Build a fault-tolerant Logging Pipeline
Last updated: October 21, 2025
Quick Overview
Design a fault-tolerant logging system that handles millions of requests. Discuss trade-offs in consistency, availability, and performance.
50
13
4,547 solved
Design a fault-tolerant logging system that handles millions of requests. Discuss trade-offs in consistency, availability, and performance.
This ML system design question from LinkedIn's Onsite tests your ability to think about ML systems at scale. The interviewer expects discussion of data quality, feature stores, model serving infrastructure, and A/B testing strategy.
What the Interviewer Expects
- Design the full ML lifecycle from data collection to model monitoring
- Address cold start, exploration/exploitation, and model freshness
- Discuss multi-objective optimization and ranking systems
- Plan for model debugging, fairness, and bias mitigation
- Design the feature store and training pipeline for scale
- Address model versioning, canary deployments, and rollback strategies
- Discuss the data flywheel and long-term system evolution
Key Topics to Cover
How to Approach This
- Start by clarifying functional and non-functional requirements with the interviewer.
- Estimate the scale: QPS, storage, bandwidth. This drives your design decisions.
- Draw a high-level architecture first, then deep dive into 1-2 critical components.
- Discuss trade-offs explicitly (e.g., consistency vs availability, SQL vs NoSQL).
- Address failure scenarios, monitoring, and how the system handles 10x traffic spikes.
Possible Follow-up Questions
- What would you do if model performance degrades over time?
- How would you handle a 10x increase in prediction requests?
- How would you handle the cold start problem?
- How would you debug a model that works well offline but poorly online?
Practice a Similar Problem on Codemia
Solve a related problem with our interactive workspace, get AI feedback, and view detailed solutions.
Solve on CodemiaSample Answer
Requirements
Functional Requirements
- Log Ingestion: The system must ingest millions of log events per second from various sources (e.g., user interactions, system metrics).
- Storage: Logs must be ...
Capacity Estimation
Assuming that LinkedIn has around 900 million users and each user generates approximately 10 log events per day, we can estimate:
- Total Logs per Day: 900 million users * 10 logs/user = 9 billio...
Submit Your Answer
