Question

A financial services company runs a globally distributed order-management system on Cloud Spanner. Their most frequent query retrieves a customer record along with all of that customer's recent orders. Currently, Customers and Orders are separate top-level tables joined on customer_id, and the team observes higher-than-expected read latency because related rows are scattered across different splits. They want to physically co-locate each customer's orders with the customer row to improve read performance for these access patterns. What should they do?

Accepted Answer

Define Orders as an interleaved child table of Customers using INTERLEAVE IN PARENT, keyed by the customer_id prefix.

Answer

Add a secondary index on Orders.customer_id and force all queries to use it with a FORCE_INDEX hint.

Answer

Denormalize by copying all order columns into the Customers table as a repeated STRUCT field.

Answer

Increase the number of Spanner nodes so that more splits can be cached in memory, reducing join latency.