Persistence
The performance of the web tier is derived from the performance of the persistence tier. While memory-caching short lived entities to reduce persistence is an option, there are still long lived objects that need to be written to the disk.
Picking the right database has a huge impact on the operational burden of a digital identity infrastructure. All database administration requires some black art--every platform has some secret knobs and levers that are critical to tuning, availability and rapid diagnosis. So beyond the performance requirements that determine the right choice for a database, you also need to consider operational concerns.
Janssen's strategy is to provide optionality for persistence. There is no one size fits all solution for databases. The following section will detail some of the pros and cons of the various databases we currently support.
-
OpenDJ (LDAP) Janssen supports Gluu's distribution of OpenDJ, and probably any other similar distributions like ForgeRock OpenDJ, Ping Directory Server, or Oracle Unified Directory. LDAP in general and OpenDJ in particular have been
successfully backing authentication service for more than 20 years. People tend to think of the LDAP tree structure as fast for reads, and slow for writes. That's just not true anymore--OpenDJ is able to perform quite well for write operations as well. OpenDJ has mature replication support, excellent command line tools for administration, and excellent stability. The main disadvantage of OpenDJ is scaling large datasets for high concurrency. While you can get around this shortcoming with a global LDAP proxy, such a topology gets complicated and costly to operate, as you have to break up the data and configure multiple replicated topologies. As a rule of thumb, if concurrency of more than 120 OpenID code flow authentications per second are needed, you should consider another database. But for concurrency less than this, OpenDJ is an excellent choice. Owing to these limitations, LDAP is not supported in production deployments using Kubernetes, 1.0.23 and forward. -
MySQL You know it... you love it. That's the biggest advantage. Performance is great out of the box. But if you have high concurrency, you'll have to figure out a plan for replication, and horizontal scaling. MySQL is our default persistence for Production deployments using Kubernetes.
-
Postgres Same as MySQL above, but there are some great commercial distributions of Postgres like EnterpriseDB. Postgres is our default persistence for VM based non-production deployments.
-
Couchbase A JSON NoSQL database that supports automatic distribution of data for auto-scaling multi-region sharded cloud-native deployments. Janssen only supports the commercial distribution--but the previously mentioned database properties are enterprise class. If you need to host your own database, and you need infinite horizontal scalability, Couchbase should be your go-to choice.
-
Aurora So you want MySQL, but you want Amazon to handle some of the care and feeding? Aurora enables you to consume database as a cloud service. Scalability is excellent and multi-region deployments are possible. The main catch is that write operations are limited to one region, with the ability to failover to another region. But to accomplish this, you need a cloud engineer to implement it.
-
Spanner Google's multi-region cloud database as a service, Spanner was purpose-built for auto-scaling, and multi-region persistence. It has its own API, although recently Google added support for MySQL and Postgres drivers.
Created: 2022-07-21
