Approach: Start with core requirements (latency, throughput, consistency), then design layered architecture with caching, distributed database sharding, and decoupled feed ranking service.

Considerations: Balance consistency vs. availability for billions of users. Consider cross-region replication and trade-offs between personalization depth and latency.

Key Points: Mention caching strategies, fan-out patterns, and handling 100K+ requests per second per feed with sub-100ms latency.

Approach: Gather metrics and logs to isolate the service, profile CPU usage with perf or flame graphs, identify the hot code path, then verify the fix.

Considerations: Prioritize visibility without impacting production. Determine if the issue is algorithmic, resource contention, or infrastructure-related.

Key Points: Reference perf, CPU profilers, and APM dashboards. Apply 80/20 rule for bottlenecks and test the fix safely in staging before rollout.

Approach: Implement token bucket or sliding window algorithm with distributed state (Redis), multiple limit tiers (per-client, per-endpoint, global), and clear configuration framework.

Considerations: Balance strict enforcement with developer experience. Handle burst spikes and graceful degradation when rate limiter becomes a bottleneck.

Key Points: Fast decision-making (<1ms), idempotent limit checks, different SLAs per team, and self-service dashboard for monitoring and adjustment.