Database & Storage
Postgres requirements, backups, migrations, and operational posture for Cascades durable state—not ORM internals.
Cascades persists durable application state primarily in PostgreSQL: operator accounts, tenancy metadata, workflow definitions, executions, hashed proof payloads, entitlement tables, observability breadcrumbs. Operators should plan provisioned Postgres with TLS-enabled connectivity, sane connection pools, WAL durability, snapshots, failover targets, replicas as business continuity demands.
Avoid treating the database artifact as interchangeable with general object storage—you still orchestrate blobs/payload proofs via application logic and cloud storage gateways when large artifacts leave Postgres row limits.
Connection string
Expose a single Postgres DSN DATABASE_URL (TLS recommended):
DATABASE_URL="postgresql://USER:PASSWORD@HOST:PORT/DATABASE?sslmode=require"Treat credentials as Tier-1 secrets—rotate exposures aggressively.
Provisioning playbook
Size & SLA
Estimate working set for workflow cardinality, proof append volume, dashboards query patterns. Provision autoscaling replicas or read replicas for analytics if heavy BI extracts become hot.
Migrations / schema rollout
Advance schema revisions through your release pipeline (CI migration jobs). Freeze destructive operations behind maintenance banners. Dry-run migrations on staging clones before production rollout; keep rollback manifests handy.
Backups / PITR
Enable automated snapshots + WAL archiving; document RPO/RTO expectations in your internal runbooks. Quarterly restore rehearsals recommended for regulated footprints.
Storage expectations beyond OLTP rows
Operational attachments (exported proofs, zipped logs) ideally land in guarded object-storage buckets keyed by tenancy—keep Postgres authoritative for lineage references.
Observability hooks
Expose slow-query logs / pg_stat_statements ingestion to your monitoring backbone; escalate noisy workflow proof writes during incident investigations.
Call out performance ceilings when operators attempt large fan-out executions without queue backpressure safeguards.
Health checks ahead of rollout
- Verify migration history applied cleanly in your automation.
- Confirm read/write probes against critical tables (users/tenancy, workflow definitions, runs, proof records, metering fields where used).
- Load-test representative orchestration bursts before executive demos.
Operational ownership remains with your infra team—these guidelines describe requirements, not Cascades internals.