Key Concepts of Managed PostgreSQL

This section presents the fundamental concepts of our Managed PostgreSQL service. Understanding these principles will help you get the most out of your managed database by aligning its capabilities with your application requirements and security standards.

Sovereignty and SecNumCloud Compliance

At the core of our product lies digital sovereignty. The Managed PostgreSQL service is fully hosted on the Cloud Temple infrastructure, certified SecNumCloud 3.2 by ANSSI.

• 100% France-based hosting: Your data remains on national territory, shielded from extraterritorial laws.
• Native compliance: The solution is designed to meet the strictest regulatory requirements (GDPR, HDS, LPM, NIS2, PCI-DSS).
• Open Source and Reversibility: By leveraging open standards such as PostgreSQL Community Server and Patroni, we guarantee no technological dependency (vendor lock-in) and full portability of your data.

High Availability Architecture: Patroni Cluster

For distributed environments, our product relies on Patroni Cluster to provide high availability without data loss.

• Synchronous Replication: Unlike traditional asynchronous replication, each transaction is committed on all cluster nodes before being confirmed. This ensures a Zero Recovery Point Objective (RPO): no committed data can be lost in the event of a failure.
• Multi-AZ Distribution: The cluster is distributed across three distinct Availability Zones (AZ). The failure of an entire datacenter results in no service interruption or data loss.
• Automatic Failover: In the event of a node failure, traffic is automatically redirected to healthy nodes, ensuring a minimal Recovery Time Objective (RTO).

Deployment Models

We offer two models to accommodate the criticality of your workloads.

1. StandAlone

This model deploys a single instance of the PostgreSQL engine.

• Use case : This deployment model is perfectly suited for simple applications, such as CMSs, which use only a single endpoint to connect to databases.
• Resilience : Although it is a single instance, the underlying storage is replicated across 3 AZs, allowing automatic restart on another AZ in case of hardware failure.
• SLA : 99.9% (excluding maintenance windows).

2. Distributed

This model deploys a Patroni cluster with 3 instances of the PostgreSQL engine, complemented by a PgBouncer proxy.

• Use case: This deployment model is ideally suited for applications with distributed access, such as data or business intelligence applications, which benefit from read-only access without impacting data ingestion.
• Components :
  • 3 PostgreSQL Nodes : One primary read-write (RW) node and two secondary read-only (RO) nodes.
  • PgBouncer Proxy : An intelligent router that distributes queries. It routes writes to the primary node and distributes reads across the secondary nodes (read/write splitting), thereby optimizing performance.
• SLA : 99.9% (excluding maintenance windows).

Important Note : It is not possible to change the deployment model of an existing cluster (for example, from StandAlone to Distributed). This operation requires creating a new cluster in the desired model, via a PiTR restore.

Backup and Restoration (PITR)

Your data protection is ensured by a dual backup strategy.

1. Physical Backup (Point-in-Time Recovery - PITR):

   • We perform complete daily physical backups using pg_basebackup (without service interruption).
   • Transaction logs (WAL) are continuously archived.
   • This combination enables restoration to the exact second, up to the moment just before an incident.

2. Logical Backup (pg_dump):

   • Logical exports of the databases are also performed.
   • They offer fine-grained control to restore or export an individual database.

All backups are encrypted at rest and stored on our S3 Object Storage, which is itself SecNumCloud certified.

Multi-Level Security

Security is integrated into every layer of the service.

• Network Isolation: Database instances are never exposed to the Internet. Access is exclusively via the client's private network.
• End-to-End Encryption:
  • In transit: All connections (client to database and between cluster nodes) are encrypted using TLS 1.3.
  • At rest: Data on disk and backups are encrypted using AES-256.
• Access Management: Authentication is secured and permissions are managed according to the principle of least privilege.

Managed Service ("Zero Ops")

The objective of Managed PostgreSQL is to offload operational complexity. Our teams handle:

• Provisioning and initial configuration.
• Full lifecycle management: minor updates, security patch application.
• 24/7 monitoring of the infrastructure and service.
• Backup management and verification.

This allows your teams to focus on application development and data operations.

Version Policy & Lifecycle

The PostgreSQL community releases versions with long-term support (LTS), ensuring stability and predictability. Our service relies on these versions to ensure the longevity of your infrastructure.

Version	Type	Supported Until
PostgreSQL 15	LTS	Nov 2027
PostgreSQL 16	LTS	Nov 2028

• Minor Updates : Security patches and bug fixes are applied by our teams via rolling update (node by node) to cause no service interruption.
• Major Updates : Major version upgrades are planned in collaboration with you to align with your schedule.
• End of Support : We notify you at least 180 days before the end of support for an LTS version to plan the migration to the next version.

Instance Sizes

StandAlone and Distributed instances are available with predefined sizes:

Size	vCPU	Memory	innodb_buffer_pool_size	innodb_buffer_pool_instances	max_allowed_packet	table_open_cache	maxconn
XS	1000m	4096Mi	2458M	2	256M	800	80
S	1000m	8192Mi	4915M	4	512M	1600	150
M	2000m	8192Mi	4915M	4	512M	1600	150
L	2000m	16384Mi	9830M	8	1G	3200	250
XL	4000m	16384Mi	9830M	8	1G	3200	250
XXL	4000m	32768Mi	19660M	16	1G	6400	500
3XL	8000m	32768Mi	19660M	16	1G	6400	500
4XL	8000m	65536Mi	39320M	16	1G	10000	500

Note: Storage is provisioned separately and can be scaled up on the fly (from 2Gi to 128Gi) (but not scaled down, except by recreating a new instance.).