@armadra/agent 0.7.5 → 0.8.0

package/dist/ai/providers/max-tokens-cache.js

@@ -0,0 +1,95 @@
+/**
+ * `max_tokens` 上限的跨进程缓存：`<dataDir>/models/max-tokens-caps.json`（#152）。
+ *
+ * 被动修正（apis/max-tokens.ts）学到的 `provider/model` 上限原本只在进程内，重启后每个模型的首个请求仍被
+ * 400 拒一次。这里把它写进数据目录：组装注册表时载回进程内的表（进程内已有的键优先），学到新上限就把文件里
+ * 未过期的条目加上这一条原子写回（临时文件 + 改名；并发进程后写者赢，丢的一条下次再学）。条目 30 天过期，
+ * 过期后重新探测（中转放宽上限时不会一直压低输出）；删掉文件即重测。只存模型引用、上限与时间，没有凭据。
+ */
+import { mkdirSync, readFileSync, renameSync, writeFileSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { maxTokensCaps, onMaxTokensCap } from "../apis/max-tokens.js";
+export const MAX_TOKENS_CACHE_VERSION = 1;
+/** 条目有效期：30 天。 */
+export const MAX_TOKENS_CACHE_TTL_MS = 30 * 24 * 60 * 60 * 1000;
+/** 缓存文件位置。 */
+export function maxTokensCachePath(dataDir) {
+    return join(dataDir, "models", "max-tokens-caps.json");
+}
+/** 读文件里未过期的条目；不存在、坏 JSON、版本不符或形状不对的条目一律忽略。 */
+function readEntries(dataDir, now) {
+    const out = new Map();
+    let parsed;
+    try {
+        parsed = JSON.parse(readFileSync(maxTokensCachePath(dataDir), "utf8"));
+    }
+    catch {
+        return out;
+    }
+    if (typeof parsed !== "object" || parsed === null)
+        return out;
+    const file = parsed;
+    if (file.version !== MAX_TOKENS_CACHE_VERSION)
+        return out;
+    if (typeof file.caps !== "object" || file.caps === null)
+        return out;
+    for (const [key, value] of Object.entries(file.caps)) {
+        if (typeof value !== "object" || value === null)
+            continue;
+        const { cap, learnedAt } = value;
+        if (typeof cap !== "number" || !Number.isInteger(cap) || cap < 1)
+            continue;
+        if (typeof learnedAt !== "number" || !Number.isFinite(learnedAt))
+            continue;
+        if (now - learnedAt > MAX_TOKENS_CACHE_TTL_MS)
+            continue;
+        out.set(key, { cap, learnedAt });
+    }
+    return out;
+}
+/** 载入未过期的上限到进程内的表（已有的键不覆盖）；返回载入条数。 */
+export function loadMaxTokensCaps(dataDir, now = Date.now) {
+    let loaded = 0;
+    for (const [key, entry] of readEntries(dataDir, now())) {
+        if (maxTokensCaps.has(key))
+            continue;
+        maxTokensCaps.set(key, entry.cap);
+        loaded++;
+    }
+    return loaded;
+}
+/** 把一条新上限并入文件（顺带清掉过期条目）；写失败不抛（缓存只是省一次 400）。 */
+export function recordMaxTokensCap(dataDir, key, cap, now = Date.now) {
+    const at = now();
+    const entries = readEntries(dataDir, at);
+    entries.set(key, { cap, learnedAt: at });
+    const path = maxTokensCachePath(dataDir);
+    const tmp = `${path}.${process.pid}.tmp`;
+    try {
+        mkdirSync(dirname(path), { recursive: true, mode: 0o700 });
+        const caps = Object.fromEntries([...entries].sort(([a], [b]) => (a < b ? -1 : 1)));
+        writeFileSync(tmp, `${JSON.stringify({ version: MAX_TOKENS_CACHE_VERSION, caps }, null, 2)}\n`);
+        renameSync(tmp, path);
+    }
+    catch {
+        // 只读数据目录等：本进程内的表仍有效
+    }
+}
+const attached = new Map();
+/**
+ * 载入并订阅：之后学到的上限写回该数据目录。按 `dataDir` 幂等（同一进程重复组装注册表不重复订阅）；
+ * 返回取消函数（取消后可重新 attach）。
+ */
+export function attachMaxTokensCache(dataDir, now = Date.now) {
+    const existing = attached.get(dataDir);
+    if (existing !== undefined)
+        return existing;
+    loadMaxTokensCaps(dataDir, now);
+    const unsubscribe = onMaxTokensCap((key, cap) => recordMaxTokensCap(dataDir, key, cap, now));
+    const detach = () => {
+        unsubscribe();
+        attached.delete(dataDir);
+    };
+    attached.set(dataDir, detach);
+    return detach;
+}

package/dist/ai/providers/registry.js

@@ -256,9 +256,9 @@ export class ProviderRegistry {
         const channelsOf = (wanted, modelId) => modelChannels(wanted, provider.channels ?? [], provider.defaultChannel ?? "", (name) => this.warn(`model "${id}/${modelId}" channel "${name}" not found; ignored`));
         for (const entry of config.models ?? []) {
             // 模型级 api（第三波 §2.3）：同一中转下不同模型走不同协议；缺省沿用供应商（或渠道）的。
-            const { entry: filled, metadata } = enrichEntry(entry, this.modelsDev);
-            if (typeof entry.modelsDev === "string" && metadata.looked && metadata.match === undefined)
-                this.warn(`modelsDev "${entry.modelsDev}" for "${id}/${entry.id}" not found`);
+            const { entry: filled, metadata } = enrichEntry(entry, this.modelsDev, id);
+            for (const warning of metadata.warnings ?? [])
+                this.warn(warning);
             const model = withCustomDefaults(filled, id, entry.api ?? provider.api);
             // 内置供应商上没写 channels 的模型：定渠道时挂全部渠道（settleBuiltinChannels）
             if (provider.channels !== undefined && (entry.channels !== undefined || !provider.builtin))