bin/pip uninstall pyopenssl -y
/databricks/python/bin/pip install --upgrade pyopenssl
/databricks/python/bin/pip install --upgrade snowflake-sqlalchemy
""", True)

The last command in the file updates all outdated packages as in: Upgrading all packages with pip