[Architecture] Multi-Environment pg_cron for Automated Partition Management - Mailing list pgsql-admin

Hello Postgres Community,

I am writing to get your expert opinion on a proposed architecture for managing pg_partman automated partitioning across a multi-environment PostgreSQL setup. While this solution is primarily for cost savings and is used in our non-production environments, stability is a must. Our core goal is to create a universal scheduling solution that works identically across cloud-managed databases (AWS RDS, Google Cloud SQL, Azure Database), as well as on-premises shared and dedicated hosts.

Business Context & Core Requirements

We need to consistently automate pg_partman maintenance tasks across our PostgreSQL 17 environments. The primary challenge is maintaining a single, identical application interface so that the same SQL commands work without modification, regardless of the deployment model. A key requirement is that our in-house database schema migration tools will be used to roll out SQL scripts that directly handle job maintenance and alterations during releases. Our jobs are low frequency (monthly/quarterly/annually) but are critically important, as failures can cause serious operational issues. 

Our current constraints include:

• PostgreSQL 17 (with a planned migration to 18).

• A mix of AWS RDS, Google Cloud SQL, Azure Database, and on-premises deployments.

• Shared hosts with 50-60 databases and dedicated hosts with a single database.

Our Proposed Architecture

As the pg_cron extension requires a dedicated database to schedule jobs for other databases, we have designed an abstraction layer to address our multi-environment challenges. Our architecture places pg_cron in a designated database and uses dblink to execute partition maintenance jobs in the target application databases.

Key elements of this design are:

• Universal API Layer: Every application database would have an identical cron schema with wrapper functions.

• Identical Application Interface: Applications use the exact same SQL statement, for example: SELECT cron.schedule('monthly_partition_maintenance', '0 2 1 * *', 'SELECT partman.run_maintenance_proc()');

• Environment-Adaptive Communication: dblink is used for multi-database environments, while direct calls are used for single-database setups.

• PostgreSQL 18 Future-Proofing: We plan to leverage SCRAM pass-through authentication for dblink to eliminate the need for storing credentials.

Technical Questions & Concerns

We have detailed several technical questions below and would greatly appreciate your insights and validation.

1. Architecture Validation

• Is this universal abstraction layer a sound architectural pattern for managing pg_partman across diverse environments, especially given our focus on using in-house tools for job management?

• How should the architecture specifically adapt for AWS RDS vs. Google Cloud SQL vs. Azure Database, especially concerning their limitations on pg_cron or connection management?

• Does this design scale appropriately from a single dedicated database to a shared host with 50-60 databases and 400-900 partition jobs?

2. Connection & Performance

• Is our proposed temporary-connection-with-guaranteed-cleanup pattern robust enough for production? We are concerned about dblink connection exhaustion risks and connection leaks, especially in shared environments with many jobs.

• Given the low frequency of jobs, will the dblink overhead be significant? Are there specific performance optimizations we should consider for cloud-managed services vs. on-premises?

3. Alternative Approaches

• Should we abandon this PostgreSQL-native approach and instead consider cloud-native job schedulers (e.g, AWS EventBridge, Google Cloud Scheduler, Azure Logic Apps) to trigger maintenance jobs?

• Are there existing enterprise scheduling solutions that are purpose-built for this kind of multi-cloud, on-premises PostgreSQL automation?

We are most concerned about connection leaks, unnoticed maintenance failures, and ensuring a single, identical SQL interface is truly achievable. We believe this solution could significantly simplify our operations, but we want to validate its viability and get feedback on potential pitfalls from the community.

Thank you for your time and any insights you can provide.Best Regards,
Jishnu Sygal