fix(server): refine error handling and PM2 connection logic

- Reduced the maximum restarts for PM2 processes from 20 to 8 and added a restart delay of 2000ms for better stability.
- Improved error handling during server startup and PM2 connection, ensuring clearer logging and graceful exits on failure.
- Introduced a utility function to filter out PM2 metadata from environment variables, preventing conflicts during process management.

.gitea/workflows/deploy.yaml

@@ -0,0 +1,33 @@
+name: Deploy to Raspberry Pi
+
+on:
+  push:
+    branches: [dev]
+
+jobs:
+  deploy:
+    runs-on: home   # must match a label on your act_runner
+    steps:
+      - name: Deploy
+        run: |
+          set -euo pipefail
+
+          # act_runner uses bash --noprofile --norc; load nvm/pnpm explicitly
+          export NVM_DIR="/home/nsrb/.nvm"
+          export PNPM_HOME="/home/nsrb/.local/share/pnpm"
+          # shellcheck disable=SC1091
+          [ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"
+          export PATH="$PNPM_HOME/bin:$PATH"
+
+          APP_DIR=/home/nsrb/apps/jerapah-flow
+          cd "$APP_DIR"
+
+          git fetch origin dev
+          git checkout dev
+          git pull --ff-only origin dev
+
+          pnpm install --frozen-lockfile
+          pnpm build
+
+          pm2 startOrReload ecosystem.config.cjs --update-env
+          pm2 save

README.md

@@ -105,8 +105,9 @@ Optional `script.meta.reads = "ctx"` documents expression hosts. `meta.input` /
 | `pnpm dev:server` | Monolith API/runner only |
 | `pnpm dev:web` | UI only (proxies `/api` → :8700, `/ops` → :8600) |
 | `pnpm build` | Production UI build |
-| `pnpm start` | Monolith: API + worker + built UI |
+| `pnpm start` | Monolith: API + worker + built UI (serves `dist` on :8700) |
 | `pnpm start:control` | Control plane only (migrates, manages PM2 children) |
+| `pnpm start:web` | Production UI on :8500 (`dist` + proxies to control/HTTP) |
 | `pnpm start:api` | HTTP API + cron enqueue (`JFLOW_ROLE=api`) |
 | `pnpm start:worker` | BullMQ worker only |
 | `pnpm migrate` | Apply SQLite migrations |
@@ -140,24 +141,45 @@ Desired state is stored in `packages/server/data/control-state.json` (generation
 | `JFLOW_ROLE` | `all` | `all` (HTTP + cron + worker), `api`, or `worker`. Prefer `pnpm start:api` / `start:worker` under control. |
 | `JFLOW_CONFIG_GENERATION` | `1` | Set by control/PM2 so children report config generation in heartbeats. |
 | `JFLOW_CONTROL_PORT` | `8600` | Control ops API port. |
+| `JFLOW_UI_PORT` | `8500` | Production UI server (`web-server.js`) port. |
+| `JFLOW_HTTP_PORT` | `8700` | HTTP API port (PM2 children / UI proxy target). |
 | `JFLOW_LOG_LEVEL` | `debug` | Pino level |
 | `JFLOW_RETENTION_DAYS` | `30` | Run history prune |
-| `JFLOW_CORS_ORIGIN` | `http://localhost:8500` | Vite origin in dev |
-| `PORT` | `8700` | HTTP API port |
-| `NODE_ENV` | — | Set `production` for secure cookies |
+| `JFLOW_CORS_ORIGIN` | `http://localhost:8500` | Browser origin (Vite in dev, UI server in prod) |
+| `PORT` | `8700` | HTTP API port (alias; prefer `JFLOW_HTTP_PORT` under control) |
+| `NODE_ENV` | — | Set `production` for secure cookies (unless overridden) |
+| `COOKIE_SECURE` | (from `NODE_ENV`) | `true`/`false` — force Secure cookie flag. Use `false` for plain HTTP LAN access (`http://192.168.x.x`) |
 
 Workflow runs are **queued** via BullMQ. HTTP and manual triggers return `202 { runId, status: "queued" }` immediately; poll `GET /api/runs/:id` for progress (`queued` → `running` → `success` \| `failed`). Cron remains an in-process producer that enqueues jobs on each tick.
 
 ## Production
 
+Control-plane topology (same ports as `pnpm dev:pm2`):
+
+| Process | Port | Role |
+|---|---|---|
+| `jflow-web` | **8500** | Built UI + proxies `/api` → :8700, `/ops` + `/api/auth` → :8600 |
+| `jflow-control` | **8600** | Migrations, Ops API, starts/stops PM2 HTTP + workers |
+| `jflow-http` | **8700** | API + cron enqueue (managed by control) |
+| `jflow-worker` | — | BullMQ workers (managed by control) |
+
 ```bash
 pnpm install
 pnpm build
 # Redis must be reachable at REDIS_URL (set REDIS_PASS if Redis requires AUTH)
-# Recommended: run control (migrates + manages PM2 HTTP/workers)
-JFLOW_JWT_SECRET=... JFLOW_SECRETS_KEY=... REDIS_URL=redis://127.0.0.1:6379 REDIS_PASS=... NODE_ENV=production pnpm start:control
-# Or monolith (dev-style):
-# ... pnpm start
+# Put secrets in .env (JFLOW_JWT_SECRET, JFLOW_SECRETS_KEY, REDIS_URL, …)
+# Use in-tree PM2 6.x (same module control.js requires). A global `pm2` 7.x
+# against a 6.x daemon pegs CPU even when ls shows only 2 fork instances.
+pnpm start:pm2
+# UI: http://localhost:8500
+# If you already mixed versions: pnpm pm2 -- kill && pnpm start:pm2
 ```
 
-With control, serve the built UI from Vite preview, a reverse proxy, or set `JFLOW_SERVE_UI=1` on the HTTP process.
+Or without the ecosystem file:
+
+```bash
+NODE_ENV=production pnpm start:control   # :8600 + PM2 children
+NODE_ENV=production pnpm start:web       # :8500
+```
+
+Monolith (no Ops stop/scale): `pnpm build && pnpm start` serves the UI from the API process on :8700. Optional `JFLOW_SERVE_UI=1` on `start:api` does the same when you run HTTP alone — do **not** use that under control-plane mode (stopping HTTP would take down the UI).

ecosystem.config.cjs

@@ -1,7 +1,12 @@
 /**
- * Production PM2 ecosystem (monolith runner).
- * Use for deployed/single-process starts. For local multi-process Ops UI, use
- * `pnpm dev:pm2` → packages/server/ecosystem.dev.cjs / control.js instead.
+ * Production PM2 ecosystem (control plane + UI).
+ * Starts always-on processes only; HTTP (:8700) and workers are owned by
+ * control.js via PM2 (same as `pnpm dev:pm2`).
+ *
+ * Use `pnpm start:pm2` (in-tree PM2 6.x). Do not use a global `pm2` 7.x —
+ * a CLI/daemon version mismatch pegs CPU even with instances: 1.
+ *
+ * Prerequisites: `pnpm build` (packages/web/dist), Redis, .env secrets.
  */
 const fs = require("fs");
 const path = require("path");
@@ -28,20 +33,42 @@ function loadEnv(file) {
 }
 
 const root = __dirname;
+const env = {
+  NODE_ENV: "production",
+  ...loadEnv(path.join(root, ".env")),
+};
 
 module.exports = {
   apps: [
     {
-      name: "jerapah-flow",
+      name: "jflow-control",
       cwd: root,
-      script: "packages/server/runner.js",
-      interpreter: "node",
+      script: "packages/server/control.js",
+      interpreter: process.execPath,
       instances: 1,
+      exec_mode: "fork",
+      autorestart: true,
+      max_restarts: 8,
+      restart_delay: 2000,
+      env: {
+        ...env,
+        JFLOW_CONTROL_PORT: env.JFLOW_CONTROL_PORT ?? "8600",
+      },
+    },
+    {
+      name: "jflow-web",
+      cwd: root,
+      script: "packages/server/web-server.js",
+      interpreter: process.execPath,
+      instances: 1,
+      exec_mode: "fork",
       autorestart: true,
       max_restarts: 20,
       env: {
-        NODE_ENV: "production",
-        ...loadEnv(path.join(root, ".env")),
+        ...env,
+        JFLOW_UI_PORT: env.JFLOW_UI_PORT ?? "8500",
+        JFLOW_CONTROL_PORT: env.JFLOW_CONTROL_PORT ?? "8600",
+        JFLOW_HTTP_PORT: env.JFLOW_HTTP_PORT ?? "8700",
       },
     },
   ],

package.json

@@ -14,6 +14,9 @@
     "start:api": "pnpm --filter @jerapah-flow/server start:api",
     "start:worker": "pnpm --filter @jerapah-flow/server start:worker",
     "start:control": "pnpm --filter @jerapah-flow/server start:control",
+    "start:web": "pnpm --filter @jerapah-flow/server start:web",
+    "pm2": "node scripts/pm2.mjs",
+    "start:pm2": "node scripts/pm2.mjs start ecosystem.config.cjs",
     "migrate": "pnpm --filter @jerapah-flow/server migrate",
     "test": "pnpm --filter @jerapah-flow/shared test && pnpm --filter @jerapah-flow/web test",
     "lint": "pnpm --filter @jerapah-flow/web lint"

packages/server/control.js

@@ -64,13 +64,6 @@ try {
   process.exit(1);
 }
 
-try {
-  await connectPm2();
-} catch (err) {
-  log.error({ err }, "failed to connect to PM2 — is pm2 installed?");
-  process.exit(1);
-}
-
 async function applyDesiredState() {
   const state = readControlState();
   await ensureHttp({
@@ -98,8 +91,6 @@ async function applyDesiredState() {
   );
 }
 
-await applyDesiredState();
-
 const server = fastify({ loggerInstance: log });
 await server.register(cookie);
 await server.register(jwt, {
@@ -484,12 +475,24 @@ process.on("SIGINT", shutdown);
 process.on("SIGTERM", shutdown);
 
 const port = Number(process.env.JFLOW_CONTROL_PORT ?? process.env.PORT ?? 8600);
-server
-  .listen({ host: "0.0.0.0", port })
-  .then(() => {
-    log.info(`Control is running on port ${port}`);
-  })
-  .catch((err) => {
-    log.error({ err }, "failed to start control");
-    process.exit(1);
-  });
+try {
+  await server.listen({ host: "0.0.0.0", port });
+  log.info(`Control is running on port ${port}`);
+} catch (err) {
+  log.error({ err }, "failed to start control");
+  process.exit(1);
+}
+
+try {
+  await connectPm2();
+} catch (err) {
+  log.error({ err }, "failed to connect to PM2 — is pm2 installed?");
+  process.exit(1);
+}
+
+try {
+  await applyDesiredState();
+} catch (err) {
+  log.error({ err }, "failed to apply desired state");
+  process.exit(1);
+}

packages/server/ops-proxy.js

@@ -0,0 +1,97 @@
+const HOP_BY_HOP = new Set([
+  "connection",
+  "keep-alive",
+  "proxy-authenticate",
+  "proxy-authorization",
+  "te",
+  "trailer",
+  "transfer-encoding",
+  "upgrade",
+  "host",
+  "content-length",
+]);
+
+const REPLY_SKIP = new Set([
+  "connection",
+  "keep-alive",
+  "transfer-encoding",
+  "content-encoding",
+  "content-length",
+]);
+
+/**
+ * Control origin used when the UI or HTTP process proxies to control.
+ * @returns {string}
+ */
+export function controlOrigin() {
+  const explicit = process.env.JFLOW_CONTROL_URL?.trim();
+  if (explicit) return explicit.replace(/\/$/, "");
+  const port = Number(process.env.JFLOW_CONTROL_PORT ?? 8600);
+  return `http://127.0.0.1:${port}`;
+}
+
+/**
+ * HTTP API origin (workflow triggers + REST).
+ * @returns {string}
+ */
+export function httpOrigin() {
+  const explicit = process.env.JFLOW_HTTP_URL?.trim();
+  if (explicit) return explicit.replace(/\/$/, "");
+  const port = Number(process.env.JFLOW_HTTP_PORT ?? process.env.PORT ?? 8700);
+  return `http://127.0.0.1:${port}`;
+}
+
+/**
+ * Forward the incoming request to `origin`, preserving path + query.
+ * @param {import("fastify").FastifyRequest} req
+ * @param {import("fastify").FastifyReply} reply
+ * @param {string} origin
+ * @param {{ unreachableMessage?: string }} [opts]
+ */
+export async function proxyToOrigin(req, reply, origin, opts = {}) {
+  const target = `${origin.replace(/\/$/, "")}${req.raw.url ?? "/"}`;
+  const headers = {};
+  for (const [key, value] of Object.entries(req.headers)) {
+    if (value == null || HOP_BY_HOP.has(key.toLowerCase())) continue;
+    headers[key] = Array.isArray(value) ? value.join(", ") : String(value);
+  }
+
+  const method = req.method.toUpperCase();
+  const hasBody = method !== "GET" && method !== "HEAD";
+  let body;
+  if (hasBody) {
+    if (Buffer.isBuffer(req.body)) body = req.body;
+    else if (typeof req.body === "string") body = req.body;
+    else if (req.body != null) {
+      body = JSON.stringify(req.body);
+      if (!headers["content-type"]) headers["content-type"] = "application/json";
+    }
+  }
+
+  let res;
+  try {
+    res = await fetch(target, { method, headers, body });
+  } catch (err) {
+    const message = opts.unreachableMessage ?? "upstream unreachable";
+    req.log.warn({ err, target }, `proxy: ${message}`);
+    return reply.code(502).send({ error: message });
+  }
+
+  reply.code(res.status);
+  res.headers.forEach((value, key) => {
+    if (REPLY_SKIP.has(key.toLowerCase())) return;
+    reply.header(key, value);
+  });
+  return reply.send(Buffer.from(await res.arrayBuffer()));
+}
+
+/**
+ * Forward `/ops/*` to the control plane (same-origin UI in production).
+ * @param {import("fastify").FastifyRequest} req
+ * @param {import("fastify").FastifyReply} reply
+ */
+export async function proxyOpsToControl(req, reply) {
+  return proxyToOrigin(req, reply, controlOrigin(), {
+    unreachableMessage: "control plane unreachable",
+  });
+}

packages/server/package.json

@@ -11,6 +11,7 @@
     "start:api": "node server.js",
     "start:worker": "node worker.js",
     "start:control": "node control.js",
+    "start:web": "node web-server.js",
     "migrate": "node -e \"import('./db.js').then((m) => m.migrate().then(() => process.exit(0)))\"",
     "test:plugins": "JFLOW_PLUGINS_DIR=./data/plugins-smoke-test JFLOW_DB_PATH=./data/plugins-smoke.db node test/plugins-smoke.js",
     "test:workflow-history": "JFLOW_WORKFLOWS_DIR=./data/workflow-history-smoke JFLOW_DB_PATH=./data/workflow-history-smoke.db node test/workflow-history-smoke.js",
@@ -41,7 +42,7 @@
     "nodemailer": "^9.0.5",
     "pino": "^10.3.1",
     "pino-roll": "^4.0.0",
-    "pm2": "^6.0.13",
+    "pm2": "6.0.14",
     "rss-parser": "^3.13.0",
     "ssh2-sftp-client": "^12.1.1",
     "webdav": "^5.10.0",

packages/server/pm2-bridge.js

@@ -164,13 +164,57 @@ export async function restartPm2Process(pmId) {
   return { name: proc.name, pmId: id };
 }
 
+/**
+ * PM2 injects these into process.env of a managed app. Spreading them into
+ * `pm2.start({ env })` overwrites `name` / `pm_exec_path` so God restarts
+ * jflow-control instead of launching http/worker (EADDRINUSE :8600 loop).
+ */
+const PM2_META_KEYS = new Set([
+  "name",
+  "namespace",
+  "exec_mode",
+  "exec_interpreter",
+  "instances",
+  "instance_var",
+  "node_app_instance",
+  "unique_id",
+  "status",
+  "username",
+  "windowsHide",
+  "merge_logs",
+  "vizion",
+  "vizion_running",
+  "autostart",
+  "autorestart",
+  "automation",
+  "km_link",
+]);
+
+/**
+ * @param {NodeJS.ProcessEnv} env
+ * @returns {NodeJS.ProcessEnv}
+ */
+export function withoutPm2Meta(env) {
+  /** @type {NodeJS.ProcessEnv} */
+  const out = {};
+  for (const [key, val] of Object.entries(env)) {
+    if (val == null) continue;
+    if (PM2_META_KEYS.has(key)) continue;
+    if (key.startsWith("pm_") || key.startsWith("axm_") || key.startsWith("PM2_")) {
+      continue;
+    }
+    out[key] = val;
+  }
+  return out;
+}
+
 /**
  * Shared env for child processes.
  * @param {{ generation: number }} opts
  */
 export function childEnv(opts) {
   return {
-    ...process.env,
+    ...withoutPm2Meta(process.env),
     JFLOW_CONFIG_GENERATION: String(opts.generation),
     JFLOW_CORS_ORIGIN: process.env.JFLOW_CORS_ORIGIN ?? "http://localhost:8500",
     PORT: process.env.JFLOW_HTTP_PORT ?? "8700",

packages/server/registry.js

@@ -36,7 +36,9 @@ import {
   resolveFailureTriggerConfig,
 } from "./trigger-failure.js";
 import { enqueueWorkflowJob } from "./workflow-queue.js";
-import { ensureInitialRevision } from "./workflow-history.js";
+import { ensureInitialRevision, recordRevision } from "./workflow-history.js";
+import { workflowIdFromFile } from "./workflow-normalize.js";
+import { publishReload } from "./control-bus.js";
 
 /**
  * @typedef {{ owner: string, file: string, workflow: any }} WorkflowEntry
@@ -553,6 +555,45 @@ export function createRegistry(server, opts = {}) {
     }
   }
 
+  /**
+   * Persist `enabled: false` for a workflow and reload registries across processes.
+   * @param {string} owner
+   * @param {string} file
+   * @param {string} key
+   */
+  async function disableWorkflowForConsecutiveFailures(owner, file, key) {
+    const content = fsStore.readWorkflowYaml(owner, file);
+    if (content == null) {
+      throw new Error(`workflow file missing for ${key}`);
+    }
+    const doc = yaml.parseDocument(content);
+    if (doc.errors?.length) {
+      throw new Error(doc.errors[0]?.message ?? "invalid yaml");
+    }
+    const parsed = doc.toJSON();
+    if (parsed?.enabled === false) {
+      log.debug({ workflow: key }, "workflow already disabled");
+      return;
+    }
+    doc.set("enabled", false);
+    const nextContent = String(doc);
+    fsStore.writeWorkflowYaml(owner, file, nextContent);
+    await recordRevision({
+      workflowId: workflowIdFromFile(file),
+      owner,
+      file,
+      content: nextContent,
+      reason: "disable-on-consecutive-failures",
+    });
+    reregister();
+    try {
+      await publishReload({ type: "workflows" });
+    } catch {
+      // Redis may be briefly unavailable; local reload already applied.
+    }
+    log.warn({ workflow: key }, "disabled workflow after consecutive failures");
+  }
+
   /**
    * @param {{
    *   key: string,
@@ -589,6 +630,17 @@ export function createRegistry(server, opts = {}) {
       return;
     }
 
+    if (failureConfig.disableOnConsecutiveFailures) {
+      const entry = workflows.get(opts.key);
+      if (entry) {
+        await disableWorkflowForConsecutiveFailures(entry.owner, entry.file, opts.key);
+      } else {
+        log.warn({ workflow: opts.key }, "cannot disable missing workflow entry");
+      }
+    }
+
+    if (!failureConfig.workflowName) return;
+
     const destKey = resolveWorkflowTriggerKey(opts.owner, failureConfig.workflowName);
     const alertData = buildFailureAlertData({
       sourceKey: opts.key,

packages/server/src/api/workflows.js

@@ -76,6 +76,7 @@ function triggerSummary(owner, workflow, nameById) {
       schedule: t?.schedule ?? null,
       onConsecutiveFailures: t?.onConsecutiveFailures ?? null,
       onFailureWorkflow: t?.onFailureWorkflow ?? null,
+      disableOnConsecutiveFailures: t?.disableOnConsecutiveFailures === true,
       auth: isHttp ? authLabel(t?.auth, nameById) : null,
     };
   });

packages/server/trigger-failure.js

@@ -47,12 +47,18 @@ export function resolveFailureTriggerConfig(workflow, owner, runtimeTrigger) {
 
     const threshold = Number(spec.onConsecutiveFailures);
     const workflowName = onFailureWorkflowName(spec);
-    if (!Number.isFinite(threshold) || threshold < 1 || workflowName.length === 0) {
+    const disableOnConsecutiveFailures = isDisableOnConsecutiveFailures(spec);
+    if (
+      !Number.isFinite(threshold) ||
+      threshold < 1 ||
+      (workflowName.length === 0 && !disableOnConsecutiveFailures)
+    ) {
       return null;
     }
     return {
       threshold: Math.floor(threshold),
-      workflowName,
+      workflowName: workflowName.length > 0 ? workflowName : null,
+      disableOnConsecutiveFailures,
     };
   }
   return null;
@@ -66,6 +72,13 @@ function onFailureWorkflowName(trigger) {
   return typeof value === "string" ? value.trim() : "";
 }
 
+/**
+ * @param {Record<string, unknown>} trigger
+ */
+export function isDisableOnConsecutiveFailures(trigger) {
+  return trigger?.disableOnConsecutiveFailures === true;
+}
+
 /**
  * @param {unknown} workflow
  */
@@ -81,12 +94,13 @@ export async function validateWorkflowFailureTriggers(workflow) {
     const hasThreshold =
       trigger.onConsecutiveFailures != null && trigger.onConsecutiveFailures !== "";
     const hasWorkflow = onFailureWorkflowName(trigger).length > 0;
+    const hasDisable = isDisableOnConsecutiveFailures(trigger);
 
-    if (!hasThreshold && !hasWorkflow) continue;
+    if (!hasThreshold && !hasWorkflow && !hasDisable) continue;
 
-    if (!hasThreshold || !hasWorkflow) {
+    if (!hasThreshold || (!hasWorkflow && !hasDisable)) {
       const err = new Error(
-        "onConsecutiveFailures and onFailureWorkflow must both be set on a trigger",
+        "onConsecutiveFailures requires onFailureWorkflow and/or disableOnConsecutiveFailures",
       );
       err.statusCode = 400;
       throw err;

packages/server/web-server.js

@@ -0,0 +1,116 @@
+/**
+ * Production UI server (:8500).
+ * Serves packages/web/dist and proxies /api, /ops, /admin, /u like Vite in dev.
+ * Always-on — survives Ops stop of jflow-http.
+ */
+import fs from "fs";
+import fastify from "fastify";
+import fastifyStatic from "@fastify/static";
+import { log } from "./logger.js";
+import { WEB_DIST } from "./paths.js";
+import {
+  controlOrigin,
+  httpOrigin,
+  proxyToOrigin,
+} from "./ops-proxy.js";
+
+if (!fs.existsSync(WEB_DIST)) {
+  log.error(
+    { WEB_DIST },
+    "web dist missing — run `pnpm build` before starting the UI server",
+  );
+  process.exit(1);
+}
+
+const port = Number(process.env.JFLOW_UI_PORT ?? 8500);
+const control = controlOrigin();
+const http = httpOrigin();
+
+const server = fastify({ loggerInstance: log });
+
+/**
+ * @param {import("fastify").FastifyRequest} req
+ * @param {import("fastify").FastifyReply} reply
+ */
+async function proxyApi(req, reply) {
+  const url = req.raw.url ?? "";
+  // Match Vite: /api/auth → control (login works when HTTP is stopped).
+  if (url === "/api/auth" || url.startsWith("/api/auth/") || url.startsWith("/api/auth?")) {
+    return proxyToOrigin(req, reply, control, {
+      unreachableMessage: "control plane unreachable",
+    });
+  }
+  return proxyToOrigin(req, reply, http, {
+    unreachableMessage: "HTTP API unreachable",
+  });
+}
+
+/**
+ * @param {import("fastify").FastifyRequest} req
+ * @param {import("fastify").FastifyReply} reply
+ */
+async function proxyOps(req, reply) {
+  return proxyToOrigin(req, reply, control, {
+    unreachableMessage: "control plane unreachable",
+  });
+}
+
+/**
+ * @param {import("fastify").FastifyRequest} req
+ * @param {import("fastify").FastifyReply} reply
+ */
+async function proxyHttp(req, reply) {
+  return proxyToOrigin(req, reply, http, {
+    unreachableMessage: "HTTP API unreachable",
+  });
+}
+
+server.all("/api", proxyApi);
+server.all("/api/*", proxyApi);
+server.all("/ops", proxyOps);
+server.all("/ops/*", proxyOps);
+server.all("/admin", proxyHttp);
+server.all("/admin/*", proxyHttp);
+server.all("/u", proxyHttp);
+server.all("/u/*", proxyHttp);
+
+await server.register(fastifyStatic, {
+  root: WEB_DIST,
+  wildcard: false,
+});
+
+server.setNotFoundHandler((req, reply) => {
+  const url = req.raw.url ?? "";
+  if (
+    url.startsWith("/api") ||
+    url.startsWith("/u/") ||
+    url.startsWith("/admin") ||
+    url.startsWith("/ops")
+  ) {
+    return reply.code(404).send({ error: "not found" });
+  }
+  return reply.sendFile("index.html");
+});
+
+async function shutdown() {
+  try {
+    await server.close();
+  } catch (err) {
+    log.error({ err }, "web-server shutdown error");
+  }
+  process.exit(0);
+}
+
+process.on("SIGINT", shutdown);
+process.on("SIGTERM", shutdown);
+
+try {
+  await server.listen({ host: "0.0.0.0", port });
+  log.info(
+    { port, WEB_DIST, control, http },
+    "UI server listening (static + proxy)",
+  );
+} catch (err) {
+  log.error({ err }, "failed to start UI server");
+  process.exit(1);
+}

packages/web/src/components/workflow/TriggerCard.jsx

@@ -143,10 +143,15 @@ function TriggerCardView({
 
 function triggerSummary(trigger, owner) {
   const type = trigger?.type;
-  const failure =
-    trigger.onConsecutiveFailures && trigger.onFailureWorkflow
-      ? ` · onFailure@${trigger.onFailureWorkflow}`
-      : "";
+  /** @type {string[]} */
+  const failureParts = [];
+  if (trigger.onConsecutiveFailures && trigger.onFailureWorkflow) {
+    failureParts.push(`onFailure@${trigger.onFailureWorkflow}`);
+  }
+  if (trigger.disableOnConsecutiveFailures) {
+    failureParts.push("auto-disable");
+  }
+  const failure = failureParts.length ? ` · ${failureParts.join(", ")}` : "";
   if (type === "HTTP") {
     return `${trigger.method || "POST"} ${namespacedPath(owner || "owner", trigger.path || "/")}${failure}`;
   }
@@ -383,6 +388,20 @@ function FailureAlertFields({ trigger, disabled, onChange, alertDestinations })
           ) : null}
         </FormSelect>
       </Field>
+      <label className="label cursor-pointer justify-start gap-3 py-0">
+        <input
+          type="checkbox"
+          className="checkbox checkbox-sm"
+          checked={Boolean(trigger.disableOnConsecutiveFailures)}
+          disabled={disabled}
+          onChange={(e) =>
+            onChange({ ...trigger, disableOnConsecutiveFailures: e.target.checked })
+          }
+        />
+        <span className="label-text">
+          Disable this workflow when the consecutive failure threshold is reached
+        </span>
+      </label>
     </div>
   );
 }

packages/web/src/lib/workflow-doc.js

@@ -171,6 +171,7 @@ export function newHttpTrigger() {
     unauthorized: null,
     onConsecutiveFailures: "",
     onFailureWorkflow: "",
+    disableOnConsecutiveFailures: false,
   };
 }
 
@@ -186,6 +187,7 @@ export function newCronTrigger() {
     unauthorized: null,
     onConsecutiveFailures: "",
     onFailureWorkflow: "",
+    disableOnConsecutiveFailures: false,
   };
 }
 
@@ -201,6 +203,7 @@ export function newWorkflowTrigger() {
     unauthorized: null,
     onConsecutiveFailures: "",
     onFailureWorkflow: "",
+    disableOnConsecutiveFailures: false,
   };
 }
 
@@ -303,9 +306,16 @@ function normalizeTrigger(raw) {
           "unauthorized",
           "onConsecutiveFailures",
           "onFailureWorkflow",
+          "disableOnConsecutiveFailures",
         ])
       : type === "cron"
-        ? new Set(["type", "schedule", "onConsecutiveFailures", "onFailureWorkflow"])
+        ? new Set([
+            "type",
+            "schedule",
+            "onConsecutiveFailures",
+            "onFailureWorkflow",
+            "disableOnConsecutiveFailures",
+          ])
         : new Set(["type"]);
   /** @type {Record<string, unknown>} */
   const extra = {};
@@ -323,6 +333,7 @@ function normalizeTrigger(raw) {
         ? ""
         : String(raw.onConsecutiveFailures),
     onFailureWorkflow: readOnFailureWorkflow(raw),
+    disableOnConsecutiveFailures: raw.disableOnConsecutiveFailures === true,
     auth: Array.isArray(raw.auth) ? raw.auth : null,
     response: typeof raw.response === "string" ? raw.response : "",
     unauthorized: raw.unauthorized ?? null,
@@ -377,6 +388,9 @@ function dumpFailureTriggerFields(t, out) {
   if (typeof t.onFailureWorkflow === "string" && t.onFailureWorkflow.trim()) {
     out.onFailureWorkflow = t.onFailureWorkflow.trim();
   }
+  if (t.disableOnConsecutiveFailures === true) {
+    out.disableOnConsecutiveFailures = true;
+  }
 }
 
 function dumpTrigger(t) {

pnpm-lock.yaml

@@ -77,7 +77,7 @@ importers:
         specifier: ^4.0.0
         version: 4.0.0
       pm2:
-        specifier: ^6.0.13
+        specifier: 6.0.14
         version: 6.0.14(supports-color@7.2.0)
       rss-parser:
         specifier: ^3.13.0

scripts/pm2.mjs

@@ -0,0 +1,48 @@
+#!/usr/bin/env node
+/**
+ * Run the same PM2 binary control.js `require("pm2")` uses.
+ * A global `pm2` 7.x talking to an in-memory 6.x daemon pegs CPU on start.
+ */
+import { spawn, spawnSync } from "node:child_process";
+import { createRequire } from "node:module";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+
+const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
+const require = createRequire(
+  path.join(root, "packages/server/package.json"),
+);
+const pm2Root = path.dirname(require.resolve("pm2/package.json"));
+const pm2Bin = path.join(pm2Root, "bin/pm2");
+const args = process.argv.slice(2);
+
+function run(pm2Args, opts = {}) {
+  return spawnSync(process.execPath, [pm2Bin, ...pm2Args], {
+    cwd: root,
+    encoding: "utf8",
+    ...opts,
+  });
+}
+
+const cmd = args[0];
+if (cmd === "start" || cmd === "restart" || cmd === "reload") {
+  const probe = run(["ls"], { stdio: ["ignore", "pipe", "pipe"] });
+  const text = `${probe.stdout ?? ""}${probe.stderr ?? ""}`;
+  const mem = text.match(/In memory PM2 version:\s*(\S+)/);
+  const loc = text.match(/Local PM2 version:\s*(\S+)/);
+  if (mem && loc && mem[1] !== loc[1]) {
+    console.error(
+      `[jflow] PM2 daemon ${mem[1]} != CLI ${loc[1]}. Killing the daemon so control.js and the CLI share one version.`,
+    );
+    run(["kill"], { stdio: "inherit" });
+  }
+}
+
+const child = spawn(process.execPath, [pm2Bin, ...args], {
+  cwd: root,
+  stdio: "inherit",
+});
+child.on("exit", (code, signal) => {
+  if (signal) process.kill(process.pid, signal);
+  process.exit(code ?? 1);
+});