Ads Click Aggregator

Apple's "Ads Click Aggregator" interview question focuses on designing a scalable system for real-time ad click processing and aggregation. It typically appears in data engineering, system design, backend, and real-time processing interviews at Apple.

Problem Statement

Design a system that ingests millions of ad click events per second from users across web and mobile platforms, performs real-time aggregation (e.g., by ad campaign, region, time window), ensures no duplicates or data loss, and serves low-latency queries for advertisers (sub-second for 1-minute granularity metrics). Key challenges include handling spikes up to 10K+ clicks/second, idempotency for retries, fault tolerance, and balancing accuracy with latency. The system must support global scale with regional variations.[1][3]

Input Format

Click events arrive via HTTP/gRPC POST requests as JSON payloads. Example:

{ "click_id": "C123-abc456", "ad_id": "A789", "campaign_id": "C789", "user_id": "U567", "timestamp": "2025-10-13T10:30:00Z", "region": "US", "device_type": "mobile", "ip_address": "192.168.1.1" }

• click_id: Unique identifier for deduplication.
• timestamp: For time-window bucketing (e.g., 1-minute intervals).
• Other fields enable segmentation (e.g., unique users per campaign).[3][1]

Output Format

Query APIs return aggregated metrics. Example response for advertiser query (GET /metrics?campaign_id=C789&region=US&start=2025-10-13T10:00:00Z&end=2025-10-13T11:00:00Z):

{ "campaign_id": "C789", "region": "US", "time_window": "10:30-10:31", "total_clicks": 54321, "unique_users": 12345 }

Aggregates use keys like (campaign_id, region, time_window). Served via dashboards or APIs.[1][3]

Constraints

• Scale: Peak 10K clicks/second globally; average 1K-5K.
• Latency: <1s for queries; near-real-time aggregation (<1min delay).
• Reliability: 99.99% uptime; no data loss (at-least-once semantics with idempotency).
• Idempotency: Handle duplicates from retries/network issues via click_id.
• Storage: Hot path for recent data (e.g., TTL 24h); cold for historical.
• Out-of-scope: Ad serving/targeting, fraud detection, conversions.[3][1]