Use cloud services for GitLab components
- Tier: Free, Premium, Ultimate
- Offering: GitLab Self-Managed
Instead of managing PostgreSQL, Redis/Valkey, and object storage yourself, you can use managed cloud provider services for them.
For cloud-native deployments using the GitLab Helm chart, external PostgreSQL and Redis/Valkey services are required. These components are not bundled in the chart.
Use managed cloud PostgreSQL
Use an external PostgreSQL service running a supported version. For setup instructions, see using an external PostgreSQL database.
Only full PostgreSQL deployments are supported. Services that implement the PostgreSQL wire protocol but are not full PostgreSQL deployments, such as Amazon Aurora and Google AlloyDB, are incompatible with GitLab.
Services known to work include:
Performance and high availability
For larger environments, enable
database load balancing with read replicas.
Match replica counts to those used in equivalent Linux package deployments.
When using read replicas, ensure all replica nodes have hot_standby_feedback = on to prevent
replication lag buildup.
For GCP Cloud SQL in larger environments, use the Enterprise Plus edition for optimal performance.
GCP Cloud SQL does not support statement_timeout as a database flag. Set it per database or user
instead: ALTER DATABASE gitlab SET statement_timeout = '60s';
GitLab Geo
GitLab Geo requires cross-region PostgreSQL replication between the primary and secondary sites. Not all managed database services support this.
Known limitations:
- Amazon RDS Multi-AZ DB cluster: cross-region replication is not supported. Use a standard RDS Multi-AZ DB instance with a cross-region read replica instead.
- GCP Cloud SQL: read replicas for Geo can only be created within the same VPC and the same GCP project. For secondary sites in a different project, Cloud SQL cannot create the replica directly. Configure PostgreSQL logical replication manually and then follow the GitLab Geo external PostgreSQL setup.
Connection pooling
If your workload requires additional connection pooling beyond what PostgreSQL handles directly, deploy your own PgBouncer instance.
The GitLab-bundled PgBouncer only works with the bundled PostgreSQL and cannot be used with external database services.
The following provider-managed pooling solutions are not recommended:
- AWS RDS Proxy: not validated for use with GitLab.
- Azure Database for PostgreSQL PgBouncer: single-threaded with limited observability. Can cause bottlenecks under load.
Use managed cloud Redis and Valkey
Use an external Redis or Valkey service running a supported version. For setup instructions, see Redis as a managed service.
The service must support:
- Standalone (primary with replica) mode, not Redis Cluster mode
- High availability through replication
- A configurable eviction policy
Serverless Redis and Valkey variants are not supported.
Services known to work include:
On AWS, use ElastiCache for Valkey 7.2. ElastiCache for Redis 7.2 is not available on AWS. ElastiCache for Redis 7.1 is built on Redis 7.0 OSS and is not recommended for new deployments. Neither Azure Cache for Redis nor Azure Managed Redis currently offer Redis 7.2 or a supported Valkey version. For Azure deployments, self-manage Redis or Valkey on a virtual machine.
For larger environments, run separate Redis instances for cache and persistent data. Redis is single-threaded and a single shared instance becomes a bottleneck at scale.
Use managed cloud object storage
Use any S3-compatible object storage service. For the full list of tested providers and configuration details, see object storage.