Spanner

Spanner is a distributed relational database. It offers SQL, schemas, secondary indexes and ACID transactions — while distributing data across many machines and, if you choose, across regions.

Historically, systems forced a choice: relational guarantees on one machine, or horizontal scale with weaker consistency. Spanner's significance is that it provides both, using tightly synchronised clocks to order transactions globally.

For systems that genuinely outgrow a single-machine database but cannot give up transactional correctness — payments, ledgers, inventory, global user accounts — there are very few alternatives that do not involve building the consistency logic yourself.

The trade-off is cost. Spanner has a meaningful minimum spend and is substantially more expensive than Cloud SQL for small workloads. Choosing it for a database that would fit comfortably on one machine is a classic over-engineering error, and exams test that judgement.

• Instance — the provisioned capacity, measured in nodes or processing units, plus a chosen configuration.
• Instance configuration — regional for lower cost and latency in one region; multi-region for higher availability and global reads at higher cost.
• Split — the unit of data distribution. Spanner divides tables into ranges and spreads them across servers automatically.
• Primary key design — the most consequential decision you make. Keys determine how data is distributed, and a poor choice creates a hotspot that limits throughput.
• Interleaved tables — physically storing child rows next to their parent rows so related data is fetched together efficiently.
• Strong and stale reads — strongly consistent reads always see the latest data; slightly stale reads can be cheaper and faster when a few seconds of lag is acceptable.
• External consistency — the strongest guarantee Spanner offers: transactions appear in an order consistent with real time, globally.

• Financial ledgers and payment systems needing correctness at scale.
• Global inventory where overselling must be impossible across regions.
• Gaming platforms with worldwide player state and leaderboards.
• Telecommunications and logistics systems with very high transaction volumes.
• Consolidating many sharded databases that have become unmanageable.