Release note — databases family, 2026-10-21
What the databases family covers at first release, the two facts it leads with, the absences recorded in four vendor pages including the missing error strings, and what is known to be missing.
markdownreferencev1DGYR-MAAGstage: designtrust: officialfamily: databasesservice: cloud-sqlinterface: anychangelogdatabasesrelease notecloud SQLalloydbspannerfirestorebigtable
Databases family — first release
Added
| Document | What it carries |
|---|---|
reference/databases/cloud-sql.md | three engines, Enterprise Plus, the HA shape, four connection paths, the restarting update category |
reference/databases/alloydb.md | the cluster/instance/node model, three instance types, the columnar engine |
reference/databases/spanner.md | the three configurations with their stated availability and write-latency trade |
reference/databases/firestore.md | the two modes, real-time updates, MongoDB compatibility |
reference/databases/bigtable.md | the data model, data on Colossus not on nodes, the consistency-by-cluster-count rule, HBase |
reference/databases/errors.md | indexed by symptom and configuration, with an explicit note on missing vendor strings |
guides/databases/connect-to-cloud-sql-with-iam-auth.md | seven steps to a connection with no stored credential |
guides/parity/databases.md | the two rows where no counterpart is honest, and the provisioned-capacity against request-unit trap |
The two headline facts
- Adding a second Bigtable cluster changes consistency from strong to eventually consistent by default.
It produces no error, which makes it the hardest class of failure in this family to diagnose.
- Spanner has no like-for-like counterpart on AWS or Azure — Microsoft's own table offers two unrelated
services for it. A move off Spanner is an architecture change and must be costed as one.
Absences recorded rather than filled in
- The Cloud SQL pages publish no verbatim connection error messages. The error catalogue says so and is
indexed by symptom instead, rather than inventing strings.
- The Cloud SQL overview states no version end-of-support information, which is normally one of the
highest-value facts in a database migration.
- The Firestore modes page does not say whether a database's mode can be changed after creation.
- The Spanner instance-configurations page states no minimum for nodes or processing units.
Known missing
- Memorystore and Azure/AWS cache parity beyond the name row.
- Backups, point-in-time recovery and retention on every engine.
- Read replicas on Cloud SQL (as distinct from AlloyDB read pools).
- Firestore security rules and index management.
- Spanner schema design, interleaved tables and the PostgreSQL interface.
- Engine-level error catalogues: these come from MySQL, PostgreSQL and SQL Server, not from Google Cloud.