Fas 1a punkt 3+4: worker-härdning (DLQ, larm, multi-instance), AAMOS training export

This commit is contained in:
Sven (AAMOS AI)
2026-08-05 23:32:18 +07:00
parent d1455fb581
commit b2b6601f54
15 changed files with 9355 additions and 32 deletions
+3
View File
@@ -10,6 +10,7 @@ import { SubscriptionsPage } from "./pages/Subscriptions.js";
import { AuditLogsPage } from "./pages/AuditLogs.js";
import { AnalyticsPage } from "./pages/Analytics.js";
import { ReleaseGatesPage } from "./pages/ReleaseGates.js";
import { WorkersPage } from "./pages/Workers.js";
export function App() {
const [authed, setAuthed] = useState(hasToken());
@@ -32,6 +33,7 @@ export function App() {
<NavLink to="/audit">Audit logs</NavLink>
<NavLink to="/analytics">Analytics</NavLink>
<NavLink to="/release-gates">Release gates</NavLink>
<NavLink to="/workers">Workers</NavLink>
<button
onClick={() => {
setToken(null);
@@ -51,6 +53,7 @@ export function App() {
<Route path="/audit" element={<AuditLogsPage />} />
<Route path="/analytics" element={<AnalyticsPage />} />
<Route path="/release-gates" element={<ReleaseGatesPage />} />
<Route path="/workers" element={<WorkersPage />} />
<Route path="*" element={<Navigate to="/" replace />} />
</Routes>
</main>
+204
View File
@@ -0,0 +1,204 @@
import { useEffect, useState } from "react";
import { api } from "../api.js";
interface JobCounts {
wait: number;
active: number;
delayed: number;
completed: number;
failed: number;
paused?: number;
}
interface WorkerInfo {
id: string;
addr: string;
concurrency: number;
lastSeenAt: number;
}
interface SchedulerInfo {
id: string;
pattern?: string;
every?: number;
}
interface HealthResponse {
ok: boolean;
queues: {
jobs: JobCounts;
deadLetter: JobCounts;
};
workers: WorkerInfo[];
schedulers: SchedulerInfo[];
}
interface DlqJob {
id: string;
name: string;
data: unknown;
failedReason?: string;
attemptsMade: number;
timestamp: number;
}
export function WorkersPage() {
const [health, setHealth] = useState<HealthResponse | null>(null);
const [dlq, setDlq] = useState<DlqJob[]>([]);
const [error, setError] = useState<string | null>(null);
const [loading, setLoading] = useState(false);
const [actionId, setActionId] = useState<string | null>(null);
const load = async () => {
setLoading(true);
try {
const [h, d] = await Promise.all([
api<HealthResponse>("/admin/v1/workers/health"),
api<{ jobs: DlqJob[] }>("/admin/v1/workers/dead-letter?limit=50"),
]);
setHealth(h);
setDlq(d.jobs);
setError(null);
} catch (e) {
setError(String((e as Error).message));
} finally {
setLoading(false);
}
};
useEffect(() => {
load();
const t = setInterval(load, 10_000);
return () => clearInterval(t);
}, []);
const retry = async (id: string) => {
setActionId(id);
try {
await api(`/admin/v1/workers/dead-letter/${id}/retry`, { method: "POST" });
await load();
} catch (e) {
setError(String((e as Error).message));
} finally {
setActionId(null);
}
};
const discard = async (id: string) => {
if (!confirm("Ta bort jobbet permanent?")) return;
setActionId(id);
try {
await api(`/admin/v1/workers/dead-letter/${id}/discard`, { method: "POST" });
await load();
} catch (e) {
setError(String((e as Error).message));
} finally {
setActionId(null);
}
};
return (
<div className="page">
<h2>Workers & köer</h2>
{error && <div className="alert err">{error}</div>}
{loading && !health && <div>Laddar </div>}
{health && (
<>
<section className="card">
<h3>Jobbkö</h3>
<div className="metrics">
<Metric label="Väntar" value={health.queues.jobs.wait} />
<Metric label="Aktiva" value={health.queues.jobs.active} />
<Metric label="Fördröjda" value={health.queues.jobs.delayed} />
<Metric label="Misslyckade" value={health.queues.jobs.failed} />
<Metric label="Klara" value={health.queues.jobs.completed} />
</div>
</section>
<section className="card">
<h3>Dead letter ({health.queues.deadLetter.wait + health.queues.deadLetter.failed})</h3>
{dlq.length === 0 ? (
<p>Inga jobb i dead letter.</p>
) : (
<table>
<thead>
<tr>
<th>Jobb</th>
<th>Fel</th>
<th>Försök</th>
<th>Tid</th>
<th>Åtgärder</th>
</tr>
</thead>
<tbody>
{dlq.map((job) => (
<tr key={job.id}>
<td>{job.name}</td>
<td className="ellipsis" title={job.failedReason}>
{job.failedReason ?? ""}
</td>
<td>{job.attemptsMade}</td>
<td>{new Date(job.timestamp).toLocaleString("sv-SE")}</td>
<td>
<button
onClick={() => retry(job.id)}
disabled={actionId === job.id}
className="btn small"
>
Kör igen
</button>{" "}
<button
onClick={() => discard(job.id)}
disabled={actionId === job.id}
className="btn small danger"
>
Kasta
</button>
</td>
</tr>
))}
</tbody>
</table>
)}
</section>
<section className="card">
<h3>Aktiva worker-instanser</h3>
{health.workers.length === 0 ? (
<p className="err">Inga aktiva workers!</p>
) : (
<ul>
{health.workers.map((w) => (
<li key={w.id}>
{w.id} @ {w.addr} (concurrency {w.concurrency})
</li>
))}
</ul>
)}
</section>
<section className="card">
<h3>Scheduler</h3>
<ul>
{health.schedulers.map((s) => (
<li key={s.id}>
{s.id}: {s.pattern ?? `every ${s.every}ms`}
</li>
))}
</ul>
</section>
</>
)}
</div>
);
}
function Metric({ label, value }: { label: string; value: number }) {
return (
<div className="metric">
<div className="value">{value}</div>
<div className="label">{label}</div>
</div>
);
}
+48
View File
@@ -124,3 +124,51 @@ input, select {
.stat .value { font-size: 1.6rem; font-weight: 700; }
.stat .label { color: var(--muted); font-size: 0.8rem; }
.row { display: flex; gap: 0.5rem; align-items: center; flex-wrap: wrap; }
.card {
background: var(--panel);
border: 1px solid var(--border);
border-radius: 12px;
padding: 1.25rem;
margin-bottom: 1.25rem;
}
.card h3 { margin-top: 0; }
.metrics { display: flex; gap: 1rem; flex-wrap: wrap; }
.metric {
background: var(--bg);
border: 1px solid var(--border);
border-radius: 10px;
padding: 0.75rem 1rem;
min-width: 90px;
}
.metric .value { font-size: 1.4rem; font-weight: 700; }
.metric .label { color: var(--muted); font-size: 0.8rem; }
.btn {
background: var(--brand);
color: #fff;
border: none;
padding: 0.45rem 0.9rem;
border-radius: 8px;
cursor: pointer;
font-size: 0.9rem;
}
.btn:hover { background: var(--brand-dark); }
.btn:disabled { opacity: 0.6; cursor: not-allowed; }
.btn.small { padding: 0.25rem 0.55rem; font-size: 0.8rem; }
.btn.danger { background: var(--danger); }
.alert {
padding: 0.75rem 1rem;
border-radius: 8px;
margin-bottom: 1rem;
}
.alert.err { background: #fee2e2; color: var(--danger); }
.ellipsis {
max-width: 280px;
overflow: hidden;
text-overflow: ellipsis;
white-space: nowrap;
}
+90
View File
@@ -0,0 +1,90 @@
import type { FastifyInstance } from "fastify";
import { DEAD_LETTER_QUEUE_NAME, JOB_QUEUE_NAME } from "@app/shared-types";
import { Queue, Job } from "bullmq";
interface JobSummary {
id: string | undefined;
name: string;
data: unknown;
failedReason: string | undefined;
attemptsMade: number;
timestamp: number;
}
export async function adminWorkersRoutes(app: FastifyInstance) {
const admin = { preHandler: [app.requireAdmin] };
const dlq = new Queue(DEAD_LETTER_QUEUE_NAME, { connection: app.redis });
app.addHook("onClose", async () => {
await dlq.close();
});
app.get("/admin/v1/workers/health", admin, async (_req, reply) => {
const queue = new Queue(JOB_QUEUE_NAME, { connection: app.redis });
const [jobCounts, dlqCounts, workers, schedulers] = await Promise.all([
queue.getJobCounts("wait", "active", "delayed", "completed", "failed"),
dlq.getJobCounts("wait", "active", "completed", "failed"),
queue.getWorkers(),
queue.getJobSchedulers(),
]);
await queue.close();
return reply.send({
ok: true,
queues: {
jobs: jobCounts,
deadLetter: dlqCounts,
},
workers: workers.map((w) => ({
id: w.id,
addr: w.addr,
concurrency: w.concurrency,
// eslint-disable-next-line @typescript-eslint/no-explicit-any
lastSeenAt: (w as any).ts,
})),
schedulers: schedulers.map((s) => ({ id: s.id, pattern: s.pattern, every: s.every })),
});
});
app.get("/admin/v1/workers/dead-letter", admin, async (req, reply) => {
const limit = Math.min(Number((req.query as { limit?: string }).limit ?? 50), 200);
const jobs = await dlq.getJobs(["wait", "failed"], 0, limit - 1, true);
const summary: JobSummary[] = jobs.map((job) => ({
id: job.id,
name: job.name,
data: job.data,
failedReason: job.failedReason,
attemptsMade: job.attemptsMade,
timestamp: job.timestamp,
}));
return reply.send({ jobs: summary });
});
app.post("/admin/v1/workers/dead-letter/:id/retry", admin, async (req, reply) => {
const { id } = req.params as { id: string };
const job = await Job.fromId(dlq, id);
if (!job) {
return reply.status(404).send({ error: { code: "NOT_FOUND", message: "Jobbet finns inte i dead letter queue." } });
}
const mainQueue = new Queue(JOB_QUEUE_NAME, { connection: app.redis });
await mainQueue.add(job.name, job.data, {
attempts: 3,
backoff: { type: "exponential", delay: 2_000 },
});
await job.remove();
await mainQueue.close();
return reply.send({ ok: true, message: "Jobbet har lagts tillbaka i kön." });
});
app.post("/admin/v1/workers/dead-letter/:id/discard", admin, async (req, reply) => {
const { id } = req.params as { id: string };
const job = await Job.fromId(dlq, id);
if (!job) {
return reply.status(404).send({ error: { code: "NOT_FOUND", message: "Jobbet finns inte i dead letter queue." } });
}
await job.remove();
return reply.send({ ok: true, message: "Jobbet har tagits bort från dead letter queue." });
});
}
+2
View File
@@ -23,6 +23,7 @@ import { communityRoutes } from "./routes/community.js";
import { adminRoutes } from "./routes/admin.js";
import { adminAnalyticsRoutes } from "./routes/admin-analytics.js";
import { adminReleaseGateRoutes } from "./routes/admin-release-gates.js";
import { adminWorkersRoutes } from "./routes/admin-workers.js";
import { analyticsRoutes } from "./routes/analytics.js";
declare module "fastify" {
@@ -85,6 +86,7 @@ export async function buildServer(config: AppConfig) {
await app.register(adminRoutes);
await app.register(adminAnalyticsRoutes);
await app.register(adminReleaseGateRoutes);
await app.register(adminWorkersRoutes);
await app.register(analyticsRoutes);
return app;
+114 -20
View File
@@ -1,6 +1,7 @@
import "dotenv/config";
import { Queue, Worker, type Job } from "bullmq";
import IORedis from "ioredis";
import { createServer } from "node:http";
import { createContext } from "./context.js";
import { processScanJob } from "./processors/scans.js";
import { processModerateRecipe } from "./processors/moderation.js";
@@ -17,19 +18,29 @@ import {
processTrainingExport,
} from "./processors/maintenance.js";
import {
DEAD_LETTER_QUEUE_NAME,
JOB_QUEUE_NAME as QUEUE_NAME,
WORKER_DEFAULT_ATTEMPTS,
WORKER_DEFAULT_BACKOFF_MS,
} from "@app/shared-types";
/**
* workern. Konsumerar kön "<slug>-jobs" (spec §50):
* App → signed S3 upload → API job → worker → AAMOS → result → app
* workern. Konsumerar kön "<slug>-jobs" (spec §50).
*
* Jobbnamn = jobbtyp (spec §54). Deterministiska jobb rör aldrig AAMOS;
* AI-jobb går alltid via de typade kontrakten i @app/ai-contracts.
* Flera instanser kan köra samtidigt mot samma Redis- och DB-anslutning
* utan kodändring (BullMQ hanterar fördelning). Varje instans exponerar
* en liten HTTP-healthcheck så att lastbalansering kan se om den lever.
*/
import { JOB_QUEUE_NAME as QUEUE_NAME } from "@app/shared-types";
const connection = new IORedis(process.env.REDIS_URL ?? "redis://localhost:6379", {
maxRetriesPerRequest: null,
});
const ctx = createContext();
const instanceId = process.env.WORKER_INSTANCE_ID ?? `worker-${process.pid}`;
const deadLetterQueue = new Queue(DEAD_LETTER_QUEUE_NAME, { connection });
const queue = new Queue(QUEUE_NAME, { connection });
const worker = new Worker(
QUEUE_NAME,
@@ -116,13 +127,48 @@ const worker = new Worker(
{
connection,
concurrency: Number(process.env.WORKER_CONCURRENCY ?? 5),
maxStalledCount: 2,
stalledInterval: 30_000,
limiter: {
max: Number(process.env.WORKER_RATE_LIMIT_MAX ?? 60),
duration: Number(process.env.WORKER_RATE_LIMIT_DURATION_MS ?? 60_000),
},
},
);
worker.on("completed", (job) => log(`${job.name} (${job.id})`));
worker.on("failed", (job, err) => {
console.error(`${job?.name} (${job?.id}): ${err.message}`);
void alertWebhook(`Worker-jobb misslyckades: ${job?.name} (${job?.id}): ${err.message}`);
worker.on("failed", async (job, err) => {
const jobType = (job?.data?.jobType as string) ?? job?.name ?? "unknown";
const attempts = job?.opts?.attempts ?? WORKER_DEFAULT_ATTEMPTS;
const attemptsMade = job?.attemptsMade ?? 0;
console.error(`${jobType} (${job?.id}): ${err.message} (attempt ${attemptsMade}/${attempts})`);
if (job && attemptsMade >= attempts) {
try {
await deadLetterQueue.add(
job.name,
{
...job.data,
_deadLetteredAt: new Date().toISOString(),
_instanceId: instanceId,
_failureReason: err.message,
_failedAt: new Date().toISOString(),
},
{
jobId: `${instanceId}:${job.id}`,
removeOnComplete: { age: 30 * 24 * 60 * 60 }, // 30 dagar
removeOnFail: { age: 90 * 24 * 60 * 60 }, // 90 dagar
},
);
log(`→ dead letter: ${jobType} (${job.id})`);
} catch (dlqErr) {
console.error("Kunde inte skriva till dead letter queue:", dlqErr);
}
}
await alertWebhook(
`Worker-jobb misslyckades: ${jobType} (${job?.id}): ${err.message} (attempt ${attemptsMade}/${attempts})`,
);
});
/** Larm till valfri webhook (Slack/Discord/Teams …) fire-and-forget. */
@@ -142,50 +188,98 @@ async function alertWebhook(text: string): Promise<void> {
}
// --- Återkommande jobb via BullMQ Job Schedulers (spec §54) ---
const queue = new Queue(QUEUE_NAME, { connection });
async function registerRepeatableJobs() {
const baseOpts = {
attempts: WORKER_DEFAULT_ATTEMPTS,
backoff: { type: "exponential" as const, delay: WORKER_DEFAULT_BACKOFF_MS },
removeOnComplete: { count: 100 },
removeOnFail: { count: 100 },
};
await queue.upsertJobScheduler(
"scheduler-outbox",
{ every: 30_000 },
{ name: "PUBLISH_OUTBOX", data: { jobType: "PUBLISH_OUTBOX" } },
{ name: "PUBLISH_OUTBOX", data: { jobType: "PUBLISH_OUTBOX" }, opts: baseOpts },
);
await queue.upsertJobScheduler(
"scheduler-expiry",
{ pattern: "0 7 * * *", tz: "Europe/Stockholm" },
{ name: "SEND_EXPIRY_NOTIFICATION", data: { jobType: "SEND_EXPIRY_NOTIFICATION" } },
{ name: "SEND_EXPIRY_NOTIFICATION", data: { jobType: "SEND_EXPIRY_NOTIFICATION" }, opts: baseOpts },
);
await queue.upsertJobScheduler(
"scheduler-memory",
{ pattern: "30 3 * * *", tz: "Europe/Stockholm" },
{ name: "UPDATE_USER_MEMORY", data: { jobType: "UPDATE_USER_MEMORY" } },
{ name: "UPDATE_USER_MEMORY", data: { jobType: "UPDATE_USER_MEMORY" }, opts: baseOpts },
);
await queue.upsertJobScheduler(
"scheduler-subs",
{ pattern: "15 4 * * *", tz: "Europe/Stockholm" },
{ name: "VERIFY_SUBSCRIPTION", data: { jobType: "VERIFY_SUBSCRIPTION" } },
{ name: "VERIFY_SUBSCRIPTION", data: { jobType: "VERIFY_SUBSCRIPTION" }, opts: baseOpts },
);
await queue.upsertJobScheduler(
"scheduler-retention",
{ pattern: "45 4 * * *", tz: "Europe/Stockholm" },
{ name: "RUN_RETENTION", data: { jobType: "RUN_RETENTION" } },
{ name: "RUN_RETENTION", data: { jobType: "RUN_RETENTION" }, opts: baseOpts },
);
await queue.upsertJobScheduler(
"scheduler-training",
{ pattern: "45 2 * * 0", tz: "Europe/Stockholm" },
{ name: "BUILD_TRAINING_SAMPLE", data: { jobType: "BUILD_TRAINING_SAMPLE" } },
{ name: "BUILD_TRAINING_SAMPLE", data: { jobType: "BUILD_TRAINING_SAMPLE" }, opts: baseOpts },
);
}
registerRepeatableJobs()
.then(() => log("Worker igång. Väntar på jobb …"))
.catch((err) => {
console.error("Kunde inte registrera återkommande jobb:", err);
// --- Minimal healthcheck-server så att flera instanser kan övervakas ---
const healthPort = Number(process.env.WORKER_HEALTH_PORT ?? 4001);
let processedCount = 0;
let failedCount = 0;
worker.on("completed", () => processedCount++);
worker.on("failed", () => failedCount++);
const healthServer = createServer(async (req, res) => {
if (req.url === "/healthz") {
const [queueCount, dlqCount, workers] = await Promise.all([
queue.getJobCounts("wait", "active", "delayed", "completed", "failed"),
deadLetterQueue.getJobCounts("wait", "active", "completed", "failed"),
queue.getWorkers(),
]);
res.writeHead(200, { "content-type": "application/json" });
res.end(
JSON.stringify({
ok: true,
instanceId,
uptimeSeconds: Math.floor(process.uptime()),
queue: queueCount,
deadLetter: dlqCount,
activeWorkers: workers.length,
processedSinceStart: processedCount,
failedSinceStart: failedCount,
}),
);
return;
}
res.writeHead(404);
res.end("not found");
});
async function start() {
await registerRepeatableJobs();
log("Worker igång. Väntar på jobb …");
healthServer.listen(healthPort, () => {
log(`Healthcheck på :${healthPort}`);
});
}
start().catch((err) => {
console.error("Kunde inte starta worker:", err);
process.exit(1);
});
const shutdown = async () => {
log("Stänger ner …");
healthServer.close();
await worker.close();
await queue.close();
await deadLetterQueue.close();
connection.disconnect();
await ctx.close();
process.exit(0);
@@ -194,5 +288,5 @@ process.on("SIGTERM", () => void shutdown());
process.on("SIGINT", () => void shutdown());
function log(msg: string) {
console.log(`[worker ${new Date().toISOString()}] ${msg}`);
console.log(`[worker ${instanceId} ${new Date().toISOString()}] ${msg}`);
}
+39 -10
View File
@@ -273,7 +273,7 @@ export async function processMemorySync(ctx: WorkerContext): Promise<number> {
return updates;
}
/** BUILD_TRAINING_SAMPLE (spec §33): exportera korrigeringar MED samtycke. */
/** BUILD_TRAINING_SAMPLE (spec §33): exportera korrigeringar MED samtycke till AAMOS. */
export async function processTrainingExport(ctx: WorkerContext): Promise<number> {
const corrections = await ctx.db
.select()
@@ -281,19 +281,48 @@ export async function processTrainingExport(ctx: WorkerContext): Promise<number>
.where(isNull(schema.aiCorrections.exportedToTraining))
.limit(100);
let exported = 0;
for (const correction of corrections) {
const snapshot = correction.consentSnapshot as Record<string, string>;
// Endast korrigeringar där anonymiserad förbättring var beviljad vid tillfället.
if (snapshot.anonymized_improvement !== "granted") continue;
// Här skulle exporten till AAMOS training-pipeline ske (avidentifierad).
const eligible = corrections.filter((c) => {
const snapshot = c.consentSnapshot as Record<string, string>;
return snapshot.anonymized_improvement === "granted";
});
if (eligible.length === 0) return 0;
const result = await ctx.aamos.runTask(
"EXPORT_TRAINING_SAMPLE",
{
marketLocale: "sv-SE",
samples: eligible.map((c) => ({
taskType: c.taskType,
aiOutput: c.aiOutput as Record<string, unknown>,
userCorrection: c.userCorrection as Record<string, unknown>,
modelVersion: c.modelVersion ?? null,
promptVersion: c.promptVersion ?? null,
})),
},
{
correlationId: `training-export-${Date.now()}`,
consentFlags: {
personalization: false,
anonymizedImprovement: true,
imageTraining: false,
},
},
);
if (result.status !== "ok" || !result.output) {
throw new Error(`AAMOS training export failed: ${result.error ?? "unknown"}`);
}
const batchId = result.output.batchId;
for (const correction of eligible) {
await ctx.db
.update(schema.aiCorrections)
.set({ exportedToTraining: new Date() })
.set({ exportedToTraining: new Date(), trainingBatchId: batchId })
.where(eq(schema.aiCorrections.id, correction.id));
exported++;
}
return exported;
return eligible.length;
}
/** VERIFY_SUBSCRIPTION: flagga prenumerationer som passerat expiry. */