@automagik/omni 2.260619.3 → 2.260622.1

package/dist/index.js

@@ -124960,7 +124960,7 @@ import { fileURLToPath } from "url";
 // package.json
 var package_default = {
   name: "@automagik/omni",
-  version: "2.260619.3",
+  version: "2.260622.1",
   description: "LLM-optimized CLI for Omni",
   type: "module",
   bin: {

package/dist/server/index.js

@@ -225245,7 +225245,7 @@ var init_sentry_scrub = __esm(() => {
 var require_package7 = __commonJS((exports, module) => {
   module.exports = {
     name: "@omni/api",
-    version: "2.260619.3",
+    version: "2.260622.1",
     type: "module",
     exports: {
       ".": {
@@ -336856,6 +336856,175 @@ var init_turn_events = __esm(() => {
   sc2 = import_nats5.StringCodec();
 });
 
+// ../api/src/plugins/agent-dispatch-limiter.ts
+function parsePositiveInt(env2, key, fallback, logger6) {
+  const raw2 = env2[key];
+  if (raw2 === undefined || raw2 === "")
+    return fallback;
+  const parsed = Number.parseInt(raw2, 10);
+  if (!Number.isFinite(parsed) || parsed <= 0) {
+    logger6?.warn("Invalid agent dispatch limiter env value, using default", { key, value: raw2, fallback });
+    return fallback;
+  }
+  return parsed;
+}
+function loadAgentDispatchLimiterConfig(env2 = process.env, logger6) {
+  return {
+    defaultConcurrency: parsePositiveInt(env2, "OMNI_AGENT_DISPATCH_CONCURRENCY_DEFAULT", DEFAULT_AGENT_DISPATCH_CONCURRENCY, logger6),
+    maxQueueDepth: parsePositiveInt(env2, "OMNI_AGENT_DISPATCH_QUEUE_MAX_DEPTH", DEFAULT_AGENT_DISPATCH_QUEUE_MAX_DEPTH, logger6),
+    maxQueueWaitMs: parsePositiveInt(env2, "OMNI_AGENT_DISPATCH_QUEUE_MAX_WAIT_MS", DEFAULT_AGENT_DISPATCH_QUEUE_MAX_WAIT_MS, logger6),
+    perChatConcurrency: parsePositiveInt(env2, "OMNI_AGENT_DISPATCH_PER_CHAT_CONCURRENCY", DEFAULT_AGENT_DISPATCH_PER_CHAT_CONCURRENCY, logger6)
+  };
+}
+
+class AgentDispatchLimiter {
+  config;
+  logger;
+  queues = new Map;
+  chatActiveCounts = new Map;
+  constructor(config4, logger6) {
+    this.config = config4;
+    this.logger = logger6;
+  }
+  run(context20, run) {
+    const queue = this.getQueue(context20.instanceId);
+    if (this.canStart(queue, context20)) {
+      return new Promise((resolve2, reject) => {
+        this.startItem(queue, {
+          context: context20,
+          enqueuedAt: Date.now(),
+          run,
+          resolve: resolve2,
+          reject,
+          timeout: null
+        });
+      });
+    }
+    if (queue.pending.length >= this.config.maxQueueDepth) {
+      const snapshot = this.snapshot(queue, context20);
+      this.logger?.warn("agent_dispatch_queue_full", snapshot);
+      return Promise.reject(new AgentDispatchQueueFullError(snapshot));
+    }
+    return new Promise((resolve2, reject) => {
+      const item = {
+        context: context20,
+        enqueuedAt: Date.now(),
+        run,
+        resolve: resolve2,
+        reject,
+        timeout: null
+      };
+      item.timeout = setTimeout(() => {
+        const idx = queue.pending.indexOf(item);
+        if (idx < 0)
+          return;
+        queue.pending.splice(idx, 1);
+        const snapshot = this.snapshot(queue, context20, Date.now() - item.enqueuedAt);
+        this.logger?.warn("agent_dispatch_queue_timeout", snapshot);
+        reject(new AgentDispatchQueueTimeoutError(snapshot));
+        this.drain(queue);
+      }, this.config.maxQueueWaitMs);
+      if (typeof item.timeout === "object" && item.timeout && "unref" in item.timeout) {
+        item.timeout.unref();
+      }
+      queue.pending.push(item);
+      this.logger?.info("agent_dispatch_queue_enqueued", this.snapshot(queue, context20));
+      this.drain(queue);
+    });
+  }
+  getSnapshot(instanceId) {
+    const queue = this.getQueue(instanceId);
+    return {
+      activeCount: queue.activeCount,
+      queueDepth: queue.pending.length,
+      maxConcurrency: this.config.defaultConcurrency,
+      maxQueueDepth: this.config.maxQueueDepth
+    };
+  }
+  getQueue(instanceId) {
+    let queue = this.queues.get(instanceId);
+    if (!queue) {
+      queue = { instanceId, activeCount: 0, pending: [] };
+      this.queues.set(instanceId, queue);
+    }
+    return queue;
+  }
+  canStart(queue, context20) {
+    return queue.activeCount < this.config.defaultConcurrency && (this.chatActiveCounts.get(this.chatKey(context20)) ?? 0) < this.config.perChatConcurrency;
+  }
+  startItem(queue, item) {
+    if (item.timeout) {
+      clearTimeout(item.timeout);
+      item.timeout = null;
+    }
+    const chatKey = this.chatKey(item.context);
+    queue.activeCount += 1;
+    this.chatActiveCounts.set(chatKey, (this.chatActiveCounts.get(chatKey) ?? 0) + 1);
+    const queueWaitMs = Date.now() - item.enqueuedAt;
+    this.logger?.info("agent_dispatch_queue_started", this.snapshot(queue, item.context, queueWaitMs));
+    Promise.resolve().then(item.run).then(item.resolve, item.reject).finally(() => {
+      queue.activeCount = Math.max(0, queue.activeCount - 1);
+      const chatActive = (this.chatActiveCounts.get(chatKey) ?? 1) - 1;
+      if (chatActive > 0) {
+        this.chatActiveCounts.set(chatKey, chatActive);
+      } else {
+        this.chatActiveCounts.delete(chatKey);
+      }
+      this.logger?.info("agent_dispatch_queue_finished", this.snapshot(queue, item.context, Date.now() - item.enqueuedAt));
+      this.drain(queue);
+    });
+  }
+  drain(queue) {
+    let started = true;
+    while (started && queue.activeCount < this.config.defaultConcurrency) {
+      started = false;
+      for (let i = 0;i < queue.pending.length; i += 1) {
+        const item = queue.pending[i];
+        if (!item || !this.canStart(queue, item.context))
+          continue;
+        queue.pending.splice(i, 1);
+        this.startItem(queue, item);
+        started = true;
+        break;
+      }
+    }
+  }
+  snapshot(queue, context20, waitMs) {
+    return {
+      instanceId: context20.instanceId,
+      chatId: context20.chatId,
+      traceId: context20.traceId,
+      activeCount: queue.activeCount,
+      queueDepth: queue.pending.length,
+      maxConcurrency: this.config.defaultConcurrency,
+      maxQueueDepth: this.config.maxQueueDepth,
+      waitMs
+    };
+  }
+  chatKey(context20) {
+    return `${context20.instanceId}:${context20.chatId}`;
+  }
+}
+var DEFAULT_AGENT_DISPATCH_CONCURRENCY = 8, DEFAULT_AGENT_DISPATCH_QUEUE_MAX_DEPTH = 100, DEFAULT_AGENT_DISPATCH_QUEUE_MAX_WAIT_MS = 60000, DEFAULT_AGENT_DISPATCH_PER_CHAT_CONCURRENCY = 1, AgentDispatchQueueFullError, AgentDispatchQueueTimeoutError;
+var init_agent_dispatch_limiter = __esm(() => {
+  AgentDispatchQueueFullError = class AgentDispatchQueueFullError extends Error {
+    snapshot;
+    constructor(snapshot) {
+      super(`Agent dispatch queue full for instance ${snapshot.instanceId}: ${snapshot.queueDepth}/${snapshot.maxQueueDepth} pending`);
+      this.snapshot = snapshot;
+      this.name = "AgentDispatchQueueFullError";
+    }
+  };
+  AgentDispatchQueueTimeoutError = class AgentDispatchQueueTimeoutError extends Error {
+    snapshot;
+    constructor(snapshot) {
+      super(`Agent dispatch queue timed out for instance ${snapshot.instanceId} chat ${snapshot.chatId} after ${snapshot.waitMs ?? 0}ms`);
+      this.snapshot = snapshot;
+      this.name = "AgentDispatchQueueTimeoutError";
+    }
+  };
+});
+
 // ../api/src/plugins/logger.ts
 function createPluginLogger(module) {
   if (typeof module === "object") {
@@ -339810,17 +339979,23 @@ async function processAgentResponse(services, instance4, messages4, triggerType,
     senderName,
     senderAgentId
   });
-  await sendTypingPresence(channel5, instance4.id, chatId, "composing");
   const inst = instance4;
   const agentAckMessage = inst.agentAckMessage ?? null;
-  if (agentAckMessage) {
-    sendTextMessage5(channel5, instance4.id, chatId, agentAckMessage).catch((err) => {
-      log99.warn("Failed to send agent ack message", { instanceId: instance4.id, chatId, error: String(err) });
-    });
-  }
-  const perThreadExtraContext = await resolvePerThreadExtraContext(db2, services, instance4, channel5, chatId, senderId, firstMessage);
   try {
-    await runWithTransientDispatchRetry(() => dispatchToAgent(services, instance4, messages4, triggerType, channel5, chatId, senderId, personId, senderName, traceId, senderAgentId, perThreadExtraContext, db2), { instanceId: instance4.id, chatId, traceId });
+    await agentDispatchLimiter.run({ instanceId: instance4.id, chatId, traceId }, async () => {
+      await sendTypingPresence(channel5, instance4.id, chatId, "composing");
+      try {
+        if (agentAckMessage) {
+          sendTextMessage5(channel5, instance4.id, chatId, agentAckMessage).catch((err) => {
+            log99.warn("Failed to send agent ack message", { instanceId: instance4.id, chatId, error: String(err) });
+          });
+        }
+        const perThreadExtraContext = await resolvePerThreadExtraContext(db2, services, instance4, channel5, chatId, senderId, firstMessage);
+        await runWithTransientDispatchRetry(() => dispatchToAgent(services, instance4, messages4, triggerType, channel5, chatId, senderId, personId, senderName, traceId, senderAgentId, perThreadExtraContext, db2), { instanceId: instance4.id, chatId, traceId });
+      } finally {
+        await sendTypingPresence(channel5, instance4.id, chatId, "paused");
+      }
+    });
   } catch (error3) {
     log99.error("agent_dispatch_failed_after_retries", {
       instanceId: instance4.id,
@@ -339831,7 +340006,6 @@ async function processAgentResponse(services, instance4, messages4, triggerType,
     sendErrorFeedback(channel5, instance4.id, chatId, error3).catch(() => {});
   } finally {
     ackHandle.remove();
-    await sendTypingPresence(channel5, instance4.id, chatId, "paused");
   }
 }
 function resolveRequiredAgentId(instance4, schemaConfig, providerId, fieldName = "agentId") {
@@ -340956,7 +341130,7 @@ async function setupAgentDispatcher(eventBus, services, db2) {
     log99.info("Agent dispatcher shutdown complete");
   };
 }
-var import_api32, log99, _natsGenieProviderCtor, QUOTED_MESSAGE_MAX_CHARS = 4000, DM_HISTORY_LIMIT = 20, LIFECYCLE_PREVIEW_MAX_CHARS = 160, LIFECYCLE_SENSITIVE_KEY_PARTS, TRANSIENT_DISPATCH_ERROR_PATTERNS, TRANSIENT_DISPATCH_RETRY_DELAYS_MS, CHANNEL_MESSAGE_LIMITS, DEFAULT_MESSAGE_LIMIT = 4000, MEDIA_BASE_PATH3, MEDIA_ICONS, MEDIA_WAIT_NULL, mediaCompletions, mediaResultCache, MEDIA_WAIT_TIMEOUT_MS = 30000, DEFAULT_SEND_MEDIA_PATH_TYPES, BOT_PREFIX = "\uD83E\uDD16 ", activeStreams, sessionActivityStore, PROC_REACT_START, PROC_REACT_DONE = "\u2705", providerCache, openclawClientPool, nullFilterWarnedInstances, ACTIVE_OWNER_IDENTIFIER_CACHE_TTL_MS = 1e4, cachedActiveOwnerIdentifiers = null, cachedActiveOwnerIdentifiersAt = 0, DEFAULT_GATE_MODEL = "gemini-3-flash-preview", GATE_TIMEOUT_MS = 3000, setupAgentResponder;
+var import_api32, log99, agentDispatchLimiter, _natsGenieProviderCtor, QUOTED_MESSAGE_MAX_CHARS = 4000, DM_HISTORY_LIMIT = 20, LIFECYCLE_PREVIEW_MAX_CHARS = 160, LIFECYCLE_SENSITIVE_KEY_PARTS, TRANSIENT_DISPATCH_ERROR_PATTERNS, TRANSIENT_DISPATCH_RETRY_DELAYS_MS, CHANNEL_MESSAGE_LIMITS, DEFAULT_MESSAGE_LIMIT = 4000, MEDIA_BASE_PATH3, MEDIA_ICONS, MEDIA_WAIT_NULL, mediaCompletions, mediaResultCache, MEDIA_WAIT_TIMEOUT_MS = 30000, DEFAULT_SEND_MEDIA_PATH_TYPES, BOT_PREFIX = "\uD83E\uDD16 ", activeStreams, sessionActivityStore, PROC_REACT_START, PROC_REACT_DONE = "\u2705", providerCache, openclawClientPool, nullFilterWarnedInstances, ACTIVE_OWNER_IDENTIFIER_CACHE_TTL_MS = 1e4, cachedActiveOwnerIdentifiers = null, cachedActiveOwnerIdentifiersAt = 0, DEFAULT_GATE_MODEL = "gemini-3-flash-preview", GATE_TIMEOUT_MS = 3000, setupAgentResponder;
 var init_agent_dispatcher = __esm(() => {
   init_src2();
   init_src();
@@ -340968,6 +341142,7 @@ var init_agent_dispatcher = __esm(() => {
   init_sentry_scrub();
   init_agent_runner();
   init_turn_events();
+  init_agent_dispatch_limiter();
   init_loader2();
   init_message_debouncer();
   init_message_persistence();
@@ -340975,6 +341150,7 @@ var init_agent_dispatcher = __esm(() => {
   init_src6();
   import_api32 = __toESM(require_src(), 1);
   log99 = createLogger("agent-dispatcher");
+  agentDispatchLimiter = new AgentDispatchLimiter(loadAgentDispatchLimiterConfig(process.env, log99), log99);
   _natsGenieProviderCtor = NatsGenieProvider;
   LIFECYCLE_SENSITIVE_KEY_PARTS = ["authorization", "bearer", "password", "secret", "token", "api_key", "apikey"];
   TRANSIENT_DISPATCH_ERROR_PATTERNS = [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@automagik/omni",
-  "version": "2.260619.3",
+  "version": "2.260622.1",
   "description": "LLM-optimized CLI for Omni",
   "type": "module",
   "bin": {