Partial data dumps using Postgres Row Level Security

When working with databases, it's common to create a seed.sql file which contains a subset of production data for testing.

During early development, it's fine to dump the entire database and restore it on your development machine. However, once you have production users this becomes a security issue - do you really want to dump your users' data onto your local machines?

There are many ways to solve this, but recently I stumbled upon a neat way to do it using PostgreSQL's Row Level Security (RLS).

The concept is simple:

1. Create a database user with restricted access.
2. Define some RLS rules for that user, limiting what data they can access.
3. Run pg_dump as that user.

For this scenario, let's imagine that you have a table called profiles in your database:

create table profiles (
  id serial primary key,
  name text,
  email text
);

In this case, if we ran a pg_dump we will save Jenny and Joe's personal data. We don't want that, so let's create a Postgres user called exporter, who can only dump the data we want.

Step 1: Prepare a user

Create a user to connect to the database. We'll call them exporter and grant them access to the public schema:

-- Create a new user with login privileges
create user exporter
  with password 'exporter_secure_password';

-- Allow this user to select the rows we need
grant usage on schema public to exporter;
grant select on profiles to exporter;

Step 2: Create data access rules

Let's turn on RLS for this table and limit the data which exporter can access:

-- Turn on Row Level Security
alter table profiles
  enable row level security;

-- Only dump data for internal team members 1, 2, 3
create policy "Data dump rule" on profiles
  for select
  to exporter
  using (
    id in (1, 2, 3)
  );

Step 3: Export the data

Now we can use pg_dump to get only the data that we need.

Run the dump with the exporter user that we created above and use the --enable-row-security flag to ensure that the dump succeeds.

# Dump all the data into a "seed.sql" file
# which we can use to restore our local databases.
pg_dump \
-h db.host.supabase.co \
-U exporter \
-d postgres \
-n public \
--data-only \
--enable-row-security \
--table=profiles \
> seed.sql

-h db.host.supabase.co \

And that's it. You can follow this same pattern for any tables that you want to dump.

Data access patterns

RLS is a bit like appending a “where” clause to a select, so you can create all sorts of data access patterns. Let's see a few more which are useful for extracting seed data.

Using email rules

Instead of using hardcoded numbers in our RLS policies, we could use email extensions to determine the users who we want to export:

-- Only dump data for supabase employees
create policy "Data dump rule" on profiles
  for select
  to exporter
  using (
    substring(email from '@(.*)$') = 'supabase.com'
  );

Only recent data

If we have a table with a lot of data, like an analytics table, we might only care about the last 2 months of data.

-- A fake analytics table where we store actions a user takes
create table analytics (
  id serial primary key,
  ts timestamptz default now(),
  profile_id references profiles,
  event text
);
alter table profiles
  enable row level security;

-- Here is an "age" rule so that we only dump the most recent analytics
create policy "Data dump rule" on logs
  for select
  to exporter
  using (
    profile_id in (1, 2, 3) and
    ts > now() - interval '2 MONTHS' -- here's the magic
  );

Using flags

If you don't mind having some additional columns in you database, you can add flags to each row to determine whether it's safe to export.

create table profiles (
  id serial primary key,
  name text,
  email text,
  is_exportable boolean -- make this "TRUE" if you want to allow access
);
alter table profiles
  enable row level security;

-- Only dump data for internal team members 1, 2, 3
create policy "Data dump rule" on profiles
  for select
  to exporter
  using ( is_exportable = true );

Conclusion

Using seed data isn't the only way to run development environments. It's also possible to run fully-masked copies of your database using tools like Snaplet.

We're also bullish on copy-on-write strategies which allow users to "fork" a database at a point in time, a strategy used by Database Lab Engine. DLE uses the ZFS file system to achieve this, but it's within reach of the Postgres core once alternative storage strategies become easier to implement.

If you want to try out the steps we described in this article, fire up a full PostgreSQL database: database.new

More Postgres resources

• Implementing "seen by" functionality with Postgres
• Postgres Views
• Postgres Auditing in 150 lines of SQL
• Cracking PostgreSQL Interview Questions
• What are PostgreSQL Templates?
• Realtime Postgres RLS on Supabase