Sanitizing databases: PostgreSQL and Django
Back to home
On this page
Databases of live websites often contain personally identifiable information (PII) such as full names, mailing addresses, and phone numbers. To ensure people reviewing code changes can’t access information they shouldn’t, sanitize your databases of any PII that they may contain.
This example goes through the process for a PostgreSQL database using Django.
Before you begin
You need:
- A project with a PostgreSQL database.
- A command interface installed:
- If doing it manually, the Platform CLI.
- Otherwise, make sure
pqsl
is installed in your environment.
This guide is about sanitizing PostgreSQL databases.
This guide doesn’t address:
- Sanitizing NoSQL Databases (such as MongoDB)
- Input validation and input sanitization, which both help prevent security vulnerabilities
Sanitize the database
Make sure that you only sanitize preview environments and never the production environment. Otherwise you may lose most or even all of the relevant data stored in your database.
First, take a database dump of your preview environment.
This is just a safety precaution.
Production data isn’t altered.
To get a database dump, run the following command:
platform
db:dump -e DEVELOPMENT_ENVIRONMENT_NAME
.
Assumptions:
users
is the table where all of your PII is stored in thestaging
development database.staging
is an exact copy of your production database.
-
Connect to the
staging
database by runningplatform sql -e staging
. -
Display all fields from your
users
table, to select which ones need to be redacted. Run the following query:main=> SELECT * FROM users;
You see output like the following:
id | user_email | display_name -----+-----------------------------------------+----------------------- 3501 | daniel02@yourcompany.com | Jason Brown 3502 | ismith@kim.com | Sandra Griffin 3503 | olee@coleman-rodriguez.com | Miss Christine Morgan
-
Change the fields where PII is contained with the
UPDATE
statement. For example, to change the display name of users with an email address not in your company’s domain to a random value, run the following query:UPDATE users SET display_name==substring(md5(display_name||'$PLATFORM_PROJECT_ENTROPY') for 8); WHERE email NOT LIKE '%@yourcompany%'
Adapt and run that query for all fields that you need to sanitize. If you modify fields that you shouldn’t alter, you can restore them from the dump you took in step 1.
You can create a script to automate the sanitization process to be run automatically on each new deployment. Once you have a working script, add your script to sanitize the database to a
deploy
hook:.platform.app.yamlhooks: deploy: | # ... cd /app/public if [ "$PLATFORM_ENVIRONMENT_TYPE" = production ]; then # Do whatever you want on the production site. else # The sanitization of the database should happen here (since it's non-production) sanitize_the_database.sh fi
Assumptions:
users
is the table where all of your PII is stored in thestaging
development database.database
is the relationship name for the PostgreSQL service.
Set up a script by following these steps:
-
Retrieve service credentials from the
PLATFORM_RELATIONSHIPS
environment variable to use thepsql
command interface. Export these values to a.environment
file or include them directly in the sanitization script..environment# Pull credentials from the PLATFORM_RELATIONSHIPS environment variable. DB_USER=$(echo $PLATFORM_RELATIONSHIPS | base64 --decode | jq -r '.database[0].username') DB_HOST=$(echo $PLATFORM_RELATIONSHIPS | base64 --decode | jq -r '.database[0].host') DB_PORT=$(echo $PLATFORM_RELATIONSHIPS | base64 --decode | jq -r '.database[0].port') DB_PASS=$(echo $PLATFORM_RELATIONSHIPS | base64 --decode | jq -r '.database[0].password')
-
Create an executable sanitizing script by running the following command:
touch sanitize.sh && chmod +x sanitize.sh
-
Make the script sanitize environments with an environment type other than
production
.The following example runs only in preview environments and sanitizes the
display_name
andemail
columns of theusers
table. Adjust the details to fit your data.sanitize.sh#!/usr/bin/env bash if [ "$PLATFORM_ENVIRONMENT_TYPE" != production ]; then # Sanitize data PGPASSWORD=$DB_PASS psql -c "UPDATE users SET display_name=substring(md5(display_name||'$PLATFORM_PROJECT_ENTROPY') for 8);" -U $DB_USER -h $DB_HOST -p $DB_PORT PGPASSWORD=$DB_PASS psql -c "UPDATE users SET email=substring(md5(email||'$PLATFORM_PROJECT_ENTROPY') for 8);" -U $DB_USER -h $DB_HOST -p $DB_PORT fi
To sanitize only on the initial deploy and not all future deploys, on sanitization create a file on a mount. Then add a check for the file as in the following example:
sanitize.sh#!/usr/bin/env bash if [ "$PLATFORM_ENVIRONMENT_TYPE" != production ] && [ ! -f MOUNT_PATH/is_sanitized ]; then # Sanitize data touch MOUNT_PATH/is_sanitized fi
-
Update the deploy hook to run your script on each deploy.
.platform.app.yamlhooks: build: ... deploy: | python manage.py migrate bash sanitize.sh
-
Commit your changes by running the following command:
git add .environment sanitize.sh .platform.app.yaml&& git commit -m "Add sanitization."
Push the changes to
staging
and verify that environment’s database was sanitized. Once merged to production, all data from future preview environments are sanitized on environment creation.
What’s next
You learned how to remove sensitive data from a database.
To replace sensitive data that with other meaningful data, you can add a faker
to the process.
A faker
is a program that generates fake data that looks real.
Having meaningful PII-free data allows you to keep your current Q&A, external reviews, and other processes.
To add a faker, adapt your sanitizing queries to replace each value that contains PII with a new value generated by the faker.
You might also want to make sure that you implement input validation.
If your database contains a lot of data, consider using the REINDEX
statement to help improve performance.