UHD_Media_Tool/apps/server/src/modules/file-manager/probe.ts

import { createHash, randomUUID } from "node:crypto";
import { mkdir, readFile, stat, writeFile } from "node:fs/promises";
import { join, resolve } from "node:path";
import type { ProbeResult } from "@umt/adapters";
import type {
  FileProbeItem,
  FilesProbeRequest,
  FilesProbeResult,
  FilesScrapeBatchRequest,
  FilesScrapeBatchResult,
  FilesScrapeRequest,
  FilesScrapeResult,
} from "@umt/contracts";
import { sql } from "@umt/db";
import type { Principal } from "../../http/auth.ts";
import { FilesError, type FmDeps, jsonb, resolveIn } from "./common.ts";
import { audit } from "./ledger.ts";
import { kindOfName, mediaSummary } from "./list.ts";
import { entitiesUnder } from "./records.ts";

// probe:ffprobe 摘要;已登记资产写回 media_asset.media_info.probe,未登记只缓存 24h(DATA_DIR/cache/probe)。
// scrape:只对已登记作品,等价 POST /media-entities/{id}/refresh(入队 refreshMetadata,字段锁优先、不动路径);
// NFO/图片投影 W5 实现(TODO),请求时返回 not_implemented。

const CACHE_TTL_MS = 24 * 3600_000;

export async function probeFiles(d: FmDeps, body: FilesProbeRequest): Promise<FilesProbeResult> {
  const items: FileProbeItem[] = [];
  const cacheDir = join(resolve(d.options.dataDir), "cache", "probe");
  for (const it of body.items) {
    const r = await resolveIn(d, it.root, it.path);
    const base = { root: it.root as FileProbeItem["root"], path: r.rel };
    if (!r.lst?.isFile() && !r.isSymlink) {
      items.push({
        ...base,
        status: "unknown",
        reason: "not_a_file",
        summary: null,
        container: null,
        subtitleTracks: 0,
        savedToAsset: false,
        cached: false,
      });
      continue;
    }
    if (kindOfName(r.abs) !== "video") {
      items.push({
        ...base,
        status: "unknown",
        reason: "not_video",
        summary: null,
        container: null,
        subtitleTracks: 0,
        savedToAsset: false,
        cached: false,
      });
      continue;
    }
    const st = await stat(r.abs);
    const key = createHash("sha1").update(`${r.abs}|${st.size}|${st.mtimeMs}`).digest("hex");
    const cacheFile = join(cacheDir, `${key}.json`);
    let result: ProbeResult | null = null;
    let cached = false;
    try {
      const c = JSON.parse(await readFile(cacheFile, "utf8")) as { at: number; result: ProbeResult };
      if (Date.now() - c.at < CACHE_TTL_MS) {
        result = c.result;
        cached = true;
      }
    } catch {
      // 无缓存
    }
    result ??= await d.ao.ports.probe(r.abs);
    let saved = false;
    if (result.status === "ok" && r.def.libraryRootId) {
      const upd = await sql<{ id: string }>`
        update media_asset a set media_info = a.media_info || jsonb_build_object('probe', ${jsonb(result)}::jsonb)
          from asset_location l
         where l.media_asset_id = a.id and l.library_root_id = ${r.def.libraryRootId}
           and l.relative_path = ${r.rel} and l.status <> 'removed'
        returning a.id`.execute(d.ctx.db);
      saved = upd.rows.length > 0;
    }
    if (result.status === "ok" && !saved && !cached) {
      await mkdir(cacheDir, { recursive: true }).catch(() => undefined);
      const tmp = `${cacheFile}.${randomUUID()}.tmp`;
      await writeFile(tmp, JSON.stringify({ at: Date.now(), result }))
        .then(() => Deno.rename(tmp, cacheFile))
        .catch(() => undefined);
    }
    items.push({
      ...base,
      status: result.status,
      reason: result.status === "ok" ? null : result.reason,
      summary: result.status === "ok" ? mediaSummary({ probe: result }) : null,
      container: result.status === "ok" ? (result.container ?? null) : null,
      subtitleTracks: result.status === "ok" ? result.subtitles.length : 0,
      savedToAsset: saved,
      cached,
    });
  }
  return { items };
}

export async function scrape(
  d: FmDeps,
  principal: Principal | undefined,
  body: FilesScrapeRequest,
): Promise<FilesScrapeResult> {
  const db = d.ctx.db;
  const e = await db
    .selectFrom("media_entity")
    .select(["id", "status", "field_locks"])
    .where("id", "=", body.mediaEntityId)
    .executeTakeFirst();
  if (!e) throw new FilesError("not_found", "作品不存在");
  const registered = await db
    .selectFrom("media_asset")
    .select(["id"])
    .where("media_entity_id", "=", e.id)
    .where("status", "=", "active")
    .executeTakeFirst();
  if (!registered) throw new FilesError("bad_request", "作品没有已登记资产。先识别并整理入库,再刮削");
  const { runId } = await d.ctx.engine.enqueue(
    "refreshMetadata",
    { mediaEntityId: e.id },
    { idempotencyKey: `refreshMetadata:files:${body.idempotencyKey}`, traceId: randomUUID() },
  );
  const overrides = await db
    .selectFrom("field_override")
    .select(["field"])
    .where("media_entity_id", "=", e.id)
    .where("locked", "=", true)
    .execute();
  const lockedFields = [...new Set([...(e.field_locks ?? []), ...overrides.map((o) => o.field)])].sort();
  // TODO(W5):writeNfo/images → 把最终字段投影成 NFO/海报写到绑定目录(临时文件 + 原子提交;非本系统 NFO 不覆盖)
  const result: FilesScrapeResult = {
    mediaEntityId: e.id,
    taskId: runId,
    lockedFields,
    nfo: body.writeNfo ? "not_implemented" : "not_requested",
    images: body.images ? "not_implemented" : "not_requested",
  };
  await audit(d, principal, "scrape", { ...result });
  return result;
}

/**
 * G5 批量刷新元数据:文件/目录归到其下已登记资产的作品,与显式作品合并去重,逐个入队 refreshMetadata。
 * 只对有 active 资产的作品提交;其余进 skipped(前端汇总"已提交 n 个作品,锁定字段不变")。
 */
export async function scrapeBatch(
  d: FmDeps,
  principal: Principal | undefined,
  body: FilesScrapeBatchRequest,
): Promise<FilesScrapeBatchResult> {
  const db = d.ctx.db;
  const skipped: FilesScrapeBatchResult["skipped"] = [];
  const ids = new Set(body.mediaEntityIds ?? []);
  for (const it of body.items ?? []) {
    const r = await resolveIn(d, it.root, it.path);
    const found = r.def.libraryRootId
      ? await entitiesUnder(db, r.def.libraryRootId, r.rel, !!r.lst?.isDirectory())
      : [];
    if (found.length === 0) skipped.push({ ref: `${it.root}:${it.path}`, reason: "no_entity" });
    for (const id of found) ids.add(id);
  }
  if (ids.size === 0 && skipped.length === 0) throw new FilesError("bad_request", "需要 items 或 mediaEntityIds");
  if (ids.size > 200) throw new FilesError("bad_request", `一次最多刷新 200 个作品(本次 ${ids.size})`);
  const list = [...ids];
  const ents = list.length
    ? await db.selectFrom("media_entity").select(["id", "title", "field_locks"]).where("id", "in", list).execute()
    : [];
  const active = new Set(
    list.length
      ? (
          await db
            .selectFrom("media_asset")
            .select(["media_entity_id"])
            .where("media_entity_id", "in", list)
            .where("status", "=", "active")
            .execute()
        ).map((a) => a.media_entity_id)
      : [],
  );
  const locks = list.length
    ? await db
        .selectFrom("field_override")
        .select(["media_entity_id", "field"])
        .where("media_entity_id", "in", list)
        .where("locked", "=", true)
        .execute()
    : [];
  const submitted: FilesScrapeBatchResult["submitted"] = [];
  for (const id of list) {
    const e = ents.find((x) => x.id === id);
    if (!e) {
      skipped.push({ ref: id, reason: "not_found" });
      continue;
    }
    if (!active.has(id)) {
      skipped.push({ ref: id, reason: "not_registered" });
      continue;
    }
    const { runId } = await d.ctx.engine.enqueue(
      "refreshMetadata",
      { mediaEntityId: id },
      { idempotencyKey: `refreshMetadata:files-batch:${body.idempotencyKey}:${id}`, traceId: randomUUID() },
    );
    const lockedFields = [
      ...new Set([...(e.field_locks ?? []), ...locks.filter((l) => l.media_entity_id === id).map((l) => l.field)]),
    ].sort();
    submitted.push({ mediaEntityId: id, title: e.title, taskId: runId, lockedFields });
  }
  await audit(d, principal, "scrape_batch", {
    submitted: submitted.map((s) => s.mediaEntityId),
    skipped,
  });
  return { submitted, skipped };
}