SQL Database

Like K-V Storage, the SQL Database service is not part of the Fundamental Services but is included in the Data Access Layer. It can be installed as a simplified, single-instance setup using the same Harbor Helm Chart, making it suitable for All-in-One deployments, or deployed in HA mode using a separate PostgreSQL Helm Chart. Alternatively, a separate SQL Database instance can be integrated into MSR 4 as an independent storage service. In this case, it is considered a dependency rather than part of the deployment footprint and is managed by a dedicated corporate team. While a remote service is an option, it is not part of the reference architecture and is more suited for custom deployments based on specific needs.

Single Node Deployment

This is a streamlined, single-instance PostgreSQL deployment that runs as a StatefulSet and utilizes a PVC for storage.

HA Deployment

Unlike the previous single-node deployment, this setup is more robust and comprehensive. It involves deploying PostgreSQL in replication mode across multiple worker nodes. The configuration includes two types of pods: replicas, managed as a StatefulSet, and pgpool, running as a ReplicaSet. Each pod uses a PVC for storage and a ConfigMap to store scripts and configuration files, while sensitive data, such as passwords, is securely stored in a Secret.

Pgpool operates as an efficient middleware positioned between PostgreSQL servers and PostgreSQL database clients. It maintains and reuses connections to PostgreSQL servers. When a new connection request with identical properties (such as username, database, and protocol version) is made, Pgpool reuses the existing connection. This minimizes connection overhead and significantly improves the system’s overall throughput.

PostgreSQL is a quorum-based service, so the number of replicas should always be odd—specifically 1, 3, 5, and so on.