What Managed PostgreSQL does not support
What Managed PostgreSQL does not do, one line each.
Last updated 18 November 2026View as Markdown
- Snapshots as a thing separate from backups. A snapshot is a backup: capturing a database now is Back up now (a pgBackRest full backup), and going back to one is a restore to the end of that backup. There is no volume snapshot or machine image, so the console has no Snapshots section and there are no snapshot create or delete permissions.
- Engines other than PostgreSQL, and PostgreSQL majors other than 15, 16, 17 and 18.
- Restoring into an existing database. A restore always builds a new database.
- Differential or incremental backups. Every backup is a full.
- Encrypting the data disk of a database created before disk encryption, in place. Restore it into a new database.
- Changing the KMS key of an existing database's disks. Restore into a new database with the key you want.
- Re-encrypting the data itself (the volume key). A key rotation replaces the passphrase that unlocks it.
- Requiring TLS on a database provisioned before managed PostgreSQL had certificates. Restore it into a new database.
- A login scoped to one database or schema. A login reaches every database on the server.
- Renewing or rotating a certificate on demand. Renewal is automatic.
- Shrinking storage.
- Changing the topology, zones, network, public reachability or master username after creation.
- A downgrade of the major version in place.
- Extensions outside the catalog from the console or API: among them those PGDG does not package
(
pg_hashids,pgjwt,plv8,pg_jsonschema) and those written for one other provider (pg_graphql,pg_net). The master user is a superuser and can stillCREATE EXTENSIONanything whose files are on the server. - A private endpoint for a cluster. Apps are given the cluster's load-balancer endpoint.
- A private endpoint for a single-node database created before private endpoints existed. Recreate it to get one.
- A failover without a write interruption. Every unplanned failover interrupts writes until a standby is promoted.
- Synchronous replication on a single node, or a mode that stops writes when no standby is available.
- Choosing the storage behind a database (for example the machine's local disk, or a separate volume for the write-ahead log).
- A published service-level objective for throughput. The benchmarks are measurements, not guarantees.
- A connection pooler or backups on a branch, and an unmasked branch while masking rules exist.
auto_explain(plans of slow statements) on a cluster.- Changing the machine type, storage, backup settings or disk keys from the console. Use the CLI, the API or Terraform.
Did this page help you?