Zac Gaetano
08499b93b2
#164 HLS preview uses h264_nvenc (forced-IDR, GOP=segment) when the sidecar has the GPU, else keeps libx264 fallback. #168 source-backend abstraction in capture-manager (blackmagic implemented as a behavior-preserving refactor; deltacast/aja stubbed pending hardware). #167 per-recorder gpu_uuid (migration 032) plumbed mam-api->agent-> NVIDIA_VISIBLE_DEVICES (defaults to 'all'). #166 node-agent reports encoder util + NVENC session count per GPU; Cluster screen renders per-GPU GPU/ENC util, VRAM, sessions. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
10 lines
546 B
SQL
10 lines
546 B
SQL
-- Migration 032: Per-recorder GPU affinity (Issue #167)
|
|
-- Adds a nullable GPU UUID to the recorders table so each recorder can be
|
|
-- pinned to a specific GPU on its node. The value is passed through to the
|
|
-- node-agent sidecar-start payload and becomes NVIDIA_VISIBLE_DEVICES for the
|
|
-- capture container. NULL = legacy behavior (NVIDIA_VISIBLE_DEVICES=all, i.e.
|
|
-- every GPU visible). Accepts an nvidia-smi GPU UUID (e.g. "GPU-xxxx") or a
|
|
-- numeric index string.
|
|
|
|
ALTER TABLE recorders
|
|
ADD COLUMN IF NOT EXISTS gpu_uuid TEXT DEFAULT NULL;
|