Coinbase details architecture behind 1.5m reads-per-second identity service

Coinbase has described how its Identity Platform’s Users service, a core dependency for critical user journeys such as login, multi-factor authentication and account management, is engineered to cope with “extreme, bursty traffic of the crypto market,” serving more than 1.5 million reads per second at peak, according to a company statement.

The company said the Users service underpins user management across Coinbase products via a common data model and API, and must meet requirements for scalability, write correctness, extensibility and operability. It cited goals to “sustain multi-million RPS read traffic with consistent, low latency under bursty load” and to “maintain high availability SLOs and have predictable fallbacks when dependencies degrade.”

To manage complexity and throughput, Coinbase said it organizes user data into “Fragments” such as Addresses, PhoneNumbers and TaxID, replacing a monolithic GetUser API with independent Fragment APIs backed by separate data stores. This allows high-load fragments like PhoneNumbers to scale without affecting others such as UserAgreements, while change data capture-based “Fragment Events” and Snowflake integration support event-driven use cases and offline analytics, the company said.

On the infrastructure side, Coinbase said it moved from an internal Cache Router sidecar for Memcache to ValKey, an open-source Redis fork managed by AWS, using native connections from API pods to simplify operations. The Users service also shifted from relying solely on MongoDB to a federated persistence model that adds DynamoDB for frequently accessed key-value datasets, with a custom framework using an auxiliary table and multi-item transactions to enforce unique constraints where needed, according to the statement.

For data correctness under concurrent writes, Coinbase said it uses an optimistic concurrency control protocol based on per-resource version fields and compare-and-swap semantics, alongside “Freshness Tokens” that bind to specific resource versions to guarantee strong read-after-write behavior for critical flows by bypassing caches when required. To avoid MongoDB “death loops” during persistent overload, the company introduced recurring load tests and a dedicated load-shedding mechanism that adjusts the Application Load Balancer to divert some traffic to a dummy target group so the Users service can “degrade gracefully rather than collapsing.”

“By moving from a monolithic service to a federated system of Fragment APIs, federated storage, and intelligent consistency controls, the Users service now sustains over 1.5 million reads per second during market surges,” Coinbase said, adding that the architecture is intended to provide the scalability and resilience needed to support the company’s broader cryptoeconomy mission.

The recap

• Coinbase’s Users service in Identity Platform handles critical user journeys.
• Federated Fragment APIs and polyglot storage support over 1.5M peak RPS.
• Optimistic concurrency, freshness tokens and load shedding protect correctness, availability.