Meta·Software Engineer·Onsite - System Design / Architecture·Staff

StaffPrefer not to say

May 2026

Summary

Meta system design round, one big question about ad-click tracking at massive scale. The scope was brutal and I felt like I was playing whack-a-mole with follow-ups the whole time.

Questions Asked (1)

Design an ad-click tracking system that handles around 100 million click events per second, supports a near-real-time advertiser dashboard showing impressions, clicks, CTR, and conversions, and stores data for analytics and billing purposes.

System DesignTechnical Trade-offsData Modeling

Author's notes

I started with the ingest layer which felt safe, load balancers into edge collectors into Kafka, and the interviewer just nodded and waited.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements and scale, then propose a high-level architecture that separates ingestion, real-time processing, and storage. Focus on trade-offs between latency, consistency, and cost, and detail how to achieve 100M events/sec with partitioning and distributed systems.

Pro tip: Emphasize the importance of idempotency and exactly-once processing to avoid double-counting clicks, which is critical for billing accuracy. Also, discuss how to handle late-arriving data and its impact on real-time dashboards.

1. Clarify Requirements and Scale

Ask questions to understand data volume, latency requirements for the dashboard, retention period, and consistency needs for billing. Confirm the 100M events/sec scale and estimate storage needs.

2. High-Level Architecture

Propose a pipeline: ingestion layer (e.g., Kafka) to handle high throughput, stream processing (e.g., Flink) for real-time aggregations, and storage layers for real-time and batch analytics. Include a serving layer for the dashboard.

3. Deep Dive into Components

Detail each component: how to partition Kafka topics, use of windowing in stream processing, choice of databases (e.g., Druid for real-time OLAP, HBase for raw storage), and how to ensure scalability and fault tolerance.

4. Address Trade-offs and Challenges

Discuss trade-offs: latency vs. accuracy, cost vs. performance, and how to handle late data, duplicates, and exactly-once semantics. Explain how to scale each layer and handle failures.

5. Summarize and Validate

Recap the design, ensuring it meets all requirements, and invite feedback. Mention potential optimizations and future improvements.

Key Points to Mention

Use of Kafka or similar distributed log for ingestion with partitioning to handle 100M events/sec.
Stream processing with Flink or Spark Streaming for real-time aggregations and windowing.
Lambda architecture or Kappa architecture to balance real-time and batch processing.
Storage solutions: real-time OLAP (e.g., Druid, ClickHouse) for dashboard, and data lake (e.g., S3, HDFS) for analytics and billing.
Exactly-once processing and idempotency to ensure billing accuracy.
Handling late-arriving data with watermarks and allowed lateness, and its impact on dashboard accuracy.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.