← Discord Interview Insights

Discord·Software Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
May 2026

Summary

Data engineering interview at Discord focused entirely on warehouse modeling for server activity data. One meaty design question that covered a lot of ground, felt more like a working session than a traditional interview.

Questions Asked (1)

Q1

Design a fact/dimension data model for Discord server activity, given raw event tables for server creation, server views, server joins, and messages. Define your dimension and fact tables, explain your grain choices, how you'd handle slowly-changing dimensions for users and servers, how you'd partition the data, and what query patterns the model is built to serve.

Data ModelingSystem DesignTechnical Trade-offs
Author's notes

This was the whole interview basically.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business questions the model must answer, then design a star schema with conformed dimensions and fact tables at the appropriate grain. Walk through grain choices, SCD handling, partitioning, and query patterns, explicitly tying each decision to Discord's scale and analytical needs.

Pro tip: Emphasize that the fact grain should be the most atomic event possible (e.g., one row per message) to maximize flexibility, and use surrogate keys in dimensions to handle SCDs without impacting fact table joins.

1. Clarify business questions and scope

Ask what analyses the model should support (e.g., daily active servers, user engagement trends, retention) and confirm the available raw event tables and their schemas.

2. Identify dimensions and facts

List the core dimensions (user, server, date, channel, etc.) and fact tables (server_activity, message_activity) that capture measurable events at the desired grain.

3. Define grain and SCD strategy

Choose the grain for each fact table (e.g., one row per message, one row per server join) and decide how to handle slowly-changing dimensions for users and servers (Type 2 for historical tracking).

4. Plan partitioning and storage

Partition fact tables by date (and possibly server_id) to enable efficient time-range queries and manage large data volumes; consider clustering for common filters.

5. Map query patterns and validate

Enumerate the key queries the model will serve (e.g., messages per server per day, user retention) and verify that the schema supports them efficiently without excessive joins.

Key Points to Mention

  • Star schema design with conformed dimensions for user, server, date, and channel.
  • Fact table grain: atomic events (e.g., one row per message, one row per server join) to allow flexible aggregation.
  • Slowly-changing dimensions: Type 2 for users (tracking username changes, etc.) and servers (tracking name changes, ownership transfers).
  • Partitioning strategy: partition fact tables by date (e.g., daily) and possibly sub-partition by server_id for scalability.
  • Query patterns: support for time-series analysis, funnel analysis (views → joins → messages), and user/server retention cohorts.
  • Trade-offs: denormalization for performance vs. storage, and handling late-arriving data with partitioning.

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