Skip to content

Cloud provider notes ​

Verified against official documentation on 2026-11-01, with RDS and EC2 re-read on 2026-11-04. Track H2 of the deployment guide has run on Neon's direct connection, so that choice is tested. RDS and Supabase are documentation research only: their feasibility is an inference from provider contracts and the local image, not a tested deployment claim.

MergeBrief constraints ​

The details are in compose.md, Dockerfile and docker-entrypoint.sh. Compose is the supported deployment. The image starts api, web, worker, or migrate; its default is api. Build context must include the repository root. Web defaults to port 3000; Compose configures API on 0.0.0.0:3001. The worker is continuous and has no HTTP listener. API and worker migrate at boot unless MERGEBRIEF_AUTO_MIGRATE=false.

Use PostgreSQL 16+, sslmode=verify-full, and an explicit sslrootcert path where the provider CA requires it. Dashboard examples using sslmode=require must be changed: MergeBrief refuses that mode. Mount CA files into both API and worker. Size database capacity using the pool formula in operations.md, including both worker queues, replicas, and rollout overlap.

RDS with private EC2 connectivity ​

  • Place EC2 and RDS in the same VPC; keep RDS non-public. Give EC2 an application security group and RDS a separate database security group. Permit RDS inbound TCP 5432 from the EC2 security-group ID, not from the internet. If EC2 egress is restricted, permit TCP 5432 to the RDS security group. Security groups do not replace working subnet routing and network ACLs. This follows AWS's application-server security-group pattern.
  • The console's Connect to an EC2 compute resource builds that pattern: rds-ec2-n and ec2-rds-n groups, Public access: No, and a DB subnet group of private subnets, created if the VPC has none. It is console-only. Source: automatic network connectivity.
  • A master user holds CREATEROLE and CREATEDB but is not a superuser, so CREATE DATABASE ... OWNER needs it to be able to SET ROLE to the owner, and since Postgres 16 creating a role no longer grants that. Sources: CREATE DATABASE, createrole_self_grant.
  • Connect using the RDS endpoint hostname, preserving hostname verification. Download the regional PEM bundle from AWS's CA bundle table, mount it read-only, and use sslmode=verify-full&sslrootcert=/etc/ssl/certs/provider-ca.pem. AWS advises trusting roots, not intermediate certificates, to avoid rotation failures. RDS PostgreSQL 15+ defaults rds.force_ssl to on; retain it. Sources: RDS PostgreSQL TLS, certificate lifecycle.

Neon and Supabase connection selection ​

  • Neon: select the direct connection in the Connect dialog, with pooling disabled; its hostname lacks -pooler. Neon pooling is PgBouncer transaction mode and explicitly lists LISTEN/NOTIFY and session advisory locks as unsupported. Use direct connections for this deployment, including migrations; do not interpret pooled SQL connectivity as worker compatibility. Source: Neon pooling and limitations.
  • Supabase: prefer direct db.<project-ref>.supabase.co:5432, normally IPv6. For an IPv4-only host, use the dashboard's Session pooler connection on port 5432, retaining its supplied hostname and username, or purchase the IPv4 add-on for direct access. Shared pooler transaction mode on 6543 loses session state; the dedicated PgBouncer endpoint is also a transaction-pooling choice, not the session alternative. Sources: connection methods, network compatibility.
  • Graphile compatibility: Graphile uses LISTEN/NOTIFY by default and warns about non-session pooling. Thus direct connections are the safest fit; Supabase session mode is a plausible compatible alternative requiring smoke testing. Do not recommend either provider's transaction pooler for the shipped worker merely because polling sometimes retrieves jobs. Source: Graphile Worker FAQ.