Skip to main content
Version: v0.7

Migrating to Keto v0.7

Ory Keto v0.6 used the table-separated namespaces database schema described in the Google Zanzibar paper. However, we found for various reasons outlined in #613 that this is not ideal for Keto.

Because the database schema changed significantly, and it is not possible to have SQL-only migrations, there is a special migration procedure needed to upgrade Ory Keto v0.6.

Preparations

danger

As always with database migrations, please make a backup of your data and try the migration process first on a non-production copy of the database.

This migration has to download all the data from the database, transform them, and then write them back to the database. Make sure that you run it as close to the database as possible to reduce the latency and failure rate. In case of a failure, just restart the process. It is run as one big SQL transaction.

caution

This is the recommended procedure. Please adjust to your setup and test before applying to a production system. You might want to create a migration script that applies all these steps and handles errors.

This procedure is zero-downtime only for read API requests. The write API will not be available during migration.

  1. Create a new database user (from now on keto_new) with the same privileges as the one used until now (let's say it was keto).
  2. Start a server instance of Keto v0.7 next to the already running Keto v0.6. It should have the same settings, but use keto_new for accessing the database.
  3. Change the privileges of keto to be read-only on all tables.
  4. Still route all traffic to the old Keto. Write API requests will fail from now on, but the Read API will be zero-downtime.
  5. Run keto migrate up --yes considering the points raised in preparations. This command will apply SQL schema changes, but not yet migrate data.
  6. Run keto namespace migrate legacy --yes considering the points raised in preparations. This command will migrate all namespaces and delete the old tables on success.
  7. In case of failure, re-run the command until it succeeds.
  8. In case of log statements stating Skipping relation tuple, it seems to be in a broken state. Please recreate it manually. note down the logged data somewhere and recreate the relation tuples in question once the migration is done using the API.
  9. After the migration is successfully done, route all traffic to Keto v0.7 and shut down Keto v0.6. At this point the API is fully operational again.

Options for Adjusting the Migration Process

The keto namespace migrate legacy command supports multiple options:

  • interactive mode with manual approval by not specifying --yes
  • migrate single namespace by specifying its name as an argument
  • only delete old tables by specifying --down-only; useful if you did not approve the down migration previously