What is one reason for unexpected performance scaling issues in Cloud Spanner?

When considering the performance scaling issues in Cloud Spanner, the use of Version 1 UUIDs as primary keys that increase monotonically is an important factor. Cloud Spanner is designed to be a distributed database, where data is spread across multiple nodes. The primary key is used for data distribution and indexing.

Using monotonically increasing keys, such as Version 1 UUIDs, can lead to performance bottlenecks because they tend to cause hot spots in the database. These hot spots occur when many write operations are directed toward the same location, which can lead to contention and reduced throughput. Since Cloud Spanner is built to handle distributed transactions, having a uniform distribution of keys is essential to maximize parallelism and minimize latency.

In comparison, using other types of keys or methods that do not increase monotonically can distribute the writes more evenly across the nodes, enhancing the database’s ability to scale effectively. Thus, the choice of primary key and its distribution characteristics play a critical role in performance and scalability within Cloud Spanner.