@tiens.nguyen/gonext-local-worker 1.0.45 → 1.0.47

package/gonext-local-worker.mjs

@@ -7,7 +7,7 @@
  * - `gonext-local-worker simulate-chat [text]` — claim next chat job, push fake reply like the real worker (needs GONEXT_* env)
  * - `gonext-local-worker` — starts polling loop (claims jobs and runs models)
  */
-import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { mkdir, mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
 import { execFile as execFileCallback } from "node:child_process";
 import { createHash } from "node:crypto";
 import { homedir, platform, tmpdir } from "node:os";
@@ -814,6 +814,26 @@ function sourceLabelFromBase(base) {
   }
 }
 
+function expandHomePath(rawPath) {
+  const v = String(rawPath ?? "").trim();
+  if (!v) return "";
+  return v.replace(/^~(?=\/)/, homedir());
+}
+
+async function checkModelPathExists(rawPath) {
+  const inputPath = String(rawPath ?? "").trim();
+  const resolvedPath = expandHomePath(inputPath);
+  if (!resolvedPath) {
+    return { path: inputPath, resolvedPath: "", exists: false };
+  }
+  try {
+    const st = await stat(resolvedPath);
+    return { path: inputPath, resolvedPath, exists: st.isDirectory() || st.isFile() };
+  } catch {
+    return { path: inputPath, resolvedPath, exists: false };
+  }
+}
+
 /** Log assistant text to stdout; cap size so huge replies do not flood the terminal. */
 function logModelResponseToWorker(jobId, modelId, text) {
   const max = 12000;
@@ -855,20 +875,26 @@ async function checkOpenAiModels(base, apiKey, source) {
     typeof source?.workerHostId === "string" ? source.workerHostId.trim() : "";
   const hostSourceLabel =
     typeof source?.hostName === "string" ? source.hostName.trim() : "";
+  const sourcePort = typeof source?.port === "number" ? source.port : null;
   try {
     const res = await fetch(endpoint, { method: "GET", headers });
     if (!res.ok) return { online: false, endpoint, models: [] };
     const j = await res.json();
-    const modelSuffix = hostSourceId ? `@@host:${hostSourceId}` : "";
-    const displaySuffix = hostSourceLabel ? ` (${hostSourceLabel})` : "";
-    const models = (j.data ?? [])
-      .map((d) => d.id)
-      .filter(Boolean)
-      .map((id) => ({
-        id: hostSourceId ? `${id}@@${hostSourceId}` : id,
-        name: `${id}${displaySuffix}`,
-        value: `mlx:${id}${modelSuffix}`,
-      }));
+    const hostSuffix = hostSourceId ? `@@host:${hostSourceId}` : "";
+    const portSuffix = sourcePort ? `@@port:${sourcePort}` : "";
+    const modelSuffix = `${hostSuffix}${portSuffix}`;
+    const portLabel = sourcePort ? `:${sourcePort}` : "";
+    const displaySuffix = hostSourceLabel ? ` (${hostSourceLabel}${portLabel})` : portLabel ? ` (${portLabel})` : "";
+    let rawIds = (j.data ?? []).map((d) => d.id).filter(Boolean);
+    // mlx_lm.server registers both the local path and the HF model ID.
+    // Prefer local absolute paths so each server shows exactly one model entry.
+    const localPaths = rawIds.filter((id) => id.startsWith("/") || id.startsWith("~"));
+    if (localPaths.length > 0) rawIds = localPaths;
+    const models = rawIds.map((id) => ({
+      id: hostSourceId ? `${id}@@${hostSourceId}` : id,
+      name: `${id}${displaySuffix}`,
+      value: `mlx:${id}${modelSuffix}`,
+    }));
     return { online: true, endpoint, models };
   } catch {
     return { online: false, endpoint, models: [] };
@@ -925,8 +951,11 @@ async function runLocalHealthJob(job) {
   const ollamaPayloadCount = Array.isArray(payload?.ollamaBaseUrls)
     ? payload.ollamaBaseUrls.length
     : 0;
+  const mlxUrlCount = Array.isArray(payload?.mlxOpenAiBaseUrls)
+    ? payload.mlxOpenAiBaseUrls.length
+    : payload?.mlxOpenAiBaseUrl ? 1 : 0;
   console.log(
-    `[gonext-worker] local_health ${jobId} start (ollamaUrls=${ollamaPayloadCount}, mlx=${payload?.mlxOpenAiBaseUrl ? "yes" : "no"})`
+    `[gonext-worker] local_health ${jobId} start (ollamaUrls=${ollamaPayloadCount}, mlxUrls=${mlxUrlCount})`
   );
   const runRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
     method: "PATCH",
@@ -960,7 +989,13 @@ async function runLocalHealthJob(job) {
         if (!dedup.has(m.value)) dedup.set(m.value, m);
       }
     }
-    const mlxRoot = normalizeOpenAiV1Root(payload?.mlxOpenAiBaseUrl);
+    // Support both single mlxOpenAiBaseUrl and multi-port mlxOpenAiBaseUrls
+    const mlxRootSingle = normalizeOpenAiV1Root(payload?.mlxOpenAiBaseUrl);
+    const mlxRootsMulti = Array.isArray(payload?.mlxOpenAiBaseUrls)
+      ? payload.mlxOpenAiBaseUrls.map(normalizeOpenAiV1Root).filter(Boolean)
+      : [];
+    const mlxRoots = mlxRootsMulti.length > 0 ? mlxRootsMulti : (mlxRootSingle ? [mlxRootSingle] : []);
+    const mlxRoot = mlxRoots[0] || null;
     const targetWorkerHostId =
       typeof payload?.targetWorkerHostId === "string"
         ? payload.targetWorkerHostId.trim()
@@ -969,20 +1004,52 @@ async function runLocalHealthJob(job) {
       typeof payload?.targetWorkerHostName === "string"
         ? payload.targetWorkerHostName.trim()
         : "";
-    let mlxHttp = null;
     let mlxNative = null;
+    const modelPathChecksRaw = Array.isArray(payload?.modelPathChecks)
+      ? payload.modelPathChecks
+      : [];
+    const modelPathChecks = [...new Set(modelPathChecksRaw)]
+      .filter((v) => typeof v === "string" && v.trim())
+      .map((v) => String(v).trim())
+      .slice(0, 8);
+    const modelPathStatus = [];
 
-    if (mlxRoot) {
+    // Collect results from all MLX endpoints (one per port)
+    const mlxHttpResults = [];
+    for (const root of mlxRoots) {
+      // Extract port from URL for model value encoding
+      let port = null;
+      try {
+        const u = new URL(root.replace(/\/v1\/?$/i, ""));
+        const p = parseInt(u.port, 10);
+        if (p > 0) port = p;
+      } catch { /* ignore */ }
       const mlxStart = Date.now();
-      console.log(`[gonext-worker] local_health ${jobId} check mlx HTTP ${mlxRoot}`);
-      mlxHttp = await checkOpenAiModels(mlxRoot, payload?.mlxApiKey ?? "", {
+      console.log(`[gonext-worker] local_health ${jobId} check mlx HTTP ${root} port=${port ?? "default"}`);
+      const result = await checkOpenAiModels(root, payload?.mlxApiKey ?? "", {
         workerHostId: targetWorkerHostId,
-        hostName: targetWorkerHostName || sourceLabelFromBase(mlxRoot),
+        hostName: targetWorkerHostName || sourceLabelFromBase(root),
+        port,
       });
       console.log(
-        `[gonext-worker] local_health ${jobId} mlx HTTP online=${mlxHttp.online} models=${mlxHttp.models.length} took=${((Date.now() - mlxStart) / 1000).toFixed(2)}s`
+        `[gonext-worker] local_health ${jobId} mlx HTTP online=${result.online} models=${result.models.length} took=${((Date.now() - mlxStart) / 1000).toFixed(2)}s`
       );
+      mlxHttpResults.push(result);
     }
+    // Merge all per-port results
+    const mlxModelDedup = new Map();
+    let mlxHttpOnline = false;
+    for (const r of mlxHttpResults) {
+      if (r.online) mlxHttpOnline = true;
+      for (const m of r.models) {
+        if (!mlxModelDedup.has(m.value)) mlxModelDedup.set(m.value, m);
+      }
+    }
+    const mlxHttp = mlxRoots.length > 0 ? {
+      online: mlxHttpOnline,
+      endpoint: mlxHttpResults[0]?.endpoint,
+      models: [...mlxModelDedup.values()],
+    } : null;
 
     const wantNativeFallback =
       mlxRoot &&
@@ -1001,6 +1068,11 @@ async function runLocalHealthJob(job) {
       );
     }
 
+    for (const rawPath of modelPathChecks) {
+      const check = await checkModelPathExists(rawPath);
+      modelPathStatus.push(check);
+    }
+
     let mlx = null;
     if (mlxRoot || mlxNative?.available) {
       const httpOk = Boolean(mlxHttp?.online && (mlxHttp?.models?.length ?? 0) > 0);
@@ -1055,6 +1127,12 @@ async function runLocalHealthJob(job) {
             }
           : undefined,
       mlx,
+      setup:
+        modelPathStatus.length > 0
+          ? {
+              modelPaths: modelPathStatus,
+            }
+          : undefined,
     };
     const totalTimeSeconds = (Date.now() - start) / 1000;
     const doneRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
@@ -1133,7 +1211,9 @@ async function pollOnce() {
     const isLocalHealthByType = job.jobType === "local_health";
     const isLocalHealthByModelKey = job.modelKey === "local_health";
     const isLocalHealthByPayload =
-      Array.isArray(job.payload?.ollamaBaseUrls) || !!job.payload?.mlxOpenAiBaseUrl;
+      Array.isArray(job.payload?.ollamaBaseUrls) ||
+      !!job.payload?.mlxOpenAiBaseUrl ||
+      Array.isArray(job.payload?.mlxOpenAiBaseUrls);
     if (isLocalHealthByType || isLocalHealthByModelKey || isLocalHealthByPayload) {
       const task = runLocalHealthJob(job).catch((e) => {
         console.error("[gonext-worker] local_health task error:", e);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.45",
+  "version": "1.0.47",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",