@tiens.nguyen/gonext-local-worker 1.0.29 → 1.0.31

package/gonext-local-worker.mjs

@@ -3,8 +3,9 @@
  * GoNext local worker:
  * - `gonext-local-worker set <workerKey> [--api-base URL] [--poll-ms 1500]`
  *   writes ~/.gonext/worker.env
- * - `gonext-local-worker` starts polling loop
- */
+ * - `gonext-local-worker ws-ping-test` — POST worker ws-ping (needs GONEXT_* env)
+ * - `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, readFile, writeFile } from "node:fs/promises";
 import { execFile as execFileCallback } from "node:child_process";
 import { homedir, platform } from "node:os";
@@ -30,13 +31,16 @@ Usage:
   gonext-local-worker
   gonext-local-worker set <workerKey> [--api-base <url>] [--poll-ms <ms>]
   gonext-local-worker ws-ping-test   # POST /api/worker/ws-ping-test (needs GONEXT_* env)
+  gonext-local-worker simulate-chat [text...]   # claim next chat job, fake reply (stop normal worker first)
 
 Examples:
   gonext-local-worker set abc123 --api-base https://hwohu56e8d.execute-api.ap-southeast-1.amazonaws.com
   gonext-local-worker
   gonext-local-worker ws-ping-test
+  gonext-local-worker simulate-chat "Hello from a fake model"
 
 Env (optional):
+  GONEXT_SIMULATE_TEXT   default body for simulate-chat when no args
   GONEXT_MLX_LM_PYTHON   Python executable for MLX LM native probe (default: python3)
 `);
 }
@@ -96,6 +100,18 @@ const apiBase = (process.env.GONEXT_API_BASE ?? "").replace(/\/+$/, "");
 const workerKey = process.env.GONEXT_WORKER_KEY ?? "";
 const pollMs = Number(process.env.GONEXT_POLL_MS ?? "1500") || 1500;
 
+const CHUNK_PATH = "/api/worker/job-chunk";
+
+async function workerFetch(path, init = {}) {
+  const url = `${apiBase}${path.startsWith("/") ? path : `/${path}`}`;
+  const headers = {
+    "Content-Type": "application/json",
+    "X-Worker-Key": workerKey,
+    ...(init.headers ?? {}),
+  };
+  return fetch(url, { ...init, headers });
+}
+
 if (args[0] === "ws-ping-test") {
   if (!apiBase || !workerKey) {
     console.error(
@@ -115,12 +131,116 @@ if (args[0] === "ws-ping-test") {
   const text = await res.text();
   if (!res.ok) {
     console.error(`[gonext-worker] ws-ping-test failed ${res.status}: ${text}`);
+    if (res.status === 404 && text.includes("Cannot POST")) {
+      console.error(
+        "[gonext-worker] DEPLOY: this API does not have POST /api/worker/ws-ping-test yet. From repo root: sam build && sam deploy (or ship updated server-bundle to your Node host)."
+      );
+    }
     process.exit(1);
   }
   console.log("[gonext-worker] ws-ping-test OK:", text);
   process.exit(0);
 }
 
+if (args[0] === "simulate-chat") {
+  if (!apiBase || !workerKey) {
+    console.error(
+      "simulate-chat needs GONEXT_API_BASE and GONEXT_WORKER_KEY (run: gonext-local-worker set <key> --api-base <url>)"
+    );
+    process.exit(1);
+  }
+  const reply =
+    args.slice(1).join(" ").trim() ||
+    String(process.env.GONEXT_SIMULATE_TEXT ?? "").trim() ||
+    "Simulated model reply: hello from gonext-local-worker simulate-chat.";
+  const jobRes = await workerFetch("/api/worker/jobs/next", { method: "POST" });
+  if (jobRes.status === 204) {
+    console.error(
+      "[gonext-worker] simulate-chat: no pending job. In the web app, send a message that returns 202 (async local chat), then run this again. Stop the normal worker so it does not claim the job first."
+    );
+    process.exit(1);
+  }
+  if (!jobRes.ok) {
+    const t = await jobRes.text().catch(() => "");
+    console.error(`[gonext-worker] simulate-chat: jobs/next failed ${jobRes.status}: ${t}`);
+    process.exit(1);
+  }
+  const job = await jobRes.json();
+  const jobId = job?.jobId;
+  if (!jobId) {
+    console.error("[gonext-worker] simulate-chat: invalid jobs/next response (missing jobId).");
+    process.exit(1);
+  }
+  const isLocalHealth =
+    job.jobType === "local_health" || job.modelKey === "local_health";
+  if (isLocalHealth) {
+    await workerFetch(`/api/worker/jobs/${jobId}`, {
+      method: "PATCH",
+      body: JSON.stringify({
+        jobStatus: "failed",
+        errorMessage:
+          "simulate-chat: claimed a local_health job. Mark failed so you can retry. Queue a chat message (not Settings refresh) and stop the normal worker before simulate-chat.",
+        totalTimeSeconds: 0,
+      }),
+    });
+    console.error(
+      "[gonext-worker] simulate-chat: claimed local_health instead of chat. Job marked failed. Retry with only a pending chat job."
+    );
+    process.exit(1);
+  }
+  const start = Date.now();
+  const runRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
+    method: "PATCH",
+    body: JSON.stringify({ jobStatus: "running" }),
+  });
+  if (!runRes.ok) {
+    const errBody = await runRes.text().catch(() => "");
+    console.error(
+      `[gonext-worker] simulate-chat: mark running failed ${runRes.status}${errBody ? `: ${errBody}` : ""}`
+    );
+    process.exit(1);
+  }
+  const sliceSize = 160;
+  for (let i = 0; i < reply.length; i += sliceSize) {
+    const slice = reply.slice(i, i + sliceSize);
+    if (!slice) continue;
+    const chunkRes = await workerFetch(CHUNK_PATH, {
+      method: "POST",
+      body: JSON.stringify({ jobId, text: slice }),
+    });
+    if (!chunkRes.ok && chunkRes.status !== 204) {
+      const snippet = (await chunkRes.text().catch(() => "")).trim().slice(0, 400);
+      console.error(
+        `[gonext-worker] simulate-chat: job-chunk failed ${chunkRes.status} jobId=${jobId}` +
+          (snippet ? ` response=${snippet}` : "")
+      );
+      process.exit(1);
+    }
+  }
+  const totalTimeSeconds = (Date.now() - start) / 1000;
+  const doneRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
+    method: "PATCH",
+    body: JSON.stringify({
+      jobStatus: "completed",
+      resultText: reply,
+      tokenCount: Math.max(1, Math.ceil(reply.length / 4)),
+      totalTimeSeconds,
+    }),
+  });
+  if (!doneRes.ok) {
+    const snippet = (await doneRes.text().catch(() => "")).trim().slice(0, 400);
+    console.error(
+      `[gonext-worker] simulate-chat: complete PATCH failed ${doneRes.status}` +
+        (snippet ? `: ${snippet}` : "")
+    );
+    process.exit(1);
+  }
+  console.log(
+    `[gonext-worker] simulate-chat OK jobId=${jobId} (${totalTimeSeconds.toFixed(2)}s, ${reply.length} chars) — same HTTP path as the real worker (API pushes worker_job_chunk / worker_job to the web app).`
+  );
+  process.exit(0);
+}
+
 if (!apiBase || !workerKey) {
   console.error(
     "Set GONEXT_API_BASE (HTTP API origin, no /api suffix) and GONEXT_WORKER_KEY."
@@ -128,7 +248,6 @@ if (!apiBase || !workerKey) {
   process.exit(1);
 }
 
-const CHUNK_PATH = "/api/worker/job-chunk";
 console.log(
   `[gonext-worker] API base ${apiBase} — streaming chunks POST ${apiBase}${CHUNK_PATH}`
 );
@@ -151,16 +270,6 @@ function toOpenAIMessages(messages) {
   });
 }
 
-async function workerFetch(path, init = {}) {
-  const url = `${apiBase}${path.startsWith("/") ? path : `/${path}`}`;
-  const headers = {
-    "Content-Type": "application/json",
-    "X-Worker-Key": workerKey,
-    ...(init.headers ?? {}),
-  };
-  return fetch(url, { ...init, headers });
-}
-
 async function runChatJob(job) {
   const { jobId, payload } = job;
   if (!payload || !Array.isArray(payload.messages)) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.29",
+  "version": "1.0.31",
   "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",