Implement an API caching layer with offline support and storage trade-offs

What's really being asked

This question evaluates whether you can design a production-ready caching architecture beyond simple key-value storage. Interviewers want to see tiered thinking, knowledge of Flutter-specific persistence options, and awareness of the hard problems in distributed systems: invalidation, consistency, and offline conflict resolution. They are looking for product sense too, meaning you choose policies based on data freshness requirements rather than applying one pattern everywhere.

The full answer

First, a two-tier hierarchy: an in-memory LRU cache for sub-millisecond access to hot data and a persistent on-disk cache for offline resilience and larger payloads. Second, storage selection aligned to data shape: use Hive or ObjectBox for fast key-value JSON blobs, SQLite for relational data requiring queries and joins, and SharedPreferences only for small flags or timestamps. Third, fetch policies per endpoint: cache-first for offline availability, network-first for critical fresh data, and stale-while-revalidate for feeds where immediate pixels matter more than absolute freshness. Fourth, invalidation mechanics: TTL for simplicity, ETag or If-None-Match to save bandwidth via 304 responses, and versioning for bulk invalidation. Fifth, offline write strategy: an optimistic local update backed by a sync queue that replays mutations when connectivity returns, using idempotency keys to make retries safe.

The mistakes people make

Treating in-memory and disk as interchangeable rather than complementary layers. Picking Hive for everything without acknowledging SQLite is better for relational queries, or vice versa. Proposing a cache without any invalidation strategy. Ignoring the difference between read caching and offline write queuing. Suggesting that cache size is unlimited or that eviction is not necessary on mobile devices with constrained storage.

What usually comes next

How would you handle cache eviction under memory pressure on Android and iOS? What is your conflict resolution strategy when two offline clients mutate the same record? How do you test offline behavior without relying solely on airplane mode? How would you migrate the cache schema when your API response shape changes? What metrics would you log to prove the cache is actually improving perceived performance?

A concrete example

A news feed endpoint should use stale-while-revalidate. On app launch, the UI renders immediately from a Hive box storing the last JSON payload. A background Dio interceptor fires a network request sending an ETag from the previous response. If the server returns 304 Not Modified, no JSON is parsed and bandwidth is saved. If new data arrives, the Hive box and an in-memory LRU cache are atomically updated and the UI refreshes. If the user likes an article while offline, the app writes the optimistic like to Hive and enqueues the mutation in a sync queue with a UUID idempotency key; when the device reconnects, the queue replays the request, and if it fails permanently the UI rolls back and surfaces a snackbar.