Files
v10/internal/design/spf/features/content-steering.md
T

12 KiB

status, date, definition
status date definition
draft 2026-05-20 coarse

Content steering

HLS content-steering protocol implementation. Server-side advertises a dynamically-updateable host pool + pathway preferences via #EXT-X-CONTENT-STEERING:SERVER-URI="..." (Apple-defined HLS spec extension); client periodically fetches the steering manifest and biases URI selection accordingly. Sister to multi-cdn-failover — multi-CDN handles static alternate-URI lists; this feature handles dynamic server-advertised host preferences that update during playback.

A Media-src feature for sources that opt into content steering (server declares EXT-X-CONTENT-STEERING in the multivariant playlist). Layered with Player feature at Tier 2 (customer- policy hooks override server-advertised priorities).

Status

  • Composition: not implemented in createSimpleHlsEngine. Parser doesn't recognize EXT-X-CONTENT-STEERING tag; no steering-manifest fetch behavior; no pathway-priority state slot.
  • Definition depth: coarse — scope from the HLS content-steering spec + composition with multi-cdn-failover; SPF touchpoints sketched at the cluster level. Implementation details (steering-manifest data shape, reload-loop primitive sharing with live-stream-support, pathway-priority composition with multi-cdn-failover's URI list) tracked as open questions.
  • Hard prerequisites: network-resilience for steering-manifest fetch retry/backoff; multi-cdn-failover for the alternate-URI rotation primitive that pathway-priority composes with.

Phases of complexity

Tier 1 / Tier 2 framing. Each phase notes Naive vs Full depth where the Naive vs Full framing applies.

Phase Tier What Notes
Parse EXT-X-CONTENT-STEERING tag Tier 1 Parser extracts SERVER-URI and optional PATHWAY-ID from the multivariant playlist's content-steering tag. Presentation data shape gains a contentSteering: { serverUri: string; pathwayId?: string } field Parser extension; presentation-modeling's multivariant-parser surface grows. Naive: support the required attributes only. Full: support optional attributes (e.g., per-rendition PATHWAY-ID) and the PATHWAY-CLONES extension
Initial steering-manifest fetch Tier 1 On presentation resolve, if content-steering is declared, fetch the steering manifest from SERVER-URI. Parse the response (JSON: { VERSION, PATHWAY-PRIORITY, RELOAD-URI?, TTL, CDN-RESET-SECONDS?, PATHWAY-CLONES? } per spec) One-shot fetch on resolve; gates first URI selection if pathway-priority differs from manifest default ordering. Uses network-resilience retry/backoff for the steering-manifest fetch
Apply pathway-priority to URI selection Tier 1 Bias multi-cdn-failover's URI selection by the server-advertised pathway-priority order. Each alternate URI tagged with a pathway-id (from multivariant PATHWAY-ID or from steering-manifest mapping); pathway-priority orders them Composition with multi-cdn-failover is the cross-cutting question. Two shapes: (a) content-steering writes a pathwayPriority slot that multi-cdn-failover's rotation reads as the ordering policy; (b) content-steering reorders the active alternate-URI list directly. Lean: (a) — keeps multi-cdn-failover's slot semantics clean; pathway-priority is the bias input, rotation is the consumer
Periodic steering-manifest refresh Tier 2 After initial fetch, periodically re-fetch the steering manifest at the TTL interval (or from RELOAD-URI if different from SERVER-URI). New pathway-priority takes effect on subsequent URI selections Reload-loop shape similar to live-stream-support's playlist reload loop but for the steering manifest. Naive: simple TTL-driven re-fetch; pause on errors. Full: jitter + error handling via network-resilience. Open: share live-stream-support's reload-loop primitive vs separate
CDN-RESET-SECONDS handling Tier 2 Server-side mechanism to instruct the client to reset per-pathway circuit-breaker / cooldown state after a duration. When the steering manifest advertises CDN-RESET-SECONDS: N, mark all CDN-rotation state (from network-resilience's circuit-breaker + multi-cdn-failover's rotation history) as reset after N seconds Composes with network-resilience's circuit-breaker state + multi-cdn-failover's rotation history. Server can recover a CDN that the client marked unhealthy; spec mechanism for server-side health-signal propagation
Customer-policy hooks Tier 2 Pluggable hooks for consumer overrides: selectPathway(steeringManifest, history) → pathwayId, shouldFetchSteeringManifest(source) → boolean. Customer can override server-advertised priority, opt out of steering for specific sources, A/B test pathways Tier 2 customer-policy surface. Above the Tier 1 server-advertised defaults; hooks override when set

What's in scope vs out of scope

In scope:

  • All six phases above for HLS sources opting into content steering
  • Parser support for #EXT-X-CONTENT-STEERING tag (multivariant playlist) + per-rendition PATHWAY-ID attribute support
  • Steering-manifest JSON parser
  • Periodic reload-loop for the steering manifest (TTL + RELOAD-URI)
  • Pathway-priority state slot + composition with multi-cdn-failover
  • CDN-RESET-SECONDS handling — reset multi-cdn-failover rotation state + network-resilience circuit-breaker state on server cue
  • PATHWAY-CLONES extension (deferred to Full depth; verify in-scope vs separate)
  • Customer-pluggable pathway-selection + opt-out hooks

Out of scope (separate cluster G sister features):

  • network-resilience (prerequisite) — retry + backoff + circuit-breaker. Content-steering consumes for steering-manifest fetch retries.
  • multi-cdn-failover (prerequisite) — alternate-URI rotation primitive. Content-steering's pathway- priority is the dynamic-input variant of multi-cdn-failover's static-manifest input.

Out of scope (different architectural layer):

  • Steering-server implementation. Service-side concern.
  • Pathway-priority decision algorithm at the server (load balancing, cost optimization, regulatory routing). Service-side.
  • Customer-facing UI for "current pathway" / "steering active" indicators.
  • Steering-manifest hosting / TLS / authentication infrastructure.

Likely cross-cutting impact

Things this feature probably forces decisions on, not just additions:

  • Composition shape with multi-cdn-failover. Per phase 3: write a pathwayPriority slot that multi-cdn-failover reads (option a) vs reorder the active alternate-URI list directly (option b). Lean (a) — keeps multi-cdn-failover's slot semantics clean. Open: should pathwayPriority be a constraint+filter (narrowing candidates) or a sort key (ordering candidates)? Sort key matches the spec's intent.
  • Reload-loop primitive sharing. Live-stream-support's reload loop (not yet implemented) and content-steering's steering-manifest reload loop share shape: periodic re-fetch with TTL pacing, error handling, source-bound lifecycle. Two implementations or a shared primitive? Worth designing the primitive at live-stream-support's landing or when content-steering implementation work begins.
  • CDN-RESET-SECONDS interaction with circuit-breaker. Server's reset instruction supersedes client-side health tracking. Cross- cutting with network-resilience's circuit-breaker (whose state reset is the implementation) and multi-cdn-failover's rotation history (which feeds preferred-URI selection). The reset clears all three state surfaces. Composition question: where does the reset behavior live? Lean: in content-steering (it's the consumer of the spec-defined reset), but the writes affect network-resilience's + multi-cdn-failover's state slots.
  • Per-rendition vs presentation-wide pathway-id. HLS spec supports per-rendition PATHWAY-ID (different bitrate variants can advertise different pathway memberships) and presentation-wide declarations. Both should be supported; affects parser shape and pathway-priority application shape (per-rendition pathway-priority resolution).
  • PATHWAY-CLONES extension. HLS spec allows the steering manifest to generate new URIs via URL-rewriting rules (PATHWAY-CLONES). Real complexity — affects parser, alternate- URI surface, multi-cdn-failover's URI list. Likely Full-depth or deferred to a follow-on.
  • Steering-manifest fetch authorization. Does the steering- manifest fetch use the same playback token as media fetches (Mux-style)? Spec doesn't mandate; customer/integration question. Lean: same auth as content URIs (consumer's existing token- handling propagates).
  • Source-replacement cascade. Steering-manifest reload loop + pathway-priority state tear down with the source via the standard resolved/unresolved cascade.

Open questions

  • Composition shape with multi-cdn-failover. Slot-bias (option a) vs URI-list-reorder (option b). Lean a.
  • Reload-loop primitive sharing. Reuse live-stream-support's primitive vs separate. Resolution likely when both features land implementation work.
  • Per-rendition vs presentation-wide pathway-id semantics. Spec supports both; implementation favors which?
  • PATHWAY-CLONES scope. In-scope (Full depth) vs deferred.
  • CDN-RESET-SECONDS write coordination. Reset clears state in this feature, network-resilience, and multi-cdn-failover. Where the reset behavior physically lives (cluster F-style absorption-in- network-resilience vs cluster A-style separate behavior) is open.
  • Steering-manifest fetch authorization. Same playback token as media fetches? Customer-configurable?
  • Customer-hook contract. Function signatures + async semantics
    • failure handling for selectPathway and shouldFetchSteeringManifest. Harmonize with network-resilience + multi-cdn-failover hook shapes.
  • Mid-stream pathway changes during live + LL-HLS. Content steering + live + LL-HLS = three reload loops + dynamic pathway- priority updates. Cross-feature coordination complexity. Out of scope for v1.
  • network-resilience (hard prerequisite) — retry/backoff for steering-manifest fetches + circuit-breaker state that CDN-RESET-SECONDS resets.
  • multi-cdn-failover (hard prerequisite) — alternate-URI rotation primitive. Content- steering's pathway-priority composes with multi-cdn-failover's static-manifest URI list.
  • presentation-modeling — multivariant-parser surface grows to recognize EXT-X-CONTENT-STEERING tag + per-rendition PATHWAY-ID.
  • live-stream-support (not implemented) — reload-loop primitive sharing question. Both features have similar periodic-fetch shape.
  • source-replacement — pathway- priority + steering-manifest state tears down via the resolved/ unresolved cascade.

See also