@pauldeng/node-red-contrib-bullmq 1.0.3 → 2.0.0

package/bull-queue.js

@@ -1,10 +1,18 @@
 "use strict";
 
 const {
+  createPostgresBackend,
+  DelayedError,
   FlowProducer,
+  MINIMUM_POSTGRES_VERSION,
   Queue,
   QueueEvents,
+  RECOMMENDED_POSTGRES_VERSION,
+  SchemaMigrationRequiredError,
+  SchemaVersionMismatchError,
   UnrecoverableError,
+  UnsupportedPostgresVersionError,
+  WaitingError,
   Worker,
 } = require("bullmq");
 const IORedis = require("ioredis");
@@ -16,12 +24,14 @@ const {
 } = require("./lib/acknowledgements");
 const { dispatchCommand } = require("./lib/commands");
 const {
+  POSTGRES_CONNECTION_TIMEOUT_MS,
   buildBullMQOptions,
   buildRedisDescriptor,
   createRedisConnection,
   normalizeQueueConfig,
 } = require("./lib/connections");
 const { serializeFlowJob, serializeJob } = require("./lib/serialization");
+const { version: PACKAGE_VERSION } = require("./package.json");
 
 const DEFAULT_EVENTS = [
   "active",
@@ -37,6 +47,7 @@ const DEFAULT_EVENTS = [
   "progress",
   "removed",
   "resumed",
+  "retries-exhausted",
   "stalled",
   "waiting",
   "waiting-children",
@@ -47,6 +58,34 @@ const DEFAULT_EVENTS = [
 // an unreachable Redis server.
 const CLOSE_GRACE_MS = 1000;
 
+// A connection that is actually ready gets a longer budget: Worker.close()
+// waits for in-flight jobs, and cutting that off abandons a running job to the
+// stalled checker, which re-runs it and can eventually fail it for exceeding
+// maxStalledCount. The ceiling stays well under Node-RED's own ~15s node close
+// timeout, and resources close concurrently, so this bounds one resource, not
+// the sum.
+const GRACEFUL_CLOSE_MS = 10000;
+
+// PostgreSQL's connection timeout is its only way to stop an outstanding
+// connect. Unlike Redis, it has no raw client that we can force closed, so its
+// close budget needs scheduling margin beyond that timeout. Derived from the
+// timeout it has to outlast rather than written as a literal, so raising
+// POSTGRES_CONNECTION_TIMEOUT_MS cannot silently leave the budget short. It
+// has a ceiling as well as a floor: Node-RED gives each node ~15s to close, so
+// POSTGRES_CONNECTION_TIMEOUT_MS must stay well under 14s or this budget stops
+// bounding anything and Node-RED's own timeout cuts the close off instead.
+const POSTGRES_CLOSE_MS = POSTGRES_CONNECTION_TIMEOUT_MS + CLOSE_GRACE_MS;
+
+function closeBudgetFor(connection) {
+  // PostgreSQL owners have no companion connection in node.resources because
+  // BullMQ owns their pool. Redis owners retain the live ioredis status that
+  // distinguishes a healthy worker drain from an unreachable fast close.
+  if (!connection) {
+    return POSTGRES_CLOSE_MS;
+  }
+  return connection.status === "ready" ? GRACEFUL_CLOSE_MS : CLOSE_GRACE_MS;
+}
+
 async function settled(promise) {
   try {
     await promise;
@@ -56,13 +95,16 @@ async function settled(promise) {
   return "settled";
 }
 
-async function timedOut(ms) {
-  await sleep(ms);
-  return "timeout";
-}
-
 async function settleWithin(promise, ms) {
-  return await Promise.race([settled(promise), timedOut(ms)]);
+  const controller = new AbortController();
+  try {
+    return await Promise.race([
+      settled(promise),
+      sleep(ms, "timeout", { signal: controller.signal }),
+    ]);
+  } finally {
+    controller.abort();
+  }
 }
 
 function disconnectClient(client) {
@@ -75,7 +117,7 @@ function disconnectClient(client) {
   }
 }
 
-function forceDisconnect(resource) {
+async function forceDisconnect(resource) {
   if (!resource) {
     return;
   }
@@ -84,18 +126,44 @@ function forceDisconnect(resource) {
     disconnectClient(resource);
     return;
   }
-  // BullMQ resource. Its public disconnect() awaits a connection promise that
-  // never settles while Redis is unreachable, so reach for the underlying
-  // ioredis clients directly (BullMQ is pinned to exactly 5.80.9).
-  if (resource.connection) {
-    disconnectClient(resource.connection._client);
+  // BullMQ 6.3.1's public disconnect() can await the same never-ready promise
+  // as close(), which would spend a second shutdown budget after the first one
+  // already expired. The exact BullMQ pin makes this one backend escape hatch
+  // deliberate and testable until upstream disconnect becomes bounded.
+  if (typeof resource.getBackend !== "function") {
+    return;
+  }
+  const backend = resource.getBackend();
+  // RedisQueueBackend is the only installed backend exposing raw `_client`
+  // handles. PostgreSQL deliberately skips this branch.
+  if (backend && backend.connection && backend.connection._client) {
+    disconnectClient(backend.connection._client);
+    disconnectClient(
+      backend.blockingConnection && backend.blockingConnection._client,
+    );
   }
-  if (resource.blockingConnection) {
-    disconnectClient(resource.blockingConnection._client);
+  // Measured on installed BullMQ 6.3.1: a Worker's close() never reaches this
+  // point on its own here, because its very first cleanup step awaits the
+  // same stuck connection above. That means the lock-renewal timer it starts
+  // on construction (independent of connection state) is never cancelled by
+  // Worker's own close() -- stop it here directly so it cannot outlive the
+  // resource and keep the process alive.
+  if (resource.lockManager && typeof resource.lockManager.close === "function") {
+    await resource.lockManager.close();
   }
+  // Same reasoning for the stalled-job checker: Worker.close() only stops it
+  // after the cleanup step that hung above, and each check reschedules its own
+  // timer. This one is only live if the worker processed at least once before
+  // Redis went away -- the common production case, unlike a connection that was
+  // never reachable at all.
+  if (typeof resource.stalledCheckStopper === "function") {
+    resource.stalledCheckStopper();
+  }
+  // PostgreSQL has no raw force-disconnect branch. closeBudgetFor therefore
+  // lets its own connection timeout settle close() before Node-RED calls done.
 }
 
-async function closeResource(resource) {
+async function closeResource(resource, connection) {
   if (!resource) {
     return;
   }
@@ -106,26 +174,29 @@ async function closeResource(resource) {
     // connections and force-disconnect everything else.
     if (resource.status === "ready" && typeof resource.quit === "function") {
       if ((await settleWithin(resource.quit(), CLOSE_GRACE_MS)) === "timeout") {
-        forceDisconnect(resource);
+        await forceDisconnect(resource);
       }
     } else {
-      forceDisconnect(resource);
+      await forceDisconnect(resource);
     }
     return;
   }
 
   // BullMQ resource: QueueEvents.close() blocks forever on a connection that
-  // never became ready, so cap the graceful close and force-disconnect the
-  // sockets when it does not settle in time.
-  if ((await settleWithin(resource.close(), CLOSE_GRACE_MS)) === "timeout") {
-    forceDisconnect(resource);
+  // never became ready, so cap the graceful close and force-disconnect when
+  // it does not settle in time (see forceDisconnect for the fallback chain).
+  if (
+    (await settleWithin(resource.close(), closeBudgetFor(connection))) ===
+    "timeout"
+  ) {
+    await forceDisconnect(resource);
   }
 }
 
 async function closeResourcePair(owner, connection) {
   let firstError;
   try {
-    await closeResource(owner);
+    await closeResource(owner, connection);
   } catch (err) {
     firstError = err;
   }
@@ -151,6 +222,62 @@ function nodeDone(node, done, err, msg) {
   }
 }
 
+// FlowProducer.addBulk(flows) accepts no options argument, so the queuesOptions
+// route withFlowJobDefaults() uses for add() is unavailable here. Stamp the
+// retention onto each job's own opts instead; anything the caller set wins.
+function withBulkFlowJobDefaults(flow, defaultJobOptions) {
+  if (!defaultJobOptions || !flow || typeof flow !== "object") {
+    return flow;
+  }
+
+  const children = Array.isArray(flow.children)
+    ? flow.children.map((child) =>
+        withBulkFlowJobDefaults(child, defaultJobOptions),
+      )
+    : flow.children;
+
+  return {
+    ...flow,
+    opts: { ...defaultJobOptions, ...flow.opts },
+    ...(children === undefined ? {} : { children }),
+  };
+}
+
+function withFlowJobDefaults(flow, flowOptions, defaultJobOptions) {
+  if (!defaultJobOptions) {
+    return flowOptions;
+  }
+
+  const options =
+    flowOptions && typeof flowOptions === "object" ? flowOptions : {};
+  // Already a fresh copy, so later queues are assigned into it rather than
+  // re-spreading the whole map per queue name.
+  const queuesOptions = { ...options.queuesOptions };
+  const pending = [flow];
+
+  while (pending.length > 0) {
+    const job = pending.pop();
+    if (!job || typeof job !== "object") {
+      continue;
+    }
+    if (typeof job.queueName === "string" && job.queueName) {
+      const queueOptions = queuesOptions[job.queueName] || {};
+      queuesOptions[job.queueName] = {
+        ...queueOptions,
+        defaultJobOptions: {
+          ...defaultJobOptions,
+          ...queueOptions.defaultJobOptions,
+        },
+      };
+    }
+    if (Array.isArray(job.children)) {
+      pending.push(...job.children);
+    }
+  }
+
+  return { ...options, queuesOptions };
+}
+
 function isPresent(value) {
   return value !== undefined && value !== null && value !== "";
 }
@@ -190,14 +317,60 @@ function setDisconnected(node) {
   node.status({ fill: "red", shape: "ring", text: "disconnected" });
 }
 
-function attachErrorListener(resource, node) {
+function attachErrorListener(resource, node, startupFailureOwner) {
   if (!resource || typeof resource.on !== "function") {
-    return;
+    return () => {};
   }
+  let ready = false;
   resource.on("error", (err) => {
     setDisconnected(node);
-    node.error(err);
+    if (startupFailureOwner && !ready) {
+      startupFailureOwner.reportBackendFailure(err);
+    } else {
+      node.error(err);
+    }
   });
+  return function markReady() {
+    ready = true;
+  };
+}
+
+// Turns a rejected backend readiness promise -- or a synchronous resource
+// construction failure, since BullMQ's postgres factory validates and loads
+// its optional `pg` dependency before any I/O -- into one of a handful of
+// actionable categories, so a Node-RED user is told what to DO rather than
+// just that the backend is unavailable. The category string is also the
+// dedup key each config node latches on; see node.reportBackendFailure.
+function describeBackendFailure(err) {
+  const message = err && err.message ? err.message : String(err);
+  if (err instanceof SchemaMigrationRequiredError) {
+    return {
+      category: "schema-migration-required",
+      message: `${message} Set the queue configuration node's "migrate" property to true so this node initialises the schema, or run BullMQ's PostgreSQL migrations against the database yourself before deploying.`,
+    };
+  }
+  if (err instanceof SchemaVersionMismatchError) {
+    return {
+      category: "schema-version-mismatch",
+      message: `${message} Upgrade this Node-RED package to a release that uses the required BullMQ major; PostgreSQL schema downgrades are not supported.`,
+    };
+  }
+  if (err instanceof UnsupportedPostgresVersionError) {
+    return {
+      category: "unsupported-postgres-version",
+      message: `${message} BullMQ's PostgreSQL backend requires server version ${MINIMUM_POSTGRES_VERSION} or newer (${RECOMMENDED_POSTGRES_VERSION}+ recommended).`,
+    };
+  }
+  // BullMQ lazily requires the optional `pg` package and, when it cannot be
+  // resolved, throws its own actionable error rather than a raw
+  // MODULE_NOT_FOUND -- reuse that message instead of wrapping it again.
+  if (message.includes("npm install pg")) {
+    return { category: "pg-missing", message };
+  }
+  return {
+    category: "generic",
+    message: `BullMQ backend is unavailable: ${message}`,
+  };
 }
 
 function createJobMessage(job, queueName, extraBull = {}) {
@@ -223,24 +396,60 @@ module.exports = function registerBullMQNodes(RED) {
     RED.nodes.createNode(this, n);
     const node = this;
 
-    node.users = {};
     node.resources = new Map();
     node.config = normalizeQueueConfig(n, node.credentials || {});
     node.queue = null;
     node.producerConnection = null;
-
-    node.register = function register(bullNode) {
-      node.users[bullNode.id] = bullNode;
-      bullNode.status({
-        fill: "grey",
-        shape: "ring",
-        text: "configured",
-      });
+    node.telemetry = undefined;
+    node.telemetryUnavailable = false;
+
+    // Categories already reported for this config node. BullMQ builds one
+    // backend per resource (Queue/Worker/QueueEvents/FlowProducer each get
+    // their own pool), so the same misconfiguration can be observed from
+    // more than one of them -- this latch is what keeps one deploy at one
+    // report per distinct failure instead of one per resource that hit it.
+    node.backendFailures = new Set();
+    node.reportBackendFailure = function reportBackendFailure(err) {
+      const { category, message } = describeBackendFailure(err);
+      if (node.backendFailures.has(category)) {
+        return;
+      }
+      node.backendFailures.add(category);
+      node.error(message);
     };
 
-    node.deregister = function deregister(bullNode, done) {
-      delete node.users[bullNode.id];
-      done();
+    // Lazily constructs (and caches) the one BullMQOtel instance shared by
+    // this config node's Queue/Worker/FlowProducer. Deferred until first
+    // resource creation rather than built in the constructor above: Node-RED
+    // builds config nodes at deploy time, which can precede the host's
+    // OpenTelemetry bootstrap, and enableMetrics needs a MeterProvider
+    // registered before BullMQOtel is constructed.
+    node.getTelemetry = function getTelemetry() {
+      if (!node.config.telemetry) {
+        return undefined;
+      }
+      if (node.telemetry || node.telemetryUnavailable) {
+        return node.telemetry;
+      }
+      let BullMQOtel;
+      try {
+        ({ BullMQOtel } = require("bullmq-otel"));
+      } catch (err) {
+        node.telemetryUnavailable = true;
+        node.error(
+          "BullMQ telemetry is enabled but bullmq-otel is not installed; run npm install bullmq-otel",
+        );
+        return undefined;
+      }
+      const serviceName =
+        node.config.telemetryServiceName || node.config.queueName;
+      node.telemetry = new BullMQOtel({
+        tracerName: serviceName,
+        meterName: serviceName,
+        version: PACKAGE_VERSION,
+        enableMetrics: node.config.telemetryMetrics,
+      });
+      return node.telemetry;
     };
 
     // owner is the node whose status should reflect connection errors. It
@@ -262,61 +471,182 @@ module.exports = function registerBullMQNodes(RED) {
       await closeResourcePair(owner, connection);
     };
 
+    // The four BullMQ owners differ only in which constructor they call and
+    // where that constructor takes the backend factory: Queue and QueueEvents
+    // take it 3rd, Worker 4th, FlowProducer 2nd. Everything around that is the
+    // same work, so `construct` receives the built options plus the factory to
+    // pass on -- undefined on Redis, where every one of those constructors
+    // defaults the parameter to BullMQ's own Redis factory.
+    //
+    // On PostgreSQL, BullMQ's factory validates the schema name and
+    // synchronously loads the optional `pg` module (createPostgresBackend ->
+    // new PostgresConnection) before any resource exists, so a bad schema name
+    // or a missing `pg` install throws out of `new Queue/Worker/QueueEvents/
+    // FlowProducer(...)` itself rather than through waitUntilReady(). Catch it
+    // here instead of crashing the owning node's constructor.
+    function createResource(role, owner, telemetry, construct, extraOptions) {
+      // BullMQ owns every PostgreSQL connection (a pool plus a dedicated
+      // LISTEN client per backend); this package builds none of its own on
+      // that path, so createConnection is skipped entirely.
+      const isPostgres = node.config.backend === "postgres";
+      const connection = isPostgres
+        ? undefined
+        : node.createConnection(role, owner);
+      const options = {
+        ...buildBullMQOptions(
+          node.config,
+          isPostgres ? node.config.postgres : connection,
+          telemetry,
+          role,
+        ),
+        ...extraOptions,
+      };
+
+      let resource;
+      if (isPostgres) {
+        try {
+          resource = construct(options, createPostgresBackend);
+        } catch (err) {
+          publishBackendStatus("disconnected");
+          node.reportBackendFailure(err);
+          return { resource: undefined, connection };
+        }
+      } else {
+        resource = construct(options);
+      }
+
+      // On postgres there is no connection of our own -- the owner is tracked
+      // with no value so close/redeploy still walks it, but has nothing raw to
+      // close.
+      node.resources.set(resource, connection);
+      return { resource, connection };
+    }
+
     node.getQueue = function getQueue() {
       if (!node.queue) {
-        node.producerConnection = node.createConnection("producer");
-        node.producerConnection.setMaxListeners(0);
-        node.queue = new Queue(
-          node.config.queueName,
-          buildBullMQOptions(node.config, node.producerConnection)
+        const { resource, connection } = createResource(
+          "producer",
+          node,
+          node.getTelemetry(),
+          (options, factory) =>
+            new Queue(node.config.queueName, options, factory),
+          // Queue-level auto-removal. Without it BullMQ keeps every completed
+          // and failed job forever, which is the unbounded growth its
+          // production guide warns about. Per-job msg.jobopts still wins.
+          node.config.defaultJobOptions
+            ? { defaultJobOptions: node.config.defaultJobOptions }
+            : undefined,
         );
-        node.resources.set(node.queue, node.producerConnection);
+        if (!resource) {
+          return null;
+        }
+        node.queue = resource;
+        node.producerConnection = connection || null;
+        node.watchBackend(node.queue.getBackend());
         attachErrorListener(node.queue, node);
       }
       return node.queue;
     };
 
-    // The producer connection backs the shared queue used by bull cmd nodes;
-    // exposing it lets those nodes mirror the real connection state.
-    node.getProducerConnection = function getProducerConnection() {
-      node.getQueue();
-      return node.producerConnection;
+    // Live reachability of the shared backend, owned here rather than read by
+    // each bullmq cmd node. waitUntilReady() cannot answer "is it reachable
+    // now": RedisQueueBackend awaits connection.client, which returns the
+    // promise built once in the constructor, and PostgresConnection memoizes
+    // readyPromise the same way. So a node deployed after an outage began would
+    // see that resolved promise and paint a false "connected". The backend's
+    // ready/error/close events are live, so they own the state after the first
+    // observation and every reader sees the current value.
+    node.backendStatus = "connecting";
+    node.backendReaders = new Set();
+
+    function publishBackendStatus(status) {
+      node.backendStatus = status;
+      for (const read of node.backendReaders) {
+        read(status);
+      }
+    }
+
+    node.watchBackend = function watchBackend(backend) {
+      backend.on("ready", () => publishBackendStatus("connected"));
+      backend.on("error", () => publishBackendStatus("disconnected"));
+      backend.on("close", () => publishBackendStatus("disconnected"));
+      // Seed the first observation once, for the whole config node. An async
+      // IIFE rather than a promise chain (forbidden in this file) so it never
+      // blocks deploy, and it must not overwrite a live event that already
+      // told us more than this memoized promise can.
+      (async () => {
+        try {
+          await backend.waitUntilReady();
+          if (node.backendStatus === "connecting") {
+            publishBackendStatus("connected");
+          }
+        } catch (err) {
+          // Do NOT assume an event already covered this. PostgresConnection's
+          // bootstrap() rejects on connect, auth, migration, or schema failure
+          // and emits nothing (its emitError only forwards idle-pool and LISTEN
+          // errors, and only when a listener is already attached), and Queue's
+          // constructor swallows the same rejection. Without this the node
+          // would sit on "connecting" forever with nothing reported. The
+          // sibling bullmq events and bullmq flow nodes handle their own
+          // waitUntilReady() rejection through node.reportBackendFailure too,
+          // so the same failure seen from more than one resource still lands
+          // as one report.
+          if (node.backendStatus === "connecting") {
+            publishBackendStatus("disconnected");
+            node.reportBackendFailure(err);
+          }
+        }
+      })();
+    };
+
+    // Readers get the current status immediately, which is what makes a
+    // late-deployed node correct: the shared state is live, unlike the
+    // already-fired "ready" event it would otherwise have missed.
+    node.readBackendStatus = function readBackendStatus(read) {
+      node.backendReaders.add(read);
+      read(node.backendStatus);
+      return function stopReading() {
+        node.backendReaders.delete(read);
+      };
     };
 
     // Runtime nodes pass themselves as owner and attach their own resource
     // error listener, so worker/events/flow errors surface on the visible
     // runtime node rather than the hidden config node.
     node.createWorker = function createWorker(processor, options, owner = node) {
-      const connection = node.createConnection("worker", owner);
-      const worker = new Worker(node.config.queueName, processor, {
-        ...buildBullMQOptions(node.config, connection),
-        ...options,
-      });
-      node.resources.set(worker, connection);
-      return worker;
+      return createResource(
+        "worker",
+        owner,
+        node.getTelemetry(),
+        (workerOptions, factory) =>
+          new Worker(node.config.queueName, processor, workerOptions, factory),
+        options,
+      ).resource;
     };
 
     node.createQueueEvents = function createQueueEvents(owner = node) {
-      const connection = node.createConnection("events", owner);
-      const queueEvents = new QueueEvents(
-        node.config.queueName,
-        buildBullMQOptions(node.config, connection)
-      );
-      node.resources.set(queueEvents, connection);
-      return queueEvents;
+      // QueueEvents deliberately gets no telemetry instance; see docs/TELEMETRY.md.
+      return createResource(
+        "events",
+        owner,
+        undefined,
+        (options, factory) =>
+          new QueueEvents(node.config.queueName, options, factory),
+      ).resource;
     };
 
     node.createFlowProducer = function createFlowProducer(owner = node) {
-      const connection = node.createConnection("producer", owner);
-      const flowProducer = new FlowProducer(
-        buildBullMQOptions(node.config, connection)
-      );
-      node.resources.set(flowProducer, connection);
-      return flowProducer;
+      return createResource(
+        "producer",
+        owner,
+        node.getTelemetry(),
+        (options, factory) => new FlowProducer(options, factory),
+      ).resource;
     };
 
     node.on("close", async function onClose(removed, done) {
       try {
+        node.backendFailures.clear();
         const resources = Array.from(node.resources.entries()).reverse();
         node.resources.clear();
         await Promise.all(
@@ -332,7 +662,7 @@ module.exports = function registerBullMQNodes(RED) {
     });
   }
 
-  RED.nodes.registerType("bull-queue-server", BullQueueServerSetup, {
+  RED.nodes.registerType("bullmq-queue-server", BullQueueServerSetup, {
     credentials: {
       password: { type: "password" },
       sentinelPassword: { type: "password" },
@@ -350,32 +680,44 @@ module.exports = function registerBullMQNodes(RED) {
 
     if (!node.bullConn) {
       node.status({ fill: "red", shape: "ring", text: "missing queue" });
-      node.error("Missing bull-queue-server config node");
+      node.error("Missing bullmq-queue-server config node");
       return;
     }
 
-    node.bullConn.register(node);
-
-    // Watch the shared producer connection so the visible status reflects
-    // whether Redis is actually reachable instead of a static "configured".
-    const connection = node.bullConn.getProducerConnection();
-    const connectionListeners = {
-      ready: () => setConnected(node),
-      error: () => setDisconnected(node),
-      close: () => setDisconnected(node),
+    // Mirror the config node's live view of the shared backend. Reading the
+    // shared state rather than subscribing to the backend directly keeps the
+    // listener count on the backend constant no matter how many bullmq cmd
+    // nodes a flow has, and gives a node deployed mid-outage the current
+    // status instead of a stale one.
+    const applyBackendStatus = (status) => {
+      if (status === "connected") {
+        setConnected(node);
+      } else if (status === "disconnected") {
+        setDisconnected(node);
+      } else {
+        setConnecting(node);
+      }
     };
-    for (const [event, listener] of Object.entries(connectionListeners)) {
-      connection.on(event, listener);
-    }
-    if (connection.status === "ready") {
-      setConnected(node);
-    } else {
-      setConnecting(node);
-    }
-
+    // Create the shared queue up front, as this node has always done, so the
+    // status reflects real reachability instead of a static "configured" the
+    // moment the flow deploys. This is the call that builds the queue, its
+    // connection, and (once PostgreSQL is wired) its pool; reading the status
+    // afterwards only subscribes.
+    node.bullConn.getQueue();
+    const stopReadingBackend =
+      node.bullConn.readBackendStatus(applyBackendStatus);
     node.on("input", async function onInput(msg, send, done) {
       try {
-        const result = await dispatchCommand(node.bullConn.getQueue(), msg);
+        const queue = node.bullConn.getQueue();
+        if (!queue) {
+          // Construction failed synchronously (missing pg, bad schema name).
+          // The config node already reported why; say something useful here
+          // rather than letting queue.add() throw a bare TypeError per message.
+          throw new Error(
+            "BullMQ queue is unavailable; see the queue configuration node's error",
+          );
+        }
+        const result = await dispatchCommand(queue, msg);
         msg.payload = result;
         nodeSend(node, send, msg);
         nodeDone(node, done);
@@ -385,10 +727,8 @@ module.exports = function registerBullMQNodes(RED) {
     });
 
     node.on("close", function onClose(removed, done) {
-      for (const [event, listener] of Object.entries(connectionListeners)) {
-        connection.removeListener(event, listener);
-      }
-      node.bullConn.deregister(node, done);
+      stopReadingBackend();
+      done();
     });
   }
 
@@ -401,14 +741,17 @@ module.exports = function registerBullMQNodes(RED) {
 
     if (!node.bullQueue) {
       node.status({ fill: "red", shape: "ring", text: "missing queue" });
-      node.error("Missing bull-queue-server config node");
+      node.error("Missing bullmq-queue-server config node");
       return;
     }
 
-    node.bullQueue.register(node);
-
     const workerOptions = {
       concurrency: parsePositiveInteger(n.concurrency, 1, "Concurrency"),
+      maxStartedAttempts: parsePositiveInteger(
+        n.maxStartedAttempts,
+        100,
+        "Max Started Attempts"
+      ),
     };
     const hasLimiterMax = isPresent(n.limiterMax);
     const hasLimiterDuration = isPresent(n.limiterDuration);
@@ -425,8 +768,21 @@ module.exports = function registerBullMQNodes(RED) {
         ),
       };
     }
+    if (node.bullQueue.config.backend === "postgres") {
+      // A permanent readiness error (for example an unmigrated schema) makes
+      // BullMQ's autorun loop retry without delay. Start only after the one
+      // readiness promise succeeds so a configuration error cannot spin the
+      // Node-RED runtime.
+      workerOptions.autorun = false;
+    }
 
-    const processor = async (job) => {
+    // Arity 3 tells BullMQ to create and track a per-job AbortController
+    // (worker.js: processorAcceptsSignal = processor.length >= 3), which is
+    // what makes cancelJob/cancelAllJobs able to reach this job at all.
+    // The token is the job's lock: bullmq job needs it for moveToWait and
+    // moveToDelayed, and it stays in the acknowledgement registry -- never in
+    // a message.
+    const processor = async (job, token, signal) => {
       if (node.completionMode === "manual") {
         const timeoutMs = parseAckTimeoutMs(n.ackTimeout);
         const acknowledgement = acknowledgements.create(
@@ -435,6 +791,9 @@ module.exports = function registerBullMQNodes(RED) {
             queue: node.bullQueue.getQueue(),
             queueName: node.bullQueue.config.queueName,
             runNodeId: node.id,
+            worker: node.worker,
+            signal,
+            token,
           },
           timeoutMs
         );
@@ -453,8 +812,52 @@ module.exports = function registerBullMQNodes(RED) {
     };
 
     node.worker = node.bullQueue.createWorker(processor, workerOptions, node);
-    attachErrorListener(node.worker, node);
-    node.worker.on("ready", () => setConnected(node));
+    // createWorker reports and returns undefined on a synchronous backend
+    // construction failure (e.g. postgres selected with pg not installed);
+    // nothing was created, so there is nothing to close on this node's own
+    // close, and no listener setup below would have anything to attach to.
+    if (!node.worker) {
+      setDisconnected(node);
+      return;
+    }
+    const workerStartupFailureOwner =
+      node.bullQueue.config.backend === "postgres" ? node.bullQueue : undefined;
+    const markWorkerReady = attachErrorListener(
+      node.worker,
+      node,
+      workerStartupFailureOwner,
+    );
+    if (workerStartupFailureOwner) {
+      let workerReady = false;
+      async function startPostgresWorker() {
+        try {
+          await node.worker.waitUntilReady();
+          if (node.worker.closing) {
+            return;
+          }
+          workerReady = true;
+          markWorkerReady();
+          setConnected(node);
+          await node.worker.run();
+        } catch (err) {
+          if (node.worker.closing) {
+            return;
+          }
+          setDisconnected(node);
+          if (workerReady) {
+            node.error(err);
+          } else {
+            workerStartupFailureOwner.reportBackendFailure(err);
+          }
+        }
+      }
+      void startPostgresWorker();
+    } else {
+      node.worker.on("ready", () => {
+        markWorkerReady();
+        setConnected(node);
+      });
+    }
     node.worker.on("closed", () => setDisconnected(node));
     setConnecting(node);
 
@@ -465,7 +868,6 @@ module.exports = function registerBullMQNodes(RED) {
       );
       try {
         await node.bullQueue.releaseResource(node.worker);
-        node.bullQueue.deregister(node, () => {});
         done();
       } catch (err) {
         done(err);
@@ -543,8 +945,63 @@ module.exports = function registerBullMQNodes(RED) {
             context.fail(Worker.RateLimitError());
             nodeDone(node, done);
             return;
+          // Step/retry transitions. Each hands the job's lock token back to
+          // BullMQ and then settles the acknowledgement with the error class
+          // the worker special-cases, so the job is NOT moved to failed
+          // (worker.js checks DelayedError and WaitingError).
+          case "moveToWait": {
+            await context.job.moveToWait(context.token);
+            context.fail(new WaitingError());
+            msg.payload = true;
+            nodeSend(node, send, msg);
+            nodeDone(node, done);
+            return;
+          }
+          case "moveToDelayed": {
+            const delay = Number(msg.delay);
+            if (!Number.isFinite(delay) || delay < 0) {
+              throw new Error(
+                "moveToDelayed requires msg.delay in milliseconds"
+              );
+            }
+            await context.job.moveToDelayed(Date.now() + delay, context.token);
+            context.fail(new DelayedError());
+            msg.payload = true;
+            nodeSend(node, send, msg);
+            nodeDone(node, done);
+            return;
+          }
+          case "updateData":
+            await context.job.updateData(
+              Object.hasOwn(msg, "jobData") ? msg.jobData : msg.payload
+            );
+            msg.payload = serializeJob(context.job);
+            nodeSend(node, send, msg);
+            nodeDone(node, done);
+            return;
+          case "cancelJob": {
+            const reason = msg.reason || "BullMQ job cancelled";
+            const cancelled = context.worker.cancelJob(context.job.id, reason);
+            if (!cancelled) {
+              throw new Error(
+                `BullMQ found no cancellable processor for job ${context.job.id}`
+              );
+            }
+            msg.payload = cancelled;
+            nodeSend(node, send, msg);
+            nodeDone(node, done);
+            return;
+          }
+          case "cancelAllJobs": {
+            const reason = msg.reason || "BullMQ job cancelled";
+            context.worker.cancelAllJobs(reason);
+            msg.payload = true;
+            nodeSend(node, send, msg);
+            nodeDone(node, done);
+            return;
+          }
           default:
-            throw new Error(`Unsupported bull job action: ${action}`);
+            throw new Error(`Unsupported bullmq job action: ${action}`);
         }
       } catch (err) {
         nodeDone(node, done, err, msg);
@@ -560,13 +1017,23 @@ module.exports = function registerBullMQNodes(RED) {
 
     if (!node.bullConn) {
       node.status({ fill: "red", shape: "ring", text: "missing queue" });
-      node.error("Missing bull-queue-server config node");
+      node.error("Missing bullmq-queue-server config node");
       return;
     }
 
-    node.bullConn.register(node);
     node.queueEvents = node.bullConn.createQueueEvents(node);
-    attachErrorListener(node.queueEvents, node);
+    if (!node.queueEvents) {
+      // Construction failed synchronously and already reported why.
+      setDisconnected(node);
+      return;
+    }
+    const eventsStartupFailureOwner =
+      node.bullConn.config.backend === "postgres" ? node.bullConn : undefined;
+    const markQueueEventsReady = attachErrorListener(
+      node.queueEvents,
+      node,
+      eventsStartupFailureOwner,
+    );
     const events = parseEventFilter(n.events);
     for (const event of events) {
       node.queueEvents.on(event, (payload, eventId) => {
@@ -585,10 +1052,15 @@ module.exports = function registerBullMQNodes(RED) {
       try {
         setConnecting(node);
         await node.queueEvents.waitUntilReady();
+        markQueueEventsReady();
         setConnected(node);
       } catch (err) {
         setDisconnected(node);
-        node.error(err);
+        if (eventsStartupFailureOwner) {
+          eventsStartupFailureOwner.reportBackendFailure(err);
+        } else {
+          node.error(err);
+        }
       }
     }
     updateReadyStatus();
@@ -596,7 +1068,6 @@ module.exports = function registerBullMQNodes(RED) {
     node.on("close", async function onClose(removed, done) {
       try {
         await node.bullConn.releaseResource(node.queueEvents);
-        node.bullConn.deregister(node, () => {});
         done();
       } catch (err) {
         done(err);
@@ -612,21 +1083,36 @@ module.exports = function registerBullMQNodes(RED) {
 
     if (!node.bullConn) {
       node.status({ fill: "red", shape: "ring", text: "missing queue" });
-      node.error("Missing bull-queue-server config node");
+      node.error("Missing bullmq-queue-server config node");
       return;
     }
 
-    node.bullConn.register(node);
     node.flowProducer = node.bullConn.createFlowProducer(node);
-    attachErrorListener(node.flowProducer, node);
+    if (!node.flowProducer) {
+      // Construction failed synchronously and already reported why.
+      setDisconnected(node);
+      return;
+    }
+    const flowStartupFailureOwner =
+      node.bullConn.config.backend === "postgres" ? node.bullConn : undefined;
+    const markFlowProducerReady = attachErrorListener(
+      node.flowProducer,
+      node,
+      flowStartupFailureOwner,
+    );
     async function updateReadyStatus() {
       try {
         setConnecting(node);
         await node.flowProducer.waitUntilReady();
+        markFlowProducerReady();
         setConnected(node);
       } catch (err) {
         setDisconnected(node);
-        node.error(err);
+        if (flowStartupFailureOwner) {
+          flowStartupFailureOwner.reportBackendFailure(err);
+        } else {
+          node.error(err);
+        }
       }
     }
     updateReadyStatus();
@@ -634,11 +1120,29 @@ module.exports = function registerBullMQNodes(RED) {
     node.on("input", async function onInput(msg, send, done) {
       try {
         if (!msg.payload || typeof msg.payload !== "object") {
-          throw new Error("bull flow requires msg.payload to contain a flow tree");
+          throw new Error(
+            "bullmq flow requires msg.payload to contain a flow tree, or an array of flow trees"
+          );
+        }
+        const retention = node.bullConn.config.defaultJobOptions;
+        if (Array.isArray(msg.payload)) {
+          // addBulk creates every tree or none of them, which is the whole
+          // reason to use it instead of one add() per tree.
+          if (msg.payload.length === 0) {
+            throw new Error("bullmq flow requires at least one flow tree");
+          }
+          const trees = await node.flowProducer.addBulk(
+            msg.payload.map((flow) => withBulkFlowJobDefaults(flow, retention))
+          );
+          msg.payload = trees.map(serializeFlowJob);
+        } else {
+          msg.payload = serializeFlowJob(
+            await node.flowProducer.add(
+              msg.payload,
+              withFlowJobDefaults(msg.payload, msg.flowopts, retention)
+            )
+          );
         }
-        msg.payload = serializeFlowJob(
-          await node.flowProducer.add(msg.payload, msg.flowopts)
-        );
         nodeSend(node, send, msg);
         nodeDone(node, done);
       } catch (err) {
@@ -649,7 +1153,6 @@ module.exports = function registerBullMQNodes(RED) {
     node.on("close", async function onClose(removed, done) {
       try {
         await node.bullConn.releaseResource(node.flowProducer);
-        node.bullConn.deregister(node, () => {});
         done();
       } catch (err) {
         done(err);
@@ -657,9 +1160,9 @@ module.exports = function registerBullMQNodes(RED) {
     });
   }
 
-  RED.nodes.registerType("bull cmd", BullQueueCmdNode);
-  RED.nodes.registerType("bull run", BullQueueRunNode);
-  RED.nodes.registerType("bull job", BullJobNode);
-  RED.nodes.registerType("bull events", BullEventsNode);
-  RED.nodes.registerType("bull flow", BullFlowNode);
+  RED.nodes.registerType("bullmq cmd", BullQueueCmdNode);
+  RED.nodes.registerType("bullmq run", BullQueueRunNode);
+  RED.nodes.registerType("bullmq job", BullJobNode);
+  RED.nodes.registerType("bullmq events", BullEventsNode);
+  RED.nodes.registerType("bullmq flow", BullFlowNode);
 };