← Professional Cloud Architect: architecture and operations
02 / 8 · 40 MIN

Data, resilience, and events

Distinguish availability, recovery, event identity, and access patterns.

Concept and mechanism

Data design must address infrastructure loss, logical error, and repeated operations. These are different problems. Regional Cloud SQL HA uses primary and standby in distinct zones with synchronous replication for the described configuration. This alone does not cover regional loss. A replica in another region can have asynchronous lag; promoting that state can lose commits acknowledged by the primary. RPO bounds intended data loss and RTO bounds recovery time. Replicating an incorrect deletion to several destinations does not create a recoverable earlier state. Backups, retention, decryption capability, and recovery exercises complete the design. The application must also reconnect and validate outcomes after failure.

Guided application

In a fictional order flow, Pub/Sub exactly-once does not replace business idempotency. The guarantee applies to pull subscriptions and the documented regional scope, using message identity. Two publications of one order can have distinct IDs. Connect deduplication to the persistent effect to avoid a race between checking and debiting. For Bigtable telemetry, a timestamp-leading key can concentrate writes; distribute prefixes according to query patterns and measure the result. For object archives, compare total cost and read frequency rather than just price per GB. Cloud Storage Archive retains online access with cost and minimum-duration tradeoffs. Apply the same design discipline to an APS agent: restrict tools, preserve context, and evaluate errors before authorizing impactful actions.

IN PRACTICE

An operational replica can contain the same incorrect deletion; data presence does not prove historical recovery.

Common pitfalls

Replica as backup; acknowledgment as a unique debit; reversed timestamp as automatic distribution.

Related topics: Requirements, costs, and platform selection · Networking, provisioning, and capacity · Identity, perimeters, and keys

Take this idea with you

Design for infrastructure, logic, and event-identity failures.

Create account

Reference: Disaster recovery planning · Current linked standard guide; edition date unconfirmed (2026-09-30 inspection)