Managed Postgres Best Practices

Managed PostgreSQL removes patching toil but not performance engineering. Queries, indexes, connection pools, and restore drills remain your job on RDS, Cloud SQL, Aurora, Neon, and Supabase.

How to Use This List

Run before production cutover and after any provider incident
Pair with Cloud Selection ADR for platform choice context
Assume managed fixes neither missing indexes nor connection storms

A - Provisioning

Encryption at rest enabled at create. KMS key documented; no unencrypted snapshots.
Private networking only. No public IP unless exception approved.
Multi-AZ / regional HA for production. Single-AZ dev/staging only.
Deletion protection on production. Terraform prevent_destroy aligned.
Major version pinned explicitly. PostgreSQL 18.x, not "default" floating.

B - Configuration

Custom parameter group: logging, pg_stat_statements, timeouts. Not default group.
PgBouncer or provider pooler for app traffic. Direct connection for migrations only.
max_connections vs pool math documented. See connection math article.
SSL verify-full with provider CA bundle in apps. Rotated before CA expiry.
Extensions allowlist in repo; CI tests CREATE EXTENSION on staging.

C - Application Ownership

Weekly pg_stat_statements review. Top 10 by total_exec_time.
Indexes owned by migrations repo. Not console-only changes.
Idempotent app retry on failover. TCP disconnects expected.
Read replica routing with lag guard. Not all reads blindly to replica.
No superuser/master user in application DSN. Per-service roles.

D - Reliability and DR

Backup retention meets RPO policy. PITR window tested.
Quarterly restore drill to non-prod. Document minutes to readable DB.
Failover game day annually. Measure RTO; fix pooler/DNS gaps.
Replication slots monitored. CDC consumers cannot fill disk.
Maintenance window announced; staging upgraded first.

E - Cost and Governance

Right-size review quarterly. CPU credit, storage growth, idle replicas.
Tags: env, service, owner, cost-center. Finance chargeback ready.
IaC source of truth. Console drift corrected or forbidden.
Egress minimized: app and DB same region/AZ where possible.
ADR on file when platform changes. Link in team wiki.

FAQs

Managed means no DBA?

You still need Postgres expertise for queries, locks, extensions, and incidents.

Console hotfix OK?

Emergency only; backport change to Terraform within 24h or revert.

When add read replica?

When primary CPU > 60% sustained and read-heavy workload isolated.

Neon branch for prod?

No - branches for CI/preview; prod on stable endpoint with suspend policy reviewed.

Supabase service role?

Server-only; never in browser bundle.

Performance Insights enough?

Complements pg_stat_statements; not replacement for EXPLAIN discipline.

Cross-cloud DR?

Expensive; document RPO from async replication lag realistically.

Upgrade major version?

Blue/green or logical replication migration; test extension compatibility first.

Who owns parameter changes?

Platform applies; service team requests with load test evidence.

Minimum monitoring?

Provider metrics + pg_stat_statements + app p95 + lag + disk alerts.

Managed Postgres Basics - responsibility
AWS RDS & Aurora Postgres - AWS
Monitoring Best Practices - observability
Connection Math - pool sizing
Backup Verification - restore proof

Stack versions: This page was written for PostgreSQL 18.4 (stable 18, maintenance 17), pgvector 0.8+, PgBouncer 1.x, Patroni 3.x, and PostGIS 3.5+.

Managed Postgres Best Practices

Managed PostgreSQL removes patching toil but not performance engineering. Queries, indexes, connection pools, and restore drills remain your job on RDS, Cloud SQL, Aurora, Neon, and Supabase.

How to Use This List

Run before production cutover and after any provider incident
Pair with Cloud Selection ADR for platform choice context
Assume managed fixes neither missing indexes nor connection storms

A - Provisioning

Encryption at rest enabled at create. KMS key documented; no unencrypted snapshots.
Private networking only. No public IP unless exception approved.
Multi-AZ / regional HA for production. Single-AZ dev/staging only.
Deletion protection on production. Terraform prevent_destroy aligned.
Major version pinned explicitly. PostgreSQL 18.x, not "default" floating.

B - Configuration

Custom parameter group: logging, pg_stat_statements, timeouts. Not default group.
PgBouncer or provider pooler for app traffic. Direct connection for migrations only.
max_connections vs pool math documented. See connection math article.
SSL verify-full with provider CA bundle in apps. Rotated before CA expiry.
Extensions allowlist in repo; CI tests CREATE EXTENSION on staging.

C - Application Ownership

Weekly pg_stat_statements review. Top 10 by total_exec_time.
Indexes owned by migrations repo. Not console-only changes.
Idempotent app retry on failover. TCP disconnects expected.
Read replica routing with lag guard. Not all reads blindly to replica.
No superuser/master user in application DSN. Per-service roles.

D - Reliability and DR

Backup retention meets RPO policy. PITR window tested.
Quarterly restore drill to non-prod. Document minutes to readable DB.
Failover game day annually. Measure RTO; fix pooler/DNS gaps.
Replication slots monitored. CDC consumers cannot fill disk.
Maintenance window announced; staging upgraded first.

E - Cost and Governance

Right-size review quarterly. CPU credit, storage growth, idle replicas.
Tags: env, service, owner, cost-center. Finance chargeback ready.
IaC source of truth. Console drift corrected or forbidden.
Egress minimized: app and DB same region/AZ where possible.
ADR on file when platform changes. Link in team wiki.

FAQs

Managed means no DBA?

You still need Postgres expertise for queries, locks, extensions, and incidents.

Console hotfix OK?

Emergency only; backport change to Terraform within 24h or revert.

When add read replica?

When primary CPU > 60% sustained and read-heavy workload isolated.

Neon branch for prod?

No - branches for CI/preview; prod on stable endpoint with suspend policy reviewed.

Supabase service role?

Server-only; never in browser bundle.

Performance Insights enough?

Complements pg_stat_statements; not replacement for EXPLAIN discipline.

Cross-cloud DR?

Expensive; document RPO from async replication lag realistically.

Upgrade major version?

Blue/green or logical replication migration; test extension compatibility first.

Who owns parameter changes?

Platform applies; service team requests with load test evidence.

Minimum monitoring?

Provider metrics + pg_stat_statements + app p95 + lag + disk alerts.

Managed Postgres Basics - responsibility
AWS RDS & Aurora Postgres - AWS
Monitoring Best Practices - observability
Connection Math - pool sizing
Backup Verification - restore proof

Stack versions: This page was written for PostgreSQL 18.4 (stable 18, maintenance 17), pgvector 0.8+, PgBouncer 1.x, Patroni 3.x, and PostGIS 3.5+.

Managed Postgres Best Practices

Search Documentation

Managed Postgres Best Practices