Cloud Spanner Complete Guide (2026)

Spanner is Google's globally distributed NewSQL database, delivering SQL, ACID transactions, strong consistency, and horizontal scale all at once. Battle-tested inside Google (AdWords, Gmail, YouTube) for years, it is one of the very few global databases that offer a 99.999% SLA.

Key Features of Spanner

Strong consistency + horizontal scale: effectively sidesteps the CAP theorem (built on TrueTime)
SQL + ACID: transactions, JOINs, and aggregate queries are all supported
Global distribution: multi-region configurations can span 5 continents
99.999% SLA: for multi-region configurations (under 5 minutes of downtime per year)
Zero-downtime scaling: just add more PUs (Processing Units)
Two SQL dialects: GoogleSQL and the PostgreSQL Interface

TrueTime

The core technology behind Spanner. By combining GPS receivers and atomic clocks deployed across Google data centers, TrueTime represents distributed time as a bounded interval. Standard NTP has uncertainty of several to tens of milliseconds, but TrueTime keeps the uncertainty strictly bounded (typically under 7 ms). That bound is what lets Spanner provide global External Consistency (a guarantee of global ordering).

Pricing

Item	Price (us-central1, 2026)
Processing Unit (PU)	$0.90 / 100 PU / hour (Regional)
Storage	$0.30 / GB / month (Regional)
Multi-region PU	$3 / 100 PU / hour
Multi-region storage	$0.50 / GB / month
Backup	$0.10 / GB / month
Network egress	Standard egress rates

You can start at the 100 PU minimum (formerly 0.1 node). 1000 PU is equivalent to 1 node, which translates to roughly 10,000 QPS as a throughput guideline.

Configuration Patterns

Configuration	Use case	SLA
Regional	Regional apps	99.99%
Dual-region	Two-region HA	99.999%
Multi-region (3+)	Global deployment	99.999%

The Two SQL Dialects

GoogleSQL (formerly ZetaSQL): the same dialect as BigQuery. Strong support for arrays, STRUCTs, and window functions
PostgreSQL Interface: PostgreSQL wire-compatible, so existing OSS tooling (psql, pg_dump, SQLAlchemy) works out of the box

Spanner Graph (2024+)

Run Property Graph queries directly on Spanner
ISO GQL standard with openCypher compatibility
Used for knowledge graphs, fraud detection, social networks, and recommendation systems
Coexists with existing relational data in the same database

Comparison with Other NewSQL Databases

Item	Spanner	AWS Aurora DSQL	CockroachDB	Azure Cosmos DB
Consistency	External Consistency	Serializable	Serializable	5 selectable levels
SQL	GoogleSQL / PostgreSQL	PostgreSQL	PostgreSQL	NoSQL-centric + SQL API
SLA	99.999%	99.99%	Self-managed	99.999%
Pricing	PU hours	Pay-as-you-go	VM hours	RU hours
Track record	10+ years inside Google	GA in 2024	OSS / commercial	Numerous

Typical Use Cases

Player data for global games
Financial transactions (low latency + strong consistency + compliance)
E-commerce order management
Multi-tenant SaaS (tenant isolation + global rollout)
Ad serving (real-time bidding + global reach)
Knowledge graphs (via Spanner Graph)

Best Practices

Primary key design: avoid hotspots (don't use sequential numbers or timestamps; prefer UUIDs or hashes)
Interleaved Table: physically co-locate parent-child rows to speed up JOINs
Mutation limit: 80,000 mutations per transaction is the hard cap
Query optimization: use Query Insights to detect slow queries
Backup: automatic backups plus Point-in-time Recovery (PITR window of 1 hour to 7 days)
Multi-region: serve reads from the local region; writes are funneled to the leader

Should I choose Spanner or Cloud SQL?

Choose Spanner if you need global distribution, horizontal scaling, and a 99.999% SLA. Choose Cloud SQL for region-bound workloads or when you want to keep using MySQL, PostgreSQL, or SQL Server. Spanner is pricier, so the right pick depends on your requirements.

What is the difference between Spanner and Bigtable?

Spanner is a NewSQL database (SQL + ACID transactions + strong consistency). Bigtable is a NoSQL wide-column store (high-throughput time-series workloads). The standard practice is to keep them strictly separated by use case.

What is TrueTime?

A distributed time API backed by GPS receivers and atomic clocks in Google data centers. It is the foundational technology that lets Spanner provide global ordering (External Consistency).

How is Spanner priced?

You pay for compute by the hour (node/PU time), plus storage and network. The minimum is 100 PU (Processing Units, formerly 0.1 node). Global multi-region configurations are notably more expensive.

What is Spanner Graph?

A 2024 feature that lets you run Property Graph queries (Cypher compatible) directly on Spanner. Common use cases include knowledge graphs, fraud detection, and social networks.

How does the PostgreSQL Interface differ from GoogleSQL?

Spanner's PostgreSQL Interface supports the PostgreSQL wire protocol and a syntax subset, making it easy to migrate existing PostgreSQL apps. You pick one of two dialects: PostgreSQL Interface or GoogleSQL (formerly ZetaSQL).

How does Spanner compare to other cloud NewSQL databases?

AWS Aurora DSQL (GA in 2024) is Spanner-like, Azure Cosmos DB is NoSQL-centric, and CockroachDB is an OSS Spanner clone. For sustaining a 99.999% SLA, Spanner has by far the longest track record.

Is there a free trial for learning Spanner?

You can experiment with the 90-day Free Trial and $300 in credits. At the 100 PU minimum, you can start for just a few thousand yen (a few tens of dollars) per month.

Check what you learned with practice questions

Practice with certification-focused question sets

View GCP Exam Prep

Author

NicheeLab Editorial Team

NicheeLab editorial team focused on data engineering and cloud certification learning. Content is structured around practical study needs and official exam domains.

Spanner Complete Guide: Global Distributed NewSQL Features, Pricing & DB Comparisons