Design Video Manifest Server - System Design | WinJob

Manifest Server Context and Platform Goals

A video manifest server is the control plane between packaged segments and every player session. It materializes HLS master/media playlists and DASH MPD documents, binds signed segment URIs, and must stay coherent when ladders, DRM keys, or live windows advance.

Focus

Separate manifest bytes (kilobytes) from segment payloads (megabytes). Players poll manifests frequently; segments are fetched far less often per second of wall clock. The server therefore optimizes for read-heavy, cache-friendly text generation—not transcoding.

Metrics and assumptions

Assume 500M MAU, 22M peak concurrent playbacks, 1.8M manifest RPS at peak (master + media refresh). Target p95 manifest assembly < 35ms origin, < 8ms edge cache hit. Master playlist < 4KB; media playlist window 6 segments for live.

Failure modes

Serving a media playlist whose segment list lags the packager by one GOP causes visible stalls—publish manifests only after segment index commit is durable.

Product parallels

Netflix Open Connect pushes immutable segment paths; YouTube uses versioned path prefixes; Mux documents static renditions with dynamic signed URLs in playlists.

Whiteboard checkpoint: draw Manifest Server Context and Platform Goals with packager index, manifest assembler, CDN cache keys, and signing boundary unique to this slice.

Operational note 1: canary profile v1 before fleet-wide pointer swap.

Capacity lens 1: recompute manifest RPS when concurrent playbacks shift above 22M.

Interview pivot 1: if challenged on piracy, explain short-lived signed URLs and ASN rate limits.

javaOne Dark Pro

1 public record ManifestSlo(Duration p95Assembly, double minCacheHitRatio) {}

pythonOne Dark Pro

1 @dataclass(frozen=True)
2 class ManifestSlo:
3     p95_assembly_ms: int
4     min_edge_hit: float

typescriptOne Dark Pro

1 interface ManifestSlo { p95AssemblyMs: number; minEdgeHitRatio: number; }

Closing Manifest Server Context and Platform Goals: bind decisions to assembly latency, stale-manifest SLI, edge hit ratio, and clean rollback via publication pointer.

Why interviewers care

Video Manifest Server interviews reward crisp scope, explicit trade-offs, and failure stories—not generic microservice diagrams.

Interview checkpoint

Name one failure story for Manifest Server Context and Platform Goals that proves you understand real outages, not happy-path diagrams.

Key Highlights

•Manifest server sits between packager index and CDN edge
•Master vs media playlist responsibilities differ
•Signed URLs are capabilities, not identities

1	@dataclass(frozen=True)
2	class ManifestSlo:
3	p95_assembly_ms: int
4	min_edge_hit: float