Managed Databases
PostgreSQL
Use your cloud provider's managed PostgreSQL service for automated backups, failover, and patching:
| Provider | Service | Minimum Version |
|---|---|---|
| AWS | RDS for PostgreSQL | 15+ |
| GCP | Cloud SQL for PostgreSQL | 15+ |
| Azure | Azure Database for PostgreSQL Flexible Server | 15+ |
Configuration recommendations:
| Parameter | Recommendation | Rationale |
|---|---|---|
| Instance size | 2 vCPU / 8 GB RAM minimum | Workflow state queries are CPU-bound |
| Storage | 100 GB SSD (gp3 or equivalent) | Journal and execution logs grow over time |
max_connections | 200+ | Support PgBouncer pool + direct admin connections |
shared_buffers | 25% of RAM | Standard PostgreSQL tuning |
| High availability | Multi-AZ / regional | Automated failover for production |
| Backups | Daily automated, 7–30 day retention | Point-in-time recovery |
Connection string format:
1DATABASE_URL="postgresql://orcher:PASSWORD@your-instance.region.rds.amazonaws.com:5432/orcher?sslmode=require"Always enable SSL (sslmode=require or sslmode=verify-full) for managed database connections.
PgBouncer
For high-concurrency workloads (100+ concurrent workflows), deploy PgBouncer between the orchestrator and managed PostgreSQL. Some managed services include built-in connection pooling:
- AWS RDS Proxy — Managed connection pooler for RDS
- GCP Cloud SQL — Built-in connection pooling via Cloud SQL Auth Proxy
- Azure — Built-in PgBouncer in Flexible Server (enable via server parameters)
If your provider doesn't include pooling, deploy PgBouncer as a sidecar container or separate service. See Docker deployment for PgBouncer configuration.
Redis
Optional but recommended for production. Wakes workers via LPUSH/BRPOP as soon as work is queued.
| Provider | Service |
|---|---|
| AWS | ElastiCache for Redis |
| GCP | Memorystore for Redis |
| Azure | Azure Cache for Redis |
Configuration recommendations:
| Parameter | Recommendation | Rationale |
|---|---|---|
| Instance size | 1–2 GB | Dispatch queues are small; most data is in PostgreSQL |
| Version | 7.0+ | Required for full feature support |
| Persistence | AOF enabled | Survives restarts without losing pending dispatches |
| Cluster mode | Disabled | Orcher uses simple key operations, not sharded |
| High availability | Replica with automatic failover | Avoid dispatch interruptions |
Connection string format:
1REDIS_URL="redis://your-instance.region.cache.amazonaws.com:6379"If Redis is unavailable, the orchestrator automatically falls back to PostgreSQL-based polling. The orcher_postgres_fallback_total metric tracks fallback events.
Next Steps
- Cloud Providers — AWS, GCP, and Azure deployment guides
- Operations — High availability, monitoring, and secrets management
- Cloud Platforms Overview — Architecture and container image setup