Migrate from Aurora to PlanetScale
Use this guide to migrate an existing Aurora (Postgres) database to PlanetScale for Postgres.
This guide will cover a no-downtime approach to migrating using Postgres logical replication. If you are willing to tolerate downtime during a maintenance window, you may also use pg_dump and restore. The pg_dump/restore approach is simpler, but is only for applications where downtime is acceptable.
This guide assumes that public internet access is enabled on your Aurora database, as it will be needed to connect and replicate to the new PlanetScale host. If you cannot enable this due to security policies, consider using AWS DMS for your migration, or contact support for more specific guidance.
These instructions work for all versions of Postgres that support logical replication (version 10+) If you have an older version you want to bring to PlanetScale, contact us for guidance.
Before beginning a migration, you should check our extensions documentation to ensure that all of the extensions you rely on will work on PlanetScale.
As an alternative to this guide, you can also try our Postgres migration scripts. These allow you to automate some of the manual steps that we describe in this guide.
1. Prepare your PlanetScale database
Go to app.planetscale.com and create a new database. A few things to check when configuring your database:
- Ensure you select the correct cloud region. You typically want to use the same region that you deploy your other application infrastructure to.
- This guide assumes you are migrating from a Postgres Aurora database, so also choose the Postgres option in PlanetScale.
- Choose the best storage option for your needs. For applications needing high-performance and low-latency I/O, use PlanetScale Metal. For applications that need more flexible storage options or smaller compute instances, choose "Elastic Block Storage" or "Persistent Disk."
Once the database is created and ready, navigate to your dashboard and click the "Connect" button.
From here, follow the instructions to create a new default role. This role will act as your admin role, with the highest level of privileges.
Though you may use this one for your migration, we recommend you use a separate role with lesser privileges for your migration and general database connections.
To create a new role, navigate to the Role management page in your database settings. Click "New role" and give the role a memorable name. By default, pg_read_all_data and pg_write_all_data are enabled. In addition to these, enable pg_create_subscription and postgres, and then create the role.
Copy the password and all other connection credentials into environment variables for later use:
PLANETSCALE_USERNAME=pscale_api_XXXXXXXXXX.XXXXXXXXXX PLANETSCALE_PASSWORD=pscale_pw_XXXXXXXXXXXXXXXXXXXXXXX PLANETSCALE_HOST=XXXX.pg.psdb.cloud PLANETSCALE_DBNAME=postgres
We also recommend that you increase max_worker_processes for the duration of the migration, in order to speed up data copying. Go to the "Parameters" tab of the "Cluster configuration" page:
On this page, increase this value from the default of 4 to 10 or more:
You can decrease these values after the migration is complete.
2. Prepare the Aurora database
For PlanetScale to import your database, it needs to be publicly accessible. You can check this in your AWS dashboard.
In the writer instance of your database cluster, go to the “Connectivity & security” tab, and under “Security” you will see if your database is publicly accessible. If it says “No,” you will need to change it to be publicly accessible through the “Modify” button. If this is an issue, you cannot do this, or you have questions, please contact support to explore your migration options.
You will also need to change some parameters and ensure that logical replication is enabled. If you don't already have a parameter group for your Aurora cluster, create one from the "Parameter groups" page in the AWS console:
From here, click the button to create a new group. Choose whichever name and description you want. Set the Engine type to Aurora Postgres and the Parameter family group to the version that matches your Aurora Postgres database. Set the Type to DB Cluster Parameter Group.
If you already have a custom parameter group for your cluster, you can use the existing one instead. The two key parameters you need to update are adding pglogical to shared_preload_libraries and setting rds.logical_replication to 1:
Once these are set, you need to make sure your Aurora database is configured to use them. Navigate to your Aurora database in the AWS console, click the "Modify" button, and then ensure your database is using the parameter group:
When you go to save the changes, select the option to either apply immediately or during your next maintenance window. The changes may take time to propagate. You can confirm that the wal_level is set to logical by running SHOW wal_level; on your Aurora database:
postgres=> SHOW wal_level; wal_level ----------- logical
If you see a result other than logical, then it is not configured correctly. If you are having trouble getting the settings to propagate, you can try restarting the Aurora instance, though that will cause a period of downtime.
3. Copy schema from Aurora to PlanetScale
Before we begin migrating data, we first must copy the schema from Aurora to PlanetScale. We do this as a distinct set of steps using pg_dump.
Warning
You should not make any schema changes during the migration process. You may continue to select, insert, update, and delete data, keeping your application fully online during this process.
For these instructions, you'll need to connect to Aurora with a role that has permissions to create replication publications and read all data. Your default role that was generated by Aurora when you first created your database should suffice here, but you may also use other roles. We will assume that the credentials for this user and other connection info are stored in the following environment variables.
AURORA_USERNAME=XXXX AURORA_PASSWORD=XXXX AURORA_HOST=XXX AURORA_DBNAME=XXX
Run the below command to take a snapshot of the full schema of the $AURORA_DBNAME that you want to migrate:
PGPASSWORD=$AURORA_PASSWORD \
pg_dump -h $AURORA_HOST \
-p 5432 \
-U $AURORA_USERNAME \
-d $AURORA_DBNAME \
--schema-only \
--no-owner \
--no-privileges \
-f schema.sql
This saves the schema into a file named schema.sql.
Note
The above command will dump the tables for all schemas in the current database. If you want to migrate only one specific schema, you can add the --schema=SCHEMA_NAME option.
The schema then needs to be loaded into your new PlanetScale database:
PGPASSWORD=$PLANETSCALE_PASSWORD \
psql -h $PLANETSCALE_HOST \
-p 5432 \
-U $PLANETSCALE_USERNAME \
-d $PLANETSCALE_DBNAME \
-f schema.sql
In the output of this command, you might see some error messages of the form:
psql:schema.sql:LINE: ERROR: DESCRIPTION
You should inspect these to see if they are of any concern. You can reach out to our support if you need assistance at this step.
4. Set up logical replication
We now must create a PUBLICATION on Aurora that the PlanetScale database can subscribe to for data copying and replication.
To create a publication for all tables in all schemas of the current database, run the following command on your Aurora database:
CREATE PUBLICATION replicate_to_planetscale FOR ALL TABLES;
You should see this if it created correctly:
CREATE PUBLICATION
Note
To publish changes for only one specific schema, run the following query:
SELECT 'CREATE PUBLICATION replicate_to_planetscale FOR TABLE ' ||
string_agg(format('%I.%I', schemaname, tablename), ', ') || ';'
FROM pg_tables
WHERE schemaname = 'YOUR_SCHEMA_NAME';
This will generate a query that looks like this:
CREATE PUBLICATION replicate_to_planetscale FOR TABLE public.table_1, public.table_2, ... public.table_n;
You can then copy/paste this and execute on Aurora. This will create a publication that only publishes changes for the tables in YOUR_SCHEMA_NAME
After creating the publication on Aurora, we then need to tell PlanetScale to SUBSCRIBE to this publication.
PGPASSWORD=$PLANETSCALE_PASSWORD psql \ -h $PLANETSCALE_HOST \ -U $PLANETSCALE_USERNAME \ -p 5432 $PLANETSCALE_DBNAME \ -c " CREATE SUBSCRIPTION replicate_from_aurora CONNECTION 'host=$AURORA_HOST dbname=$AURORA_DBNAME user=$AURORA_USERNAME password=$AURORA_PASSWORD' PUBLICATION replicate_to_planetscale WITH (copy_data = true);"
Data copying and replication will begin at this point. To check in on the row counts for the tables, you can run a query like this on your source and target databases:
SELECT table_name, row_count FROM ( SELECT 'table_name_1' as table_name, COUNT(*) as row_count FROM table_name_1 UNION ALL SELECT 'table_name_2', COUNT(*) FROM table_name_2 UNION ALL ... SELECT 'table_name_N', COUNT(*) FROM table_name_N ) t ORDER BY table_name;
When the row counts match (or nearly match) you can begin testing and prepare for your application to cutover to use PlanetScale.
5. Handling sequences
Logical replication is great at migrating all of your data over to PlanetScale. However, logical replication does not synchronize the nextval values for sequences in your database. Sequences are often used for things like auto incrementing IDs, so it's important to ensure we update this before you switch your traffic to PlanetScale.
You can see all of the sequences and their corresponding nextvals on your source Aurora database using this command:
SELECT schemaname, sequencename, last_value + increment_by AS next_value FROM pg_sequences;
An example output from this command:
schemaname | sequencename | next_value ------------+------------------+------------ public | users_id_seq | 105 public | posts_id_seq | 1417 public | followers_id_seq | 3014
What this means is that we have three sequences in our database. In this case, they are all being used for auto-incrementing primary keys. The nextval for the users_id_seq is 105, the nextval for the posts_id_seq is 1417, and the nextval for the followers_id_seq is 3014. If you run the same query on your new PlanetScale database, you'll see something like:
schemaname | sequencename | next_value ------------+------------------+------------ public | users_id_seq | 0 public | posts_id_seq | 0 public | followers_id_seq | 0
If you switch traffic over to PlanetScale in this state, you'll likely encounter errors when inserting new rows:
ERROR: duplicate key value violates unique constraint "XXXX" DETAIL: Key (id)=(ZZZZ) already exists.
Before switching over, you need to progress all of these sequences forward so that the nextvals produced will be greater than any of the values previously produced on the source Aurora database, avoiding constraint violations. There are several approaches you can take for this. A simple way to solve the problem is to first run this query on your source Aurora database:
SELECT 'SELECT setval(''' || schemaname || '.' || sequencename || ''', '
|| (last_value + 10000) || ');' AS query
FROM pg_sequences;
This will generate a sequence of queries that will advance the nextval by 10,000 for each sequence:
query
--------------------------------------------------
SELECT setval('public.users_id_seq', 10104);
SELECT setval('public.posts_id_seq', 11416);
SELECT setval('public.followers_id_seq', 13013);
You would then execute these on your target PlanetScale database. You need to ensure you advance each sequence far enough forward so that the sequences in the Aurora database will not reach these nextvals before you switch your primary to PlanetScale. For tables that have a high insertion rate, you might need to increase this by a larger value (say, 100,000 or 1,000,000).
6. Cutting over to PlanetScale
Before you cutover, it's good to have confidence that the replication is fully caught up between Aurora and PlanetScale. You can do this using Log Sequence Numbers (LSNs). The goal is to see these match up between the source Aurora database and the target PlanetScale database exactly. If they don't, it indicates that the PlanetScale database is not fully caught-up with the changes happening on Aurora.
You can run this on Aurora to see the current LSN:
postgres=> SELECT pg_current_wal_lsn(); pg_current_wal_lsn -------------------- 0/703FE460
Then on PlanetScale, you would run the following query to check for a match:
postgres=> SELECT received_lsn, latest_end_lsn
FROM pg_stat_subscription
WHERE subname = 'replicate_from_aurora';
received_lsn | latest_end_lsn
--------------+----------------
0/703FE460 | 0/703FE460
Once you are comfortable that all your data has successfully copied over and replication is sufficiently caught up, it's time to switch to PlanetScale. In your application code, prepare the cutover by changing the database connection credentials to go to PlanetScale rather than Aurora. Then, you can deploy this new version of your application, which will begin using PlanetScale as your primary database.
After doing this, new rows written to PlanetScale will not be reverse-replicated to Aurora. Thus, it's important to ensure you are fully ready for the cutover at this point.
Once this is complete, PlanetScale is now your primary database! We recommend you keep your old database around for at least a few days, just in case you discover any data or schemas you forgot to copy over to PlanetScale.
Need help?
Get help from the PlanetScale Support team, or join our GitHub discussion board to see how others are using PlanetScale.