TopolearnPicking the wrong database is the most expensive architectural mistake to undo.
Google Cloud offers a database for most shapes of data. A practical map:
- Cloud SQL: managed PostgreSQL/MySQL/SQL Server on one primary. Default for typical OLTP apps.
- AlloyDB for PostgreSQL: PostgreSQL-compatible with a disaggregated storage layer, faster transactions, read pool instances and a columnar engine for analytical queries. For demanding Postgres workloads.
- Spanner: relational, SQL, strongly consistent, scaling writes horizontally across nodes and regions, with up to 99.999% availability in multi-region configs. For global or very large OLTP that outgrows one primary.
- Firestore: serverless document database (collections of JSON-like documents), strongly consistent, with real-time listeners and offline mobile/web SDKs. Pay per operation; great for app backends.
- Bigtable: wide-column NoSQL for huge throughput at single-digit millisecond latency: time series, IoT, ad tech. Access by row key only.
- Memorystore: managed Valkey, Redis and Memcached for caching.
- BigQuery: analytics warehouse, not an OLTP store.
Questions to ask: Do I need joins and multi-row transactions (relational)? How big will writes get? Is global distribution required? What access patterns exist (key lookups, ranges, ad-hoc analytics)? Is per-operation or per-node pricing better at my traffic?
Example
# pip install google-cloud-firestore
from google.cloud import firestore
db = firestore.Client(project="shop-dev-4821")
users = db.collection("users")
users.document("ada").set({"name": "Ada", "plan": "pro", "age": 36})
users.document("bob").set({"name": "Bob", "plan": "free", "age": 29})
query = users.where(filter=firestore.FieldFilter("plan", "==", "pro"))
for doc in query.stream():
print(doc.id, doc.to_dict()["name"])
# ada Ada
# Listen for live changes (real-time updates)
watch = users.document("ada").on_snapshot(lambda docs, changes, ts: print(docs[0].to_dict()))Remember
- Cloud SQL by default; AlloyDB for heavy Postgres; Spanner when one primary is not enough
- Firestore for serverless app documents with real-time sync
- Bigtable for massive key-range workloads; BigQuery for analytics
Watch out for
- Choosing Spanner or Bigtable for a small app and paying for capacity it never uses.