Summary#
WS08-008 introduces dedicated RunPod endpoint classes for audio-heavy Isis
workflows so audio traffic can be routed to specialized serverless lanes instead
of sharing only generic image/video families.
Implemented Endpoint Classes#
audio-sfx: optimized lane fortext-to-audioworkflows.audio-voice: optimized lane forvoice-synthesisworkflows.audio-music: optimized lane formusic-generationworkflows.
Implemented for both production and staging desired-state inventories with
separate stable/canary endpoint environment variables, worker bounds, GPU class,
idle timeout, and execution timeout definitions.
Contract and Routing Updates#
- Added new feature-flag/env contracts for audio endpoint IDs and canary rollout
controls:
RUNPOD_AUDIO_SFX_ENDPOINT_IDRUNPOD_AUDIO_VOICE_ENDPOINT_IDRUNPOD_AUDIO_MUSIC_ENDPOINT_IDRUNPOD_AUDIO_SFX_CANARY_ENDPOINT_IDRUNPOD_AUDIO_SFX_CANARY_PERCENTAGERUNPOD_AUDIO_VOICE_CANARY_ENDPOINT_IDRUNPOD_AUDIO_VOICE_CANARY_PERCENTAGERUNPOD_AUDIO_MUSIC_CANARY_ENDPOINT_IDRUNPOD_AUDIO_MUSIC_CANARY_PERCENTAGE
- Expanded generation-api provider compatibility endpoint-family matrix with
audio-sfx,audio-voice, andaudio-musicsupport declarations. - Expanded RunPod operations and provider routing diagnostics payloads to expose audio endpoint IDs and canary state.
- Expanded SDK/client type contracts to include new endpoint families.
Web Exposure#
- Updated RunPod operations UI to render canary status lines for all endpoint families, including audio classes.
- Updated feature flag diagnostics UI to surface endpoint canary state for the new audio classes.
Verification#
python3 scripts/isis/validate_runpod_desired_state.py --file infra/runpod/endpoints/desired-state.jsonpnpm --filter @oshun/config exec vitest run src/features.spec.ts --reporter=dotpnpm --filter @isis/generation-api exec vitest run src/routes/jobs.spec.ts --reporter=dotpnpm --filter @isis/web exec vitest run src/pages/RunPodPage.spec.tsx src/pages/FeatureFlagsPage.spec.tsx src/pages/RunPodParityPage.spec.tsx --reporter=dot
Outcome#
Audio-heavy workflows now have dedicated RunPod endpoint classes with explicit stable/canary routing controls and end-to-end visibility in API, SDK, and web operations surfaces.