Replace the single fixed /records path with a list of storages and an
active selection (where new recordings go). Quota is now per-storage.
- config: StorageItem(id,name,type,path,quota_gb,nas-meta) + Storage keeps
global retention_days/segment_seconds; active + items. Migration from the
old single {path,quota_gb} schema → one local item.
- recorder: write to the active storage; restart_all() repoints on switch;
NAS-not-mounted guard (never write into the container overlay).
- indexer: scan ALL storage roots so recordings survive an active switch.
- retention: enforce quota PER storage via db.total_size_under /
oldest_recordings_under (path-prefix LIKE).
- api/storages: GET/POST/PUT/DELETE + activate + policy. NAS uses host-mount
(backend stores SMB params, generates the mount/fstab command, reports
ismount status). Retire old POST /api/storage.
- frontend: new "Хранилища" settings tab — table with inline per-row quota,
add form (local folder / NAS SMB), global policy (segment + retention).
- compose: add shared ./storages:/storages root for extra local folders and
host-mounted NAS shares.
VERSION 0.0.127.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live and recording both read from the go2rtc restream. If go2rtc's
runtime stream table gets emptied (camera clear via API, go2rtc restart
with a stale/empty config, or a silently-failed startup sync_all), every
restream returns 404 and both live and recording go dark until a manual
`docker restart nvr-go2rtc`.
Add a background Reconciler that every 30s diffs the desired streams
against go2rtc /api/streams and PUTs only the missing ones (existing
streams are left untouched so active producers/consumers are not
disrupted). VERSION -> 0.0.115.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
git_admin