How to use cx_Oracle within AWS SageMaker (Jupyter Notebook)

Question

I've followed the official installation guide but haven't had any luck so far. I wonder if cx_Oracle can work on AWS SageMaker's virtual environment. The steps I've used so far are:

1. Create a /opt/oracle directory and unzip the basic instantclient in it.
2. sudo yum install libaio
3. sudo sh -c "echo /opt/oracle/instantclient_18_3 > /etc/ld.so.conf.d/oracle-instantclient.conf" and sudo ldconfig
4. And finally exported the LD_LIBRARY_PATH with: export LD_LIBRARY_PATH=/opt/oracle/instantclient_18_3:$LD_LIBRARY_PATH

When trying to run a connection inside the notebook with connection = cx_Oracle.connect(usr + '/' + pwd + '@' + url) , I receive the DPI-1047 error code that says that libclntsh.so cannot be open, however that library is in the /opt/oracle folder. As another option, when running the same connection through the terminal Python console, I get the ORA-01804 error code, which says that the timezone files were not properly read, which is something I'm trying to fix also but suspect is related to cx_Oracle not finding its library folder. (Now, explain to me: why does it have to be so difficult for a billionaire company to create a decent library import and installation?)

Is there a step I'm missing? Is there a detail about AWS SageMaker that I should account for? Also, is there another option for extracting data from an oracle server through Python and AWS?

Answer 1

Hi and thank you for using SageMaker!

After some effort, I was finally able to figure out a sequence of steps that allowed me to query an Oracle 12 database from within a SageMaker notebook instance. Here are the steps that I took:

1. I created an Oracle 12 database using Amazon RDS for testing purposes. (You can of course skip this step if you already have an Oracle database available.)
2. I downloaded the Oracle 12 Instant Client RPM as described here . Note that you will need an Oracle Account in order to download this file.
3. I uploaded the RPM to my SageMaker Notebook Instance from within JupyterLab. Note that this can take 2-3 minutes to completely upload before proceeding to the next step. (I initially had problems running the installation because the upload was still in progress.)
4. I ran all of the following commands from the Jupyter terminal as prescribed in the Oracle instructions:

cd SageMaker

sudo yum install oracle-instantclient12.2-basic-12.2.0.1.0-1.x86_64.rpm
sudo sh -c "echo /usr/lib/oracle/12.2/client64/lib > /etc/ld.so.conf.d/oracle-instantclient.conf"
sudo ldconfig
sudo mkdir -p /usr/lib/oracle/12.2/client64/lib/network/admin

# Restart Jupyter...
sudo restart jupyter-server

5. I then installed the cx_Oracle library:

source activate python3
pip install cx_Oracle --upgrade

6. Then lastly, I created a new notebook using the conda_python3 kernel:

import cx_Oracle
db = cx_Oracle.connect("my_username/my_password@my-rds-instance.ccccccccccc.us-east-1.rds.amazonaws.com/ORCL")

# Example query
cursor = db.cursor()
for row in cursor.execute('select * from DBA_TABLES'):
  print(row)

...and then it worked!

Note that it took me a few tries to figure out the exact connection string, which can differ depending on how your database is configured. Unfortunately, the error messages were hard to understand - in my case, I had the error ORA-12504: TNS:listener was not given the SERVICE_NAME in CONNECT_DATA until I had specified the /ORCL at the end of the connection string.

If you need to do these steps frequently, you can add the installation and configuration of the Oracle client in a SageMaker Lifecycle Configuration script. I haven't tested that scenario, but it might be worth a try!

One last thing, I noticed in your question that you are using the Oracle 18 client. I didn't test that exact scenario, since I only have access to an Oracle 12 database. However, the Oracle 12 client should be able to connect to an Oracle 18 database, too.

Best, Kevin