persistence-production

Cloud Run has an ephemeral filesystem, so any data written locally disappears on restart or redeploy. Configure a durable external provider such as a managed database for production, and reserve filesystem storage for dev and test environments only.

Map the persistence boundary by listing every adapter and tracing how the provider is selected per environment. Then confirm whether the target runtime has a durable filesystem and verify that production always resolves to an external durable provider.

Cloud Run containers are stateless, so writes to the local filesystem are discarded when instances are replaced. The usual root cause is a filesystem adapter being selected in production, often through a silent fallback when the durable provider fails.

No. Silent fallback to filesystem storage in a stateless runtime hides failures and guarantees data loss. Production should fail closed, surfacing the error instead of degrading to non-durable storage.

Write tests that simulate provider selection per environment and verify fail-safe behavior when the durable provider is unavailable. These tests assert that production never resolves to filesystem storage, without needing live credentials.