Question 820
The service owner asks how to avoid a repeat of: A team creates two Cloud SQL read replicas but an internal load balancer sends INSERT statements to a replica. Which operational check is the best match?
Explanation
Correct Answer
Run negative write-on-replica tests after routing changes. The recurring risk is application traffic routing does not distinguish read-only endpoints from the writable primary. This check makes that risk visible before another incident.
Incorrect Answers
The proposed check would test quota and permission failures in infrastructure automation, helping with cases where a database initialization script retries failed CREATE TABLE statements indefinitely and repeatedly hits service quotas. The recurring exposure described here is that application traffic routing does not distinguish read-only endpoints from the writable primary.
The proposed check would review HA and read-serving components as independent architecture layers, helping with cases where a company claims two read replicas provide automatic HA for its Cloud SQL primary without enabling the approved failover configuration. The recurring exposure described here is that application traffic routing does not distinguish read-only endpoints from the writable primary.
This control would set per-replica lag alerts based on business use of its reads; it responds to the risk in which a read replica of a large MySQL instance increasingly lags after a nightly reconciliation writes millions of rows. It would not address the current root cause: application traffic routing does not distinguish read-only endpoints from the writable primary.