f69e24eeff
The fake Idunn signer can lose the lease once a job's meta.json exists, which pins the guards in startWorker (no logs or embeddings directory, no worker). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1065 lines
50 KiB
JavaScript
1065 lines
50 KiB
JavaScript
// node --test scripts/test-vili-daemon.mjs
|
|
// Runs the real daemon against scripts/fixtures/stub-worker.mjs as its --python/--worker-entry.
|
|
import assert from "node:assert/strict";
|
|
import { spawn } from "node:child_process";
|
|
import { existsSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, rmSync, statSync, writeFileSync } from "node:fs";
|
|
import http from "node:http";
|
|
import net from "node:net";
|
|
import os from "node:os";
|
|
import path from "node:path";
|
|
import test from "node:test";
|
|
import { fileURLToPath } from "node:url";
|
|
import { decode, encode } from "@msgpack/msgpack";
|
|
|
|
const here = path.dirname(fileURLToPath(import.meta.url));
|
|
const daemonPath = path.join(here, "vili-daemon.mjs");
|
|
const stubPath = path.join(here, "fixtures", "stub-worker.mjs");
|
|
const fakeIdunnEntry = path.join(here, "fixtures", "fake-idunn-entry.mjs");
|
|
const THREADS = 5;
|
|
|
|
async function until(condition, what, seconds = 15) {
|
|
const deadline = Date.now() + seconds * 1000;
|
|
for (;;) {
|
|
const value = await condition();
|
|
if (value) return value;
|
|
assert.ok(Date.now() < deadline, `timed out waiting for ${what}`);
|
|
await new Promise((resolve) => setTimeout(resolve, 25));
|
|
}
|
|
}
|
|
|
|
// A zombie is dead for our purposes: its parent (or init) has simply not reaped it yet.
|
|
function isAlive(pid) {
|
|
try {
|
|
process.kill(pid, 0);
|
|
} catch {
|
|
return false;
|
|
}
|
|
try {
|
|
return !/^\d+ \(.*\) Z/.test(readFileSync(`/proc/${pid}/stat`, "utf8"));
|
|
} catch {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
async function startDaemon({
|
|
omit = [], viaEnv = false, threads = String(THREADS), python = process.execPath, nested = false, host = "127.0.0.1",
|
|
relativeModels = false, extraArgs = [], root: givenRoot = null, entry = daemonPath, extraEnv = {}, hold = entry === daemonPath,
|
|
} = {}) {
|
|
const root = givenRoot ?? mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const stateRoot = nested ? path.join(root, "a", "b", "state") : root;
|
|
const modelsRoot = path.join(root, "models");
|
|
const settings = {
|
|
host,
|
|
port: "0",
|
|
"state-root": stateRoot,
|
|
python,
|
|
"worker-entry": stubPath,
|
|
"models-root": relativeModels ? "models-rel" : modelsRoot,
|
|
"kimodo-commit": "abc1234",
|
|
"python-env": "test-env",
|
|
};
|
|
if (threads !== null) settings.threads = threads;
|
|
for (const name of omit) delete settings[name];
|
|
const env = { ...process.env, LEAK_CHECK: "leaked", LANG: "xx_TEST.UTF-8", ...extraEnv };
|
|
for (const name of Object.keys(env)) if (name.startsWith("VILI_")) delete env[name];
|
|
const argv = [entry, ...extraArgs];
|
|
for (const [name, value] of Object.entries(settings)) {
|
|
if (viaEnv) env[`VILI_${{ host: "DAEMON_HOST", port: "DAEMON_PORT", "state-root": "STATE_ROOT", "worker-entry": "WORKER_ENTRY" }[name] ?? name.toUpperCase().replace("-", "_")}`] = value;
|
|
else argv.push(`--${name}`, value);
|
|
}
|
|
const proc = spawn(process.execPath, argv, { env, cwd: root, stdio: ["ignore", "pipe", "pipe"] });
|
|
let output = "";
|
|
let errors = "";
|
|
proc.stdout.on("data", (chunk) => { output += chunk; });
|
|
proc.stderr.on("data", (chunk) => { errors += chunk; });
|
|
const listening = await until(() => /listening on http:\/\/([^:]+):(\d+)/.exec(output), "the daemon to listen");
|
|
const port = listening[2];
|
|
// the hand-run daemon takes its lock after it listens; a test that wants a served state root waits for it
|
|
if (hold) await until(() => /holds the state root/.test(output), "the daemon to hold the state root");
|
|
const spawnLog = () => {
|
|
const log = path.join(stateRoot, "embeddings", "stub-spawns.log");
|
|
return existsSync(log) ? readFileSync(log, "utf8").split("\n").filter(Boolean) : [];
|
|
};
|
|
return {
|
|
root,
|
|
stateRoot,
|
|
modelsRoot,
|
|
proc,
|
|
boundHost: listening[1],
|
|
port: Number(port),
|
|
stderr: () => errors,
|
|
stdout: () => output,
|
|
spawns: () => spawnLog().filter((line) => /^\d+$/.test(line)).map(Number),
|
|
grandchildren: () => spawnLog().filter((line) => line.startsWith("grandchild ")).map((line) => Number(line.split(" ")[1])),
|
|
async stop() {
|
|
try {
|
|
if (proc.exitCode === null && proc.signalCode === null) {
|
|
proc.kill("SIGTERM");
|
|
await until(() => proc.exitCode !== null || proc.signalCode !== null, "the daemon to exit", 30);
|
|
}
|
|
} finally {
|
|
// a daemon that did not stop on SIGTERM must not outlive the test and hold the run open
|
|
if (proc.exitCode === null && proc.signalCode === null) proc.kill("SIGKILL");
|
|
rmSync(root, { recursive: true, force: true });
|
|
}
|
|
},
|
|
};
|
|
}
|
|
|
|
function request(port, method, urlPath, { body, headers = {} } = {}) {
|
|
return new Promise((resolve, reject) => {
|
|
const req = http.request({ host: "127.0.0.1", port, method, path: urlPath, headers }, (res) => {
|
|
const chunks = [];
|
|
res.on("data", (chunk) => chunks.push(chunk));
|
|
res.on("end", () => {
|
|
const raw = Buffer.concat(chunks).toString("utf8");
|
|
let parsed = null;
|
|
try {
|
|
parsed = JSON.parse(raw);
|
|
} catch {
|
|
// not JSON
|
|
}
|
|
resolve({ status: res.statusCode, headers: res.headers, raw, body: parsed });
|
|
});
|
|
});
|
|
req.on("error", reject);
|
|
req.end(body);
|
|
});
|
|
}
|
|
|
|
const generate = (daemon, payload = { prompt: "a person waves", duration: 2 }, headers = {}) =>
|
|
request(daemon.port, "POST", "/motion/generate", {
|
|
body: typeof payload === "string" ? payload : JSON.stringify(payload),
|
|
headers: { "content-type": "application/json", ...headers },
|
|
});
|
|
const operatorState = async (daemon) => (await request(daemon.port, "GET", "/operator-state")).body;
|
|
const backendOf = async (daemon) => (await operatorState(daemon)).backend;
|
|
|
|
test("the worker's port comes from its ready line", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 200);
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "ready");
|
|
assert.notEqual(daemon.port, 8824, "--port 0 binds a free port");
|
|
assert.ok(Number.isInteger(backend.port) && backend.port > 1 && backend.port !== 8825);
|
|
assert.equal(answer.body.worker.pid, backend.pid);
|
|
assert.deepEqual(answer.body.worker.argv.slice(-4), ["--host", "127.0.0.1", "--port", "0"]);
|
|
// the request reached the stub on the port the stub printed, not on a decoy line's port
|
|
assert.equal((await request(backend.port, "GET", "/health")).body.requests, 1);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("one child, respawned after exit", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const [first, second] = await Promise.all([generate(daemon), generate(daemon)]);
|
|
assert.equal(first.status, 200);
|
|
assert.equal(second.status, 200);
|
|
assert.equal(daemon.spawns().length, 1);
|
|
const [pid] = daemon.spawns();
|
|
process.kill(pid, "SIGKILL");
|
|
await until(async () => (await backendOf(daemon)).state === "stopped", "the daemon to see the exit");
|
|
assert.match((await backendOf(daemon)).lastError, /^the worker exited \(SIGKILL\)$/);
|
|
const third = await generate(daemon);
|
|
assert.equal(third.status, 200);
|
|
const spawns = daemon.spawns();
|
|
assert.equal(spawns.length, 2);
|
|
assert.notEqual(spawns[1], pid);
|
|
assert.equal(third.body.worker.pid, spawns[1]);
|
|
assert.equal((await backendOf(daemon)).lastError, null);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("environment and nice", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const { env, priority } = (await generate(daemon)).body.worker;
|
|
assert.equal(env.OMP_NUM_THREADS, String(THREADS));
|
|
assert.equal(env.MKL_NUM_THREADS, String(THREADS));
|
|
assert.equal(env.VILI_THREADS, String(THREADS));
|
|
assert.equal(env.HF_HUB_OFFLINE, "1");
|
|
assert.equal(env.TRANSFORMERS_OFFLINE, "1");
|
|
assert.equal(env.HF_HOME, daemon.modelsRoot);
|
|
assert.equal(env.VILI_KIMODO_COMMIT, "abc1234");
|
|
assert.equal(env.VILI_PYTHON_ENV, "test-env");
|
|
assert.equal(env.PYTHONDONTWRITEBYTECODE, "1");
|
|
assert.equal(env.PYTHONUNBUFFERED, "1");
|
|
assert.equal(env.PATH, process.env.PATH);
|
|
assert.equal(env.HOME, process.env.HOME);
|
|
assert.equal(env.LANG, "xx_TEST.UTF-8");
|
|
assert.ok(env.PYTHONPATH.endsWith(path.join("vendor", "CultLib", "packages", "cultcache-py", "src")));
|
|
assert.equal(env.LEAK_CHECK, undefined, "the daemon's own environment does not reach the worker");
|
|
assert.equal(priority, 10);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("every flag has an environment twin, and the thread cap defaults to 8", async () => {
|
|
const viaEnv = await startDaemon({ viaEnv: true, threads: "16" });
|
|
try {
|
|
const answer = await generate(viaEnv);
|
|
assert.equal(answer.status, 200);
|
|
assert.equal(answer.body.worker.env.OMP_NUM_THREADS, "16");
|
|
assert.deepEqual(await backendOf(viaEnv), {
|
|
state: "ready", pid: answer.body.worker.pid, port: (await backendOf(viaEnv)).port, threads: 16, pythonEnv: "test-env",
|
|
kimodoCommit: "abc1234", modelsRoot: viaEnv.modelsRoot, lastError: null,
|
|
});
|
|
} finally {
|
|
await viaEnv.stop();
|
|
}
|
|
const byDefault = await startDaemon({ threads: null });
|
|
try {
|
|
const answer = await generate(byDefault);
|
|
assert.equal(answer.body.worker.env.OMP_NUM_THREADS, "8");
|
|
assert.equal((await backendOf(byDefault)).threads, 8);
|
|
} finally {
|
|
await byDefault.stop();
|
|
}
|
|
const one = await startDaemon({ threads: "1", nested: true });
|
|
try {
|
|
assert.equal((await generate(one)).body.worker.env.OMP_NUM_THREADS, "1");
|
|
} finally {
|
|
await one.stop();
|
|
}
|
|
});
|
|
|
|
test("the worker's output lands in its log, and the ready line does not", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
const log = path.join(daemon.stateRoot, "logs", "worker.log");
|
|
const text = await until(() => {
|
|
const content = existsSync(log) ? readFileSync(log, "utf8") : "";
|
|
return content.includes("stub-stdout-line") && content.includes("stub-stderr-line") ? content : null;
|
|
}, "the worker log");
|
|
assert.ok(!text.includes(`"port":${(await backendOf(daemon)).port}`), "the real ready line is not logged");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a job is a record on disk", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const answer = await generate(daemon, { prompt: "a person waves", duration: 2, constraints: [{ type: "root2d" }] });
|
|
assert.equal(answer.status, 200);
|
|
const job = answer.body;
|
|
assert.equal(job.schema, "gamecult.vili.motion_job.v0");
|
|
assert.equal(job.request.schema, "gamecult.vili.motion_request.v0");
|
|
assert.equal(job.prompt, "a person waves");
|
|
assert.equal(job.model, "stub-model");
|
|
assert.equal(job.ok, true);
|
|
assert.equal(job.error, null);
|
|
assert.equal(job.artifactDir, path.join(daemon.stateRoot, "artifacts", job.jobId));
|
|
assert.equal(job.worker.output, path.join(job.artifactDir, "motion"));
|
|
assert.equal(job.worker.contentType, "application/json");
|
|
assert.equal(job.worker.prompt, "a person waves");
|
|
assert.equal(job.worker.constraints, path.join(job.artifactDir, "constraints.json"));
|
|
assert.deepEqual(JSON.parse(readFileSync(job.worker.constraints, "utf8")), [{ type: "root2d" }]);
|
|
assert.deepEqual(JSON.parse(readFileSync(path.join(daemon.stateRoot, "jobs", `${job.jobId}.json`), "utf8")), job);
|
|
const batch = await generate(daemon, { texts: ["a", "b"], durations: [1, 1] });
|
|
assert.equal(batch.status, 200);
|
|
assert.equal(batch.body.prompt, null);
|
|
assert.equal(batch.body.worker.prompt, undefined);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a requested model restarts the worker only when it is not the one serving", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "stub-model" })).status, 200);
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "Stub Model" })).status, 200);
|
|
assert.equal(daemon.spawns().length, 1);
|
|
const [first] = daemon.spawns();
|
|
const other = await generate(daemon, { prompt: "x", model: "other-model" });
|
|
assert.equal(other.status, 200);
|
|
assert.equal(other.body.model, "other-model");
|
|
assert.deepEqual(other.body.worker.argv.slice(-2), ["--model", "other-model"]);
|
|
assert.equal(daemon.spawns().length, 2);
|
|
await until(() => !isAlive(first), "the first worker to be gone");
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
assert.equal(daemon.spawns().length, 2, "a request that names no model keeps the one serving");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
for (const signal of ["SIGTERM", "SIGINT"]) {
|
|
test(`${signal} leaves no child`, async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
const [pid] = daemon.spawns();
|
|
assert.ok(isAlive(pid));
|
|
daemon.proc.kill(signal);
|
|
await until(() => daemon.proc.exitCode !== null, "the daemon to exit", 8);
|
|
assert.equal(isAlive(pid), false);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
}
|
|
|
|
test("a worker that is stopping gets its grace, not a SIGKILL", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "graceful" })).status, 200);
|
|
daemon.proc.kill("SIGTERM");
|
|
await until(() => daemon.proc.exitCode !== null, "the daemon to exit", 8);
|
|
assert.ok(readFileSync(path.join(daemon.stateRoot, "embeddings", "stub-spawns.log"), "utf8").includes("graceful-exit"));
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that dies while the daemon waits for its health fails the job at once", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const began = Date.now();
|
|
const failed = await generate(daemon, { prompt: "x", model: "die-after-ready" });
|
|
assert.equal(failed.status, 502);
|
|
assert.ok(Date.now() - began < 20000, "the default deadline is minutes; a dead worker does not wait for it");
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "stopped");
|
|
assert.equal(backend.lastError, "the worker exited (code 4)");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("the daemon binds loopback unless told otherwise, and ignores flags it does not know", async () => {
|
|
const byDefault = await startDaemon({ omit: ["host"], extraArgs: ["--bogus"] });
|
|
try {
|
|
assert.equal(byDefault.boundHost, "127.0.0.1");
|
|
assert.equal((await request(byDefault.port, "GET", "/health")).status, 200);
|
|
} finally {
|
|
await byDefault.stop();
|
|
}
|
|
const named = await startDaemon({ host: "127.0.0.2" });
|
|
try {
|
|
assert.equal(named.boundHost, "127.0.0.2");
|
|
assert.equal((await fetch(`http://127.0.0.2:${named.port}/health`)).status, 200);
|
|
} finally {
|
|
await named.stop();
|
|
}
|
|
});
|
|
|
|
test("a relative models root reaches the worker as an absolute path", async () => {
|
|
const daemon = await startDaemon({ relativeModels: true });
|
|
try {
|
|
const { env } = (await generate(daemon)).body.worker;
|
|
assert.equal(env.HF_HOME, path.join(path.dirname(daemon.modelsRoot), "models-rel"));
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that ignores SIGTERM is killed with everything it started", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "stubborn" })).status, 200);
|
|
const [pid] = daemon.spawns();
|
|
const [grandchild] = daemon.grandchildren();
|
|
assert.ok(isAlive(pid) && isAlive(grandchild));
|
|
daemon.proc.kill("SIGTERM");
|
|
await until(() => daemon.proc.exitCode !== null, "the daemon to exit", 30);
|
|
assert.equal(isAlive(pid), false);
|
|
assert.equal(isAlive(grandchild), false);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
// The daemon owns the worker's whole tree (motion-pipeline:finding:cut-vili-cpu-body.s1.orphaned-descendants).
|
|
const aliveWorkers = (daemon) => daemon.spawns().filter(isAlive);
|
|
|
|
test("a worker killed mid-request leaves no encoder child, and the next job has exactly one worker", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const pending = generate(daemon, { prompt: "x", model: "encoder" });
|
|
const [grandchild] = await until(() => { const found = daemon.grandchildren(); return found.length ? found : null; }, "the encoder child");
|
|
const [worker] = daemon.spawns();
|
|
assert.ok(isAlive(worker) && isAlive(grandchild));
|
|
process.kill(worker, "SIGKILL");
|
|
assert.equal(readFileSync(path.join(daemon.stateRoot, "worker.pid"), "utf8"), `${worker}\n`);
|
|
assert.equal((await pending).status, 502);
|
|
await until(() => !isAlive(grandchild), "the encoder child to die with its worker");
|
|
await until(() => !existsSync(path.join(daemon.stateRoot, "worker.pid")), "the dead worker's pid file to go");
|
|
assert.equal((await generate(daemon, { prompt: "y" })).status, 200);
|
|
assert.equal(aliveWorkers(daemon).length, 1);
|
|
assert.equal(readFileSync(path.join(daemon.stateRoot, "worker.pid"), "utf8"), `${aliveWorkers(daemon)[0]}\n`);
|
|
assert.equal(isAlive(grandchild), false);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
// "encoder" leaves a SIGTERM-proof grandchild beside the worker; "stubborn" is a worker that ignores SIGTERM; the plain stub is just a worker.
|
|
for (const model of ["encoder", "stubborn", "stub-model"]) {
|
|
test(`a restarted daemon kills the ${model} worker its crashed predecessor left, and runs exactly one`, async () => {
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const first = await startDaemon({ root });
|
|
const stray = generate(first, { prompt: "x", model }).catch(() => null);
|
|
await until(() => first.spawns().length, "the worker");
|
|
if (model === "encoder") await until(() => first.grandchildren().length, "the encoder child");
|
|
await until(() => request(first.port, "GET", "/health").then((r) => r.body?.state?.backend?.state === "ready"), "the worker to be ready");
|
|
const [worker] = first.spawns();
|
|
const strays = [worker, ...first.grandchildren()];
|
|
first.proc.kill("SIGKILL");
|
|
await until(() => first.proc.exitCode !== null || first.proc.signalCode !== null, "the first daemon to die");
|
|
await stray;
|
|
assert.ok(strays.every(isAlive), "the probe must reproduce the orphan before the fix can be measured");
|
|
const second = await startDaemon({ root });
|
|
try {
|
|
await until(() => !strays.some(isAlive), "the orphans to be killed");
|
|
assert.equal(existsSync(path.join(root, "worker.pid")), false);
|
|
assert.equal((await generate(second, { prompt: "y" })).status, 200);
|
|
assert.equal(aliveWorkers(second).length, 1);
|
|
second.proc.kill("SIGTERM");
|
|
await until(() => second.proc.exitCode !== null, "the second daemon to exit", 30);
|
|
assert.equal(aliveWorkers(second).length, 0);
|
|
} finally {
|
|
await second.stop();
|
|
}
|
|
});
|
|
}
|
|
|
|
test("a live worker outside the pid file's group is left alone", async () => {
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const first = await startDaemon({ root });
|
|
const bystander = spawn(process.execPath, ["-e", "setInterval(() => {}, 1000)"], { detached: true, stdio: "ignore" });
|
|
try {
|
|
assert.equal((await generate(first)).status, 200);
|
|
const [worker] = first.spawns();
|
|
first.proc.kill("SIGKILL");
|
|
await until(() => first.proc.exitCode !== null || first.proc.signalCode !== null, "the first daemon to die");
|
|
writeFileSync(path.join(root, "worker.pid"), `${bystander.pid}\n`);
|
|
const second = await startDaemon({ root });
|
|
assert.ok(isAlive(worker) && isAlive(bystander.pid), "only the group the pid file names, and only if it is ours, is killed");
|
|
second.proc.kill("SIGTERM");
|
|
await until(() => second.proc.exitCode !== null, "the second daemon to exit", 30);
|
|
process.kill(-worker, "SIGKILL");
|
|
} finally {
|
|
bystander.kill("SIGKILL");
|
|
await first.stop();
|
|
}
|
|
});
|
|
|
|
// The marker that proves a group is ours is this state root's own embeddings directory. A live worker of another
|
|
// state root, named by a stale pid file, carries another root's marker and is left alone.
|
|
test("a pid file naming another state root's live worker kills nothing", async () => {
|
|
const other = await startDaemon();
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
try {
|
|
assert.equal((await generate(other)).status, 200);
|
|
const [foreign] = other.spawns();
|
|
writeFileSync(path.join(root, "worker.pid"), `${foreign}\n`);
|
|
const daemon = await startDaemon({ root });
|
|
try {
|
|
assert.ok(isAlive(foreign), "a worker carrying another root's marker must survive the reclaim");
|
|
assert.equal(existsSync(path.join(root, "worker.pid")), false, "the stale pid file is still forgotten");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
} finally {
|
|
await other.stop();
|
|
}
|
|
});
|
|
|
|
test("a pid file naming a process group that is not the worker's kills nothing", async () => {
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const bystander = spawn(process.execPath, ["-e", "setInterval(() => {}, 1000)"], { detached: true, stdio: "ignore" });
|
|
writeFileSync(path.join(root, "worker.pid"), `${bystander.pid}\n`);
|
|
const daemon = await startDaemon({ root });
|
|
try {
|
|
assert.equal(isAlive(bystander.pid), true);
|
|
assert.equal(existsSync(path.join(root, "worker.pid")), false);
|
|
} finally {
|
|
bystander.kill("SIGKILL");
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("cross-site is refused before the worker", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
const withOrigin = await generate(daemon, undefined, { origin: "https://example.test" });
|
|
assert.equal(withOrigin.status, 403);
|
|
assert.deepEqual(withOrigin.body, { ok: false, code: "cross-site" });
|
|
const plain = await generate(daemon, undefined, { "content-type": "text/plain" });
|
|
assert.equal(plain.status, 403);
|
|
assert.deepEqual(plain.body, { ok: false, code: "cross-site" });
|
|
const untyped = await request(daemon.port, "POST", "/motion/generate", { body: JSON.stringify({ prompt: "x" }) });
|
|
assert.equal(untyped.status, 403);
|
|
const options = await request(daemon.port, "OPTIONS", "/motion/generate", { headers: { origin: "https://example.test" } });
|
|
assert.equal(options.status, 404);
|
|
const health = await request(daemon.port, "GET", "/health");
|
|
for (const reply of [withOrigin, plain, untyped, options, health]) {
|
|
assert.equal(Object.keys(reply.headers).filter((name) => name.startsWith("access-control-")).length, 0);
|
|
}
|
|
const backend = await backendOf(daemon);
|
|
assert.equal((await request(backend.port, "GET", "/health")).body.requests, 1, "only the same-site request reached the worker");
|
|
// the media type is matched as a type, not as a string
|
|
assert.equal((await generate(daemon, undefined, { "content-type": "Application/JSON; charset=utf-8" })).status, 200);
|
|
assert.equal((await generate(daemon, undefined, { "content-type": "application/json ; charset=utf-8" })).status, 200);
|
|
// and only the generate route answers a POST
|
|
const stray = await request(daemon.port, "POST", "/motion/other", {
|
|
body: JSON.stringify({ prompt: "x" }),
|
|
headers: { "content-type": "application/json" },
|
|
});
|
|
assert.equal(stray.status, 404);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
for (const omitted of ["models-root", "kimodo-commit", "python-env"]) {
|
|
test(`unconfigured (no ${omitted}) refuses before spawning`, async () => {
|
|
const daemon = await startDaemon({ omit: [omitted] });
|
|
try {
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 503);
|
|
assert.deepEqual(answer.body, { ok: false, code: "worker-unconfigured" });
|
|
assert.deepEqual(daemon.spawns(), []);
|
|
assert.equal((await backendOf(daemon)).state, "stopped");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
}
|
|
|
|
test("an error never echoes the request", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const malformed = await generate(daemon, "{\"prompt\": \"the launch code is swordfish\", ");
|
|
assert.equal(malformed.status, 400);
|
|
assert.deepEqual(malformed.body, { ok: false, error: "Invalid JSON." });
|
|
const missing = await generate(daemon, { texts: ["the launch code is swordfish"] });
|
|
assert.equal(missing.status, 400);
|
|
assert.ok(!missing.raw.includes("swordfish"));
|
|
// a body the handler cannot use at all: the answer names nothing, and neither does the log
|
|
const unusable = await generate(daemon, "null");
|
|
assert.equal(unusable.status, 500);
|
|
assert.deepEqual(unusable.body, { ok: false, code: "internal" });
|
|
await until(() => daemon.stderr().includes("request failed: TypeError"), "the daemon to log the class");
|
|
assert.ok(!daemon.stderr().includes("swordfish"));
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that dies before it is ready fails the job, and the next job tries again", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const failed = await generate(daemon, { prompt: "x", model: "crash-before-ready" });
|
|
assert.equal(failed.status, 502);
|
|
assert.equal(failed.body.ok, false);
|
|
assert.equal(failed.body.error, "the worker request failed (Error)");
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "failed");
|
|
assert.equal(backend.lastError, "the worker exited (code 3)");
|
|
assert.equal(backend.pid, null);
|
|
assert.equal(backend.port, null);
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
assert.equal((await backendOf(daemon)).state, "ready");
|
|
assert.equal(daemon.spawns().length, 2);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that never says it is ready is killed at the deadline", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const pending = generate(daemon, { prompt: "x", model: "hang-before-ready", workerTimeoutMs: 1500 });
|
|
const starting = await until(async () => {
|
|
const backend = await backendOf(daemon);
|
|
return backend.state === "starting" ? backend : null;
|
|
}, "the starting state");
|
|
assert.ok(Number.isInteger(starting.pid) && starting.port === null);
|
|
const failed = await pending;
|
|
assert.equal(failed.status, 502);
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "failed");
|
|
assert.equal(backend.lastError, "the worker was not ready in time");
|
|
assert.equal(isAlive(daemon.spawns()[0]), false);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that is ready but not healthy fails at the deadline; one that gets healthy in time serves", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const failed = await generate(daemon, { prompt: "x", model: "sick", workerTimeoutMs: 1500 });
|
|
assert.equal(failed.status, 502);
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "failed");
|
|
assert.equal(backend.lastError, "the worker did not become healthy in time");
|
|
assert.equal(isAlive(daemon.spawns()[0]), false);
|
|
const slow = await generate(daemon, { prompt: "x", model: "slow-health" });
|
|
assert.equal(slow.status, 200);
|
|
assert.equal((await backendOf(daemon)).state, "ready");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a missing interpreter fails the job at once", async () => {
|
|
const daemon = await startDaemon({ python: "/no/such/python" });
|
|
try {
|
|
const began = Date.now();
|
|
const failed = await generate(daemon);
|
|
assert.equal(failed.status, 502);
|
|
assert.ok(Date.now() - began < 5000, "nothing waits on a process that never existed");
|
|
const backend = await backendOf(daemon);
|
|
assert.equal(backend.state, "failed");
|
|
assert.equal(backend.lastError, "the worker could not be spawned");
|
|
assert.equal(existsSync(path.join(daemon.stateRoot, "worker.pid")), false);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker that refuses, errors or drops the request fails the job without echoing it", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
const refused = await generate(daemon, { prompt: "swordfish", model: "refuse-generate" });
|
|
assert.equal(refused.status, 502);
|
|
assert.equal(refused.body.ok, false);
|
|
assert.equal(refused.body.error, null);
|
|
const errored = await generate(daemon, { prompt: "swordfish", model: "fail-generate" });
|
|
assert.equal(errored.status, 502);
|
|
assert.equal(errored.body.error, "the worker answered 500");
|
|
const dropped = await generate(daemon, { prompt: "swordfish", model: "drop-generate" });
|
|
assert.equal(dropped.status, 502);
|
|
assert.equal(dropped.body.error, "the worker request failed (TypeError)");
|
|
const stalled = await generate(daemon, { prompt: "swordfish", model: "slow-generate", timeoutMs: 700 });
|
|
assert.equal(stalled.status, 502);
|
|
assert.equal(stalled.body.error, "the worker request failed (TimeoutError)");
|
|
for (const answer of [refused, errored, dropped, stalled]) assert.ok(!answer.body.error?.includes("swordfish"));
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("health, operator state and the unknown", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
for (const route of ["/", "/health"]) {
|
|
const answer = await request(daemon.port, "GET", route);
|
|
assert.equal(answer.status, 200);
|
|
assert.equal(answer.body.ok, true);
|
|
assert.equal(answer.body.service, "vili");
|
|
assert.equal(answer.body.state.backend.threads, THREADS);
|
|
}
|
|
assert.equal((await request(daemon.port, "GET", "/motion/generate")).status, 404);
|
|
// the retired operator route, spelled in parts so the retirement grep over scripts/ stays empty
|
|
assert.equal((await request(daemon.port, "GET", ["", "eve", "operator"].join("/"))).status, 404);
|
|
const unknown = await request(daemon.port, "GET", "/provider-advertisement");
|
|
assert.equal(unknown.status, 404);
|
|
assert.equal(unknown.raw, "not found\n");
|
|
const state = await operatorState(daemon);
|
|
assert.equal(state.host, os.hostname());
|
|
assert.deepEqual(Object.keys(state.routes), ["health", "operatorState", "generate"]);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("the typed service state is refreshed while the daemon runs", async () => {
|
|
const daemon = await startDaemon({ nested: true });
|
|
try {
|
|
const store = path.join(daemon.stateRoot, "vili.service.cc");
|
|
assert.ok(existsSync(store));
|
|
assert.ok(!readFileSync(store).includes("ready"));
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
await until(() => readFileSync(store).includes("ready"), "the store to show the worker ready", 20);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a thread count outside 1 to 16 stops the daemon without echoing it", async () => {
|
|
for (const threads of ["0", "17", "99", "100", "8x", "x8", "-1", "1.5"]) {
|
|
const proc = spawn(process.execPath, [daemonPath, "--threads", threads], { stdio: ["ignore", "pipe", "pipe"] });
|
|
let output = "";
|
|
proc.stdout.on("data", (chunk) => { output += chunk; });
|
|
proc.stderr.on("data", (chunk) => { output += chunk; });
|
|
const code = await new Promise((resolve) => proc.on("exit", resolve));
|
|
assert.equal(code, 2);
|
|
assert.match(output, /threads must be an integer from 1 to 16/);
|
|
assert.ok(!output.replace("from 1 to 16", "").includes(threads), `the refusal repeated ${threads}`);
|
|
}
|
|
});
|
|
|
|
const msgpackPost = (daemon, body, headers = {}) =>
|
|
new Promise((resolve, reject) => {
|
|
const req = http.request({ host: "127.0.0.1", port: daemon.port, method: "POST", path: "/cultnet/snapshot", headers: { "content-type": "application/msgpack", ...headers } }, (res) => {
|
|
const chunks = [];
|
|
res.on("data", (chunk) => chunks.push(chunk));
|
|
res.on("end", () => resolve({ status: res.statusCode, headers: res.headers, raw: Buffer.concat(chunks) }));
|
|
});
|
|
req.on("error", reject);
|
|
req.end(body);
|
|
});
|
|
const SECRET = "canary-9f3c-never-echoed";
|
|
const snapshotRequest = (messageId = "m1") => Buffer.from(encode({ schemaVersion: "cultnet.snapshot_request.v0", messageId }));
|
|
|
|
// A second daemon on a state root a live daemon holds must refuse, and must not touch the first one's worker.
|
|
test("a second daemon on a live state root exits 2 and kills nothing", async () => {
|
|
const first = await startDaemon();
|
|
let second = null;
|
|
try {
|
|
assert.equal((await generate(first)).status, 200);
|
|
const [worker] = first.spawns();
|
|
assert.ok(isAlive(worker));
|
|
second = await startDaemon({ root: first.root, hold: false });
|
|
await until(() => second.proc.exitCode !== null, "the second daemon to refuse", 20);
|
|
assert.equal(second.proc.exitCode, 2);
|
|
assert.match(second.stderr(), /held by a running daemon/);
|
|
assert.ok(isAlive(worker), "the first daemon's worker must survive the refusal");
|
|
assert.equal((await generate(first)).status, 200);
|
|
assert.equal(aliveWorkers(first).length, 1);
|
|
} finally {
|
|
if (second && second.proc.exitCode === null && second.proc.signalCode === null) second.proc.kill("SIGKILL"); // a daemon that failed to refuse must not outlive the test
|
|
await first.stop();
|
|
}
|
|
});
|
|
|
|
// Two daemons start together over the lock a crashed daemon left. Exactly one owns the root, every time: the
|
|
// other exits 2, and one generate leaves one worker on the root.
|
|
for (let trial = 1; trial <= 40; trial += 1) {
|
|
test(`two daemons racing a crashed daemon's lock leave one owner and one worker (trial ${trial})`, async () => {
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const crashed = await startDaemon({ root });
|
|
crashed.proc.kill("SIGKILL");
|
|
await until(() => crashed.proc.exitCode !== null || crashed.proc.signalCode !== null, "the first daemon to die");
|
|
const racers = await Promise.all([startDaemon({ root, hold: false }), startDaemon({ root, hold: false })]);
|
|
try {
|
|
await until(() => racers.every((racer) => racer.proc.exitCode !== null || /holds the state root/.test(racer.stdout())), "both racers to decide", 20);
|
|
await new Promise((resolve) => setTimeout(resolve, 300));
|
|
const owners = racers.filter((racer) => racer.proc.exitCode === null);
|
|
assert.equal(owners.length, 1, "exactly one daemon stays up");
|
|
assert.match(owners[0].stdout(), /holds the state root/);
|
|
const loser = racers.find((racer) => racer !== owners[0]);
|
|
assert.equal(loser.proc.exitCode, 2);
|
|
assert.match(loser.stderr(), /held by a running daemon/);
|
|
for (const racer of owners) assert.equal((await generate(racer)).status, 200);
|
|
assert.equal(aliveWorkers(owners[0]).length, 1, "one worker on the root");
|
|
} finally {
|
|
for (const racer of racers) if (racer.proc.exitCode === null && racer.proc.signalCode === null) racer.proc.kill("SIGKILL");
|
|
for (const pid of racers[0].spawns()) { try { process.kill(-pid, "SIGKILL"); } catch { /* gone */ } }
|
|
rmSync(root, { recursive: true, force: true });
|
|
}
|
|
});
|
|
}
|
|
|
|
const idunnDaemon = async ({ leased = false, extraEnv = {}, loseAfterMeta = false } = {}) => {
|
|
const root = mkdtempSync(path.join(os.tmpdir(), "vili-test-"));
|
|
const lease = path.join(root, "lease");
|
|
const healthLog = path.join(root, "health.log");
|
|
if (leased) writeFileSync(lease, "");
|
|
const daemon = await startDaemon({
|
|
root, nested: true, entry: fakeIdunnEntry, extraEnv: {
|
|
FAKE_IDUNN_LEASE: lease, FAKE_IDUNN_HEALTH_LOG: healthLog, ...extraEnv,
|
|
...(loseAfterMeta ? { FAKE_IDUNN_LOSE_AFTER_META: path.join(root, "a", "b", "state", "artifacts") } : {}),
|
|
},
|
|
});
|
|
return Object.assign(daemon, { lease, health: () => (existsSync(healthLog) ? readFileSync(healthLog, "utf8").split("\n").filter(Boolean) : []) });
|
|
};
|
|
|
|
test("before the lease nothing is written, spawned or accepted", async () => {
|
|
const daemon = await idunnDaemon();
|
|
try {
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 503);
|
|
assert.equal(answer.body.code, "write-lease-pending");
|
|
assert.deepEqual(daemon.spawns(), []);
|
|
assert.equal(existsSync(daemon.stateRoot), false, "the state root must not exist before the lease");
|
|
assert.deepEqual(daemon.health(), []);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("with the lease the daemon writes and serves, and no pid file exists", async () => {
|
|
const daemon = await idunnDaemon();
|
|
try {
|
|
writeFileSync(daemon.lease, "");
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 200);
|
|
assert.equal(existsSync(path.join(daemon.stateRoot, "worker.pid")), false);
|
|
assert.equal(existsSync(path.join(daemon.stateRoot, "daemon.lock")), false);
|
|
assert.deepEqual(daemon.health(), ["active"]);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("losing the lease kills the worker tree, refuses, and writes nothing more", async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "stubborn" })).status, 200);
|
|
const [worker] = daemon.spawns();
|
|
const [grandchild] = await until(() => { const found = daemon.grandchildren(); return found.length ? found : null; }, "the armed grandchild");
|
|
assert.ok(isAlive(worker) && isAlive(grandchild));
|
|
rmSync(daemon.lease);
|
|
await until(() => !isAlive(worker) && !isAlive(grandchild), "the worker tree to die", 20);
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 503);
|
|
assert.equal(answer.body.code, "write-lease-lost");
|
|
await until(() => daemon.health().join() === "active,degraded", "the health to say degraded");
|
|
const store = path.join(daemon.stateRoot, "vili.service.cc");
|
|
const before = statSync(store).mtimeMs;
|
|
const jobs = readdirSync(path.join(daemon.stateRoot, "jobs")).length;
|
|
await new Promise((resolve) => setTimeout(resolve, 6500));
|
|
assert.equal(statSync(store).mtimeMs, before);
|
|
assert.equal(readdirSync(path.join(daemon.stateRoot, "jobs")).length, jobs);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
// A generate whose headers are in and whose body is still arriving when the lease goes. The lease is gone, so
|
|
// nothing is created and nothing is spawned when the body completes, whether the daemon has noticed the loss yet
|
|
// (its 5 s tick) or not.
|
|
for (const noticed of [false, true]) {
|
|
test(`a request in flight when the lease is lost writes and spawns nothing (loss ${noticed ? "noticed" : "not yet noticed"})`, async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
const body = JSON.stringify({ prompt: "x", duration: 2 });
|
|
const socket = net.connect(daemon.port, "127.0.0.1");
|
|
let raw = "";
|
|
socket.on("data", (chunk) => { raw += chunk; });
|
|
const closed = new Promise((resolve) => socket.on("close", resolve));
|
|
socket.write(`POST /motion/generate HTTP/1.1\r\nhost: vili\r\ncontent-type: application/json\r\ncontent-length: ${body.length}\r\nconnection: close\r\n\r\n${body.slice(0, 5)}`);
|
|
await new Promise((resolve) => setTimeout(resolve, 200));
|
|
rmSync(daemon.lease);
|
|
if (noticed) await until(() => daemon.health().join() === "active,degraded", "the daemon to notice the loss", 15);
|
|
socket.write(body.slice(5));
|
|
await Promise.race([closed, new Promise((resolve) => setTimeout(resolve, 15000))]);
|
|
socket.destroy();
|
|
assert.match(raw, /^HTTP\/1\.1 503 /);
|
|
assert.match(raw, /write-lease-lost/);
|
|
assert.deepEqual(daemon.spawns(), [], "no worker is spawned after the loss");
|
|
const artifacts = path.join(daemon.stateRoot, "artifacts");
|
|
assert.deepEqual(existsSync(artifacts) ? readdirSync(artifacts) : [], [], "no artifact directory is created");
|
|
assert.deepEqual(readdirSync(path.join(daemon.stateRoot, "jobs")), []);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
}
|
|
|
|
// The lease goes after the job wrote its meta.json and before its worker would start: the spawn is gated at the
|
|
// spawn, not by the request's earlier checks.
|
|
test("a worker is not started, and its directories are not made, once the lease is gone", async () => {
|
|
const daemon = await idunnDaemon({ leased: true, loseAfterMeta: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 503);
|
|
assert.equal(answer.body.code, "write-lease-lost");
|
|
assert.deepEqual(daemon.spawns(), []);
|
|
assert.equal(existsSync(path.join(daemon.stateRoot, "logs")), false);
|
|
assert.equal(existsSync(path.join(daemon.stateRoot, "embeddings")), false);
|
|
assert.deepEqual(readdirSync(path.join(daemon.stateRoot, "jobs")), []);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a request arriving right after the lease is lost writes and spawns nothing", async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
rmSync(daemon.lease);
|
|
const answer = await generate(daemon);
|
|
assert.equal(answer.status, 503);
|
|
assert.equal(answer.body.code, "write-lease-lost");
|
|
assert.deepEqual(daemon.spawns(), []);
|
|
const artifacts = path.join(daemon.stateRoot, "artifacts");
|
|
assert.deepEqual(existsSync(artifacts) ? readdirSync(artifacts) : [], []);
|
|
assert.deepEqual(readdirSync(path.join(daemon.stateRoot, "jobs")), []);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a worker respawned under the lease is killed when the lease is lost", async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
const [first] = daemon.spawns();
|
|
process.kill(-first, "SIGKILL");
|
|
await until(async () => (await backendOf(daemon)).state === "stopped", "the daemon to see the worker gone");
|
|
assert.equal((await generate(daemon)).status, 200);
|
|
const second = daemon.spawns().find((pid) => pid !== first);
|
|
assert.ok(second && isAlive(second));
|
|
rmSync(daemon.lease);
|
|
await until(() => !isAlive(second), "the respawned worker to die", 20);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
// Idunn stops an incumbent (SIGTERM to the unit) before it revokes the lease. What Vili owns in that sequence: the
|
|
// worker tree is dead when the process ends, it exits 0, and nothing is written from the signal on.
|
|
test("on SIGTERM the daemon stops its worker tree, writes nothing more, and exits 0", async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
assert.equal((await generate(daemon, { prompt: "x", model: "stubborn" })).status, 200);
|
|
const [worker] = daemon.spawns();
|
|
const [grandchild] = await until(() => { const found = daemon.grandchildren(); return found.length ? found : null; }, "the armed grandchild");
|
|
const store = path.join(daemon.stateRoot, "vili.service.cc");
|
|
const before = statSync(store).mtimeMs;
|
|
const jobs = readdirSync(path.join(daemon.stateRoot, "jobs"));
|
|
daemon.proc.kill("SIGTERM");
|
|
await until(() => daemon.proc.exitCode !== null || daemon.proc.signalCode !== null, "the daemon to exit", 30);
|
|
assert.equal(daemon.proc.exitCode, 0);
|
|
assert.ok(!isAlive(worker) && !isAlive(grandchild), "the worker tree is dead when the daemon is gone");
|
|
assert.equal(statSync(store).mtimeMs, before, "the 5 s surface tick must not write during the stop");
|
|
assert.deepEqual(readdirSync(path.join(daemon.stateRoot, "jobs")), jobs);
|
|
assert.ok(existsSync(daemon.lease), "the daemon never touches the lease Idunn revokes");
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a job that finishes after the lease is lost is not recorded", async () => {
|
|
const daemon = await idunnDaemon({ leased: true });
|
|
try {
|
|
await until(() => existsSync(path.join(daemon.stateRoot, "vili.service.cc")), "the service store", 10);
|
|
const slow = generate(daemon, { prompt: "x", model: "slow-generate" });
|
|
slow.catch(() => {}); // a failing assertion below must not leave this request pending
|
|
await until(() => daemon.spawns().length, "the worker");
|
|
await until(() => request(daemon.port, "GET", "/health").then((r) => r.body?.state?.backend?.state === "ready"), "the worker to be ready");
|
|
rmSync(daemon.lease);
|
|
const answer = await Promise.race([slow, new Promise((resolve) => setTimeout(() => resolve({ status: "no answer in 30 s" }), 30000))]);
|
|
assert.equal(answer.status, 503);
|
|
assert.equal(answer.body.code, "write-lease-lost");
|
|
assert.deepEqual(readdirSync(path.join(daemon.stateRoot, "jobs")), []);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("the route challenge is answered ahead of the cross-site refusal, and an Origin is still refused", async () => {
|
|
const daemon = await idunnDaemon();
|
|
try {
|
|
// the challenge is how Idunn learns the incarnation is warming, so it is answered before the lease
|
|
const answered = await msgpackPost(daemon, snapshotRequest("m-42"));
|
|
assert.equal(answered.status, 200);
|
|
assert.equal(answered.headers["content-type"], "application/msgpack");
|
|
const message = decode(answered.raw);
|
|
assert.equal(message.schemaVersion, "cultnet.snapshot_response_raw.v0");
|
|
assert.equal(message.messageId, "m-42");
|
|
assert.equal(message.documents.length, 1);
|
|
assert.equal((await msgpackPost(daemon, snapshotRequest(), { origin: "https://evil.example" })).status, 403);
|
|
assert.equal((await msgpackPost(daemon, Buffer.from("not msgpack at all"))).status, 400);
|
|
assert.equal((await msgpackPost(daemon, Buffer.from(encode({ schemaVersion: "cultnet.snapshot_response_raw.v0", messageId: "x", documents: [] })))).status, 400);
|
|
assert.equal((await msgpackPost(daemon, Buffer.alloc(70 * 1024))).status, 400);
|
|
const text = await request(daemon.port, "POST", "/motion/generate", { body: "{}", headers: { "content-type": "text/plain" } });
|
|
assert.equal(text.status, 403);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("a signer that cannot answer gives 503 and echoes nothing", async () => {
|
|
const daemon = await idunnDaemon({ extraEnv: { FAKE_IDUNN_ANSWER_THROWS: "1" } });
|
|
try {
|
|
const answered = await msgpackPost(daemon, snapshotRequest(SECRET));
|
|
assert.equal(answered.status, 503);
|
|
assert.ok(!answered.raw.toString("utf8").includes(SECRET));
|
|
assert.ok(!daemon.stderr().includes(SECRET));
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
test("the hand-run daemon has no route challenge", async () => {
|
|
const daemon = await startDaemon();
|
|
try {
|
|
assert.equal((await msgpackPost(daemon, snapshotRequest())).status, 404);
|
|
assert.equal((await msgpackPost(daemon, snapshotRequest(), { origin: "https://evil.example" })).status, 404);
|
|
} finally {
|
|
await daemon.stop();
|
|
}
|
|
});
|
|
|
|
// ---- the Idunn entry's own rules: its argument contract, and what it derives for the daemon
|
|
|
|
const entryPath = path.join(here, "idunn-entry.mjs");
|
|
const runEntry = (args, env = {}) => new Promise((resolve) => {
|
|
const proc = spawn(process.execPath, [entryPath, ...args], { env: { PATH: process.env.PATH, ...env }, stdio: ["ignore", "pipe", "pipe"] });
|
|
let output = "";
|
|
proc.stdout.on("data", (chunk) => { output += chunk; });
|
|
proc.stderr.on("data", (chunk) => { output += chunk; });
|
|
proc.on("exit", (code) => resolve({ code, output }));
|
|
});
|
|
|
|
test("the entry takes exactly --state-root <absolute> --kimodo-commit <40 hex>, and echoes none of a bad value", async () => {
|
|
const commit = "6bb58488037dd65360ff0c5d1692b403a23309f7";
|
|
const good = ["--state-root", "/var/lib/gamecult/vili", "--kimodo-commit", commit];
|
|
const bads = [
|
|
[],
|
|
[...good, "--extra", "x"],
|
|
["--kimodo-commit", commit, "--state-root", "/var/lib/gamecult/vili"],
|
|
["--state-root", "canary-relative-path", "--kimodo-commit", commit],
|
|
["--state-root", "/var/lib/gamecult/vili", "--kimodo-commit", "canary-not-a-commit"],
|
|
["--state-root", "/var/lib/gamecult/vili", "--kimodo-commit", commit.toUpperCase()],
|
|
];
|
|
for (const args of bads) {
|
|
const answer = await runEntry(args, { GAMECULT_IDUNN_CANDIDATE_BIND: "127.0.0.1:18861" });
|
|
assert.equal(answer.code, 2, JSON.stringify(args));
|
|
assert.ok(!answer.output.includes("canary"), `the refusal repeated a value of ${JSON.stringify(args)}`);
|
|
}
|
|
for (const badBind of ["canary-not-a-bind", "127.0.0.1:65536", "127.0.0.1:99999", "127.0.0.1:123456", "127.0.0.1:"]) {
|
|
const bind = await runEntry(good, { GAMECULT_IDUNN_CANDIDATE_BIND: badBind });
|
|
assert.equal(bind.code, 2, badBind);
|
|
assert.match(bind.output, /CANDIDATE_BIND is not host:port/);
|
|
assert.ok(!bind.output.includes("canary"));
|
|
assert.ok(!bind.output.includes("65536") && !bind.output.includes("99999"));
|
|
}
|
|
});
|
|
|
|
test("the entry derives the daemon's host, port, interpreter and environment identity from the release", async () => {
|
|
const { createHash } = await import("node:crypto");
|
|
const { daemonEnvironment } = await import("./idunn-entry.mjs");
|
|
const release = mkdtempSync(path.join(os.tmpdir(), "vili-entry-"));
|
|
try {
|
|
mkdirSync(path.join(release, "vili", "scripts"), { recursive: true });
|
|
mkdirSync(path.join(release, "vili", "deployment", "idunn"), { recursive: true });
|
|
writeFileSync(path.join(release, "vili", "deployment", "idunn", "requirements-build.lock"), "build lock bytes\n");
|
|
writeFileSync(path.join(release, "vili", "deployment", "idunn", "requirements.lock"), "runtime lock bytes\n");
|
|
const entryDir = path.join(release, "vili", "scripts");
|
|
assert.equal(daemonEnvironment("127.0.0.1:65535", entryDir).VILI_DAEMON_PORT, "65535", "the top of the port range is accepted");
|
|
for (const bind of ["127.0.0.1:18861", "http://127.0.0.1:18861"]) {
|
|
const derived = daemonEnvironment(bind, entryDir);
|
|
assert.equal(derived.VILI_DAEMON_HOST, "127.0.0.1");
|
|
assert.equal(derived.VILI_DAEMON_PORT, "18861");
|
|
assert.equal(derived.VILI_PYTHON, path.join(release, "python", "bin", "python3.12"));
|
|
assert.equal(derived.VILI_PYTHON_ENV, `locks-sha256:${createHash("sha256").update("build lock bytes\nruntime lock bytes\n").digest("hex")}`);
|
|
}
|
|
} finally {
|
|
rmSync(release, { recursive: true, force: true });
|
|
}
|
|
});
|