@genroc/eval-node 0.0.0-edge.4348678 → 0.0.0-edge.46dbe379

package/dist/worker.js

@@ -4,7 +4,8 @@
 // nothing about the queue, which is what keeps the containment strategy swappable.
 //
 // See README.md for the contract, and specs/external-task-queue.md for the queue itself.
-import { evaluate } from "./eval.js";
+import { readFileSync } from "node:fs";
+import { Cancelled, evaluate } from "./eval.js";
 const SERVER = (process.env.GENROC_SERVER ?? "http://localhost:8448").replace(/\/$/, "");
 const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
 // The credential, when the server runs with --auth token. A worker needs exactly the `worker`
@@ -15,7 +16,11 @@ const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
 // Sent as a header rather than in the URL because Node's fetch REFUSES a URL carrying
 // credentials ("Request cannot be constructed from a URL that includes credentials"), so the
 // basic-auth-in-the-URL trick that works for genctl is not available here.
-const TOKEN = process.env.GENROC_TOKEN ?? "";
+// GENROC_TOKEN_FILE is the mounted-secret shape: a credential in a file rather than an
+// environment variable, so it stays out of `docker inspect` and out of the process environment
+// any child inherits. The inline variable wins when both are set.
+const TOKEN = process.env.GENROC_TOKEN ??
+    (process.env.GENROC_TOKEN_FILE ? readFileSync(process.env.GENROC_TOKEN_FILE, "utf8").trim() : "");
 const authHeaders = TOKEN ? { authorization: `Bearer ${TOKEN}` } : {};
 // Concurrency is the worker's to set, and that is the point of pulling: under the old fetch
 // shape genroc decided how many scripts ran at once (--max-concurrent, default 200) and the
@@ -57,15 +62,15 @@ async function fetchObject(ref) {
 /** Put each listed value back where its path says it belongs. Paths are arrays of keys, so this
  *  needs no parser: the whole reason they are not JSON Pointer strings. */
 async function resolveObjects(job) {
-    let input = job.input;
+    let input = job.external_input;
     for (const e of job.objects ?? []) {
         const value = await fetchObject(e.ref);
         if (e.path.length === 0) {
             input = value;
             continue;
         }
-        // The path is rooted at the entry and starts with "input", which is the value being rebuilt.
-        const rest = e.path[0] === "input" ? e.path.slice(1) : e.path;
+        // The path is rooted at the entry and starts with "external_input", the value being rebuilt.
+        const rest = e.path[0] === "external_input" ? e.path.slice(1) : e.path;
         if (rest.length === 0) {
             input = value;
             continue;
@@ -180,7 +185,7 @@ async function answer(token, outcome) {
     console.error(`genroc refused the outcome for ${token}: ${JSON.stringify(data)}`);
     await release(token);
 }
-async function run(job) {
+async function run(job, signal) {
     let resolved;
     try {
         resolved = await resolveObjects(job);
@@ -201,9 +206,17 @@ async function run(job) {
     }
     let result;
     try {
-        result = await evaluate(req);
+        result = await evaluate(req, signal);
     }
     catch (err) {
+        // Cancelled is not a fault: the process was stopped while this ran. The claim still goes
+        // back — the row is terminal, so nothing re-claims it, and the release is what stops it
+        // waiting out a lease nobody is serving.
+        if (err instanceof Cancelled) {
+            console.log(`cancelled: ${job.token}`);
+            await release(job.token);
+            return;
+        }
         // The RUNNER faulted, not the script — the one class where a retry can help. There is no
         // error code for it on purpose: releasing the claim is how a queue spells "retryable", and
         // it puts the task in front of a different worker instead of burning the definition's
@@ -240,11 +253,22 @@ async function renewLoop() {
             tokens,
             lease_ms: LEASE_MS,
         });
-        // A short count means a claim lapsed and was taken over. Nothing to do about it — the work
-        // continues and its answer will be refused — but say so, because it is the signal that
-        // LEASE_MS is too short for what these scripts actually take.
-        if (ok && data?.renewed < tokens.length) {
-            console.error(`renewed ${data.renewed}/${tokens.length} claims; a lease lapsed under load`);
+        if (!ok)
+            continue;
+        // Cancelled: the process was stopped, so the evaluation is abandoned and the claim handed
+        // back. Aborting first is the point of the list — waiting for the script to finish would
+        // keep burning a core on work an operator has already stopped. The release rides run()'s
+        // own catch, so nothing is released twice.
+        for (const token of (data?.cancelled ?? [])) {
+            inFlight.get(token)?.abort.abort();
+        }
+        // Lost: someone else holds this claim now. The work continues and its answer will be
+        // refused — that is not fixable here — but say so, because it is the signal that LEASE_MS
+        // is too short for what these scripts actually take. Deliberately NOT released: the claim
+        // is the new holder's, and releasing would bump the epoch out from under it.
+        const lost = (data?.lost ?? []);
+        if (lost.length) {
+            console.error(`lost ${lost.length}/${tokens.length} claims; a lease lapsed under load`);
         }
     }
 }
@@ -253,8 +277,9 @@ async function pollLoop() {
         const free = CONCURRENCY - inFlight.size;
         const jobs = free > 0 ? await claim(free) : [];
         for (const job of jobs) {
-            inFlight.set(job.token, job);
-            void run(job).finally(() => inFlight.delete(job.token));
+            const abort = new AbortController();
+            inFlight.set(job.token, { job, abort });
+            void run(job, abort.signal).finally(() => inFlight.delete(job.token));
         }
         // Only idle when there was nothing to take: a full queue should be drained at the speed the
         // realms allow, not at the poll interval.

package/eval.ts

@@ -1,6 +1,7 @@
-// Evaluation core: run a code string in its OWN realm and classify every outcome into one of
-// the failure kinds in README.md. Nothing here knows about genroc — worker.ts is the only
-// thing that talks to the queue — so this stays testable and the containment stays swappable.
+// Evaluation core: import a script module in its OWN realm, call its default export, and
+// classify every outcome into one of the failure kinds in README.md. Nothing here knows about
+// genroc — worker.ts is the only thing that talks to the queue — so this stays testable and
+// the containment stays swappable.
 //
 // The containment is a Worker per execution (realm.ts). It is what makes the budget real:
 // a synchronous busy loop never yields, so no in-process timer can interrupt it, and only a
@@ -9,6 +10,7 @@
 import { Worker } from "node:worker_threads";
 
 export type EvalRequest = {
+  /** An ES module whose default export is the function to run; `input` is its argument. */
   code: string;
   input?: unknown;
   timeout_ms?: number;
@@ -55,14 +57,32 @@ class RealmFault extends Error {
   }
 }
 
-export async function evaluate(req: EvalRequest): Promise<EvalResult> {
+/** Thrown when `signal` aborts: the work was cancelled server-side, so there is no outcome to
+ *  report and worker.ts must release rather than answer. Distinct from RealmFault because the
+ *  runner did not fault -- nothing is wrong, the answer is simply no longer wanted. */
+export class Cancelled extends Error {
+  constructor() {
+    super("cancelled");
+    this.name = "Cancelled";
+  }
+}
+
+export async function evaluate(req: EvalRequest, signal?: AbortSignal): Promise<EvalResult> {
   const budget = typeof req.timeout_ms === "number" ? req.timeout_ms : DEFAULT_TIMEOUT_MS;
 
   const worker = new Worker(REALM_URL);
   let timer: ReturnType<typeof setTimeout> | undefined;
+  let onAbort: (() => void) | undefined;
   try {
     return await new Promise<EvalResult>((resolve, reject) => {
       timer = setTimeout(() => resolve(timedOut(budget)), budget);
+      // The abort is wired to the same promise as the budget, so both settle through the one
+      // finally below -- which is what guarantees the thread is gone before either returns.
+      if (signal) {
+        if (signal.aborted) reject(new Cancelled());
+        onAbort = () => reject(new Cancelled());
+        signal.addEventListener("abort", onAbort, { once: true });
+      }
       worker.once("message", (reply: WorkerReply) => resolve(reply));
       // A script may end its own realm (`process.exit()`), which is not a throw and would
       // otherwise present as a hang until the budget expired. Our own terminate() raises
@@ -73,6 +93,7 @@ export async function evaluate(req: EvalRequest): Promise<EvalResult> {
     });
   } finally {
     clearTimeout(timer);
+    if (signal && onAbort) signal.removeEventListener("abort", onAbort);
     // Awaited, and the whole point: on the timeout path a thread is still burning a core, and
     // resolving before it is gone would report an evaluation the machine is still running.
     await worker.terminate();