Advanced GraphQL Operations

Persisted / Automatic Persisted Queries (APQ)

Mobile clients ship with query hash only; server maps hash → full query text. Benefits:

• Smaller payloads
• Allowlist — unknown hashes rejected (security)
• CDN-friendly GET requests for reads (?extensions={\"persistedQuery\":{\"sha256Hash\":\"...\"}})

Requires release pipeline to register queries.

@defer and @stream (incremental delivery)

Return partial data first, later patches for slow fields — improves perceived latency on mobile.

Operational cost: more complex execution engine, caching interactions.

Subscription scaling

Architecture:

Kafka topic → subscription service → WebSocket fan-out → N clients

Challenges:

• Connection count per pod (file descriptors, memory)
• Auth refresh on long connections
• Fan-out when one event → 10k subscribers — partition by shard/key
• Backpressure when client reads slowly — drop vs buffer (memory risk)

Federation tuning

• Entity resolver batching across subgraphs
• Query plan caching at gateway
• Schema composition CI with breaking change detection (Rover)

GraphQL at staff interviews

Discuss operational guardrails (cost analysis, APQ, timeout) not just schema design — distinguishes senior from mid.