Design Meeting Recording

Problem Scope and Interview Frame

A strong answer starts by separating live conferencing from cloud recording. The meeting service already moves RTP/WebRTC media between participants; the recording service receives authorized media taps, persists time-ordered artifacts, enriches them, and serves replay under strict access rules. Its first promise is not cinematic quality. Its first promise is that a meeting owner can trust the recording to exist after the call.

The scope includes host controls, participant consent, media capture, chunk durability, object storage, processing jobs, searchable metadata, captions, retention, permissions, audit, and replay. It excludes real-time call routing unless recording load feeds back into meeting quality. That boundary lets you protect the live call by treating recording as an attached durability pipeline with graceful degradation.

Interview stance

• State that recording is a multi-path system: hot ingest, cold processing, metadata indexing, and playback.
• Make per-track recording the default capture model, then explain when composite recording is worth the cost.
• Tie every design choice to missing-minutes risk, compliance exposure, replay latency, or storage cost.

Why interviewers care

Meeting Recording interviews reward crisp scope, explicit trade-offs, and failure stories—not generic microservice diagrams.

The failure that defines the design

The outage to narrate is the hour-long recording lost to a crash at minute fifty-nine. If the recording server buffers media in memory or holds the file open until the session ends, a single process crash or node failure destroys the entire recording — the most infuriating possible outcome, because the meeting genuinely happened and the bytes are gone. The fix is structural: capture writes media to durable storage continuously as segments, each a few seconds long with a monotonic sequence, appended to a growing manifest. A crash then costs at most the last unflushed segment, and a new capture server resumes appending. State durable-as-it-arrives segmented capture up front, because a recording you cannot trust to survive a crash is not a recording.