Spanner

Cloud Spanner is Google's globally distributed relational database — the only commercial database offering strong ACID consistency across multiple geographic regions with horizontal scaling. For most workloads Spanner is significantly more expensive than regional databases, but for applications requiring strong global consistency (financial transactions, inventory management across regions), it's uniquely capable. For agents: use the Spanner client library for reads, writes, and transactions; interleaved tables optimize related data co-location; and partitioned DML enables large-scale updates without transaction size limits. Confidence is docs-derived.

Cloud Spanner Client Library (Python). database = spanner_client.instance("my-instance").database("my-db"). Read: with database.snapshot() as snapshot: results = snapshot.execute_sql("SELECT * FROM users WHERE userId = @userId", params={"userId": user_id}, param_types={"userId": spanner.param_types.STRING}). Write: def write_transaction(transaction): transaction.insert("users", columns=["userId", "name", "email"], values=[("user-123", "Alice", "alice@example.com")]). database.run_in_transaction(write_transaction). Batch mutations: database.batch().insert("users", ...).commit(). DML: transaction.execute_update("UPDATE users SET plan = 'pro' WHERE userId = @id"). Partitioned DML: database.execute_partitioned_dml("DELETE FROM events WHERE created < TIMESTAMP_ADD(CURRENT_TIMESTAMP(), INTERVAL -30 DAY)").

Service Account auth: spanner_client = spanner.Client(project=project_id, credentials=credentials). Application Default Credentials: spanner.Client() for GCP-hosted agents. IAM roles: roles/spanner.databaseUser (read/write), roles/spanner.databaseReader (read-only). HTTPS enforced. No OAuth2. Instance and database: separate resources — instance provisions capacity, database holds schema and data. Multi-region replication: configurable at instance creation (single region, multi-region, custom).

Client library handles retry and backoff automatically. Deadlock detection: Spanner detects deadlocks and aborts one transaction (retry aborted transactions). Transaction size limits: 100MB per commit (avoid large transactions). Read-write transactions: require locking — minimize lock contention with short transactions. Read-only transactions: cheaper (no locking), use for queries that don't need to see latest writes. Spanner uptime: 99.999% SLA (5 nines) for multi-region configs. Cloud Spanner status at status.cloud.google.com. Cost: significantly higher than Cloud SQL — validate pricing before adoption.

cloud.google.com/spanner/docs covers Python client guide, SQL reference, transaction types, interleaving guide, and performance optimization. Getting started: GCP free trial ($300 credit), create Spanner instance, install google-cloud-spanner — first queries in under 15 minutes. google-cloud-spanner: pip install google-cloud-spanner. Community via Google Cloud Community and Stack Overflow (google-cloud-spanner tag). Comprehensive documentation for a technically complex globally distributed database.