@output.ai/core 0.3.5 → 0.3.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@output.ai/core",
-  "version": "0.3.5",
+  "version": "0.3.6",
   "description": "The core module of the output framework",
   "type": "module",
   "exports": {

package/src/worker/configs.js

@@ -6,7 +6,17 @@ const envVarSchema = z.object( {
   CATALOG_ID: z.string().regex( /^[a-z0-9_.@-]+$/i ),
   TEMPORAL_ADDRESS: z.string().default( 'localhost:7233' ),
   TEMPORAL_API_KEY: z.string().optional(),
-  TEMPORAL_NAMESPACE: z.string().optional().default( 'default' )
+  TEMPORAL_NAMESPACE: z.string().optional().default( 'default' ),
+  // Worker concurrency — tune these via env vars to adjust for your workload.
+  // Each step (API, LLM, etc.) call is one activity. Lower this to reduce memory pressure.
+  MAX_CONCURRENT_ACTIVITY_TASKS: z.coerce.number().int().positive().default( 40 ),
+  // Workflows are lightweight state machines — this can be high.
+  MAX_CONCURRENT_WORKFLOW_TASKS: z.coerce.number().int().positive().default( 200 ),
+  // LRU cache for sticky workflow execution. Lower values free memory faster after surges.
+  MAX_CACHED_WORKFLOWS: z.coerce.number().int().positive().default( 1000 ),
+  // How aggressively the worker pulls tasks from Temporal.
+  MAX_CONCURRENT_ACTIVITY_POLLS: z.coerce.number().int().positive().default( 5 ),
+  MAX_CONCURRENT_WORKFLOW_POLLS: z.coerce.number().int().positive().default( 5 )
 } );
 
 const { data: envVars, error } = envVarSchema.safeParse( process.env );
@@ -17,8 +27,11 @@ if ( error ) {
 export const address = envVars.TEMPORAL_ADDRESS;
 export const apiKey = envVars.TEMPORAL_API_KEY;
 export const executionTimeout = '1m';
-export const maxActivities = 100;
-export const maxWorkflows = 100;
+export const maxActivities = envVars.MAX_CONCURRENT_ACTIVITY_TASKS;
+export const maxWorkflows = envVars.MAX_CONCURRENT_WORKFLOW_TASKS;
+export const maxCachedWorkflows = envVars.MAX_CACHED_WORKFLOWS;
+export const maxActivityPolls = envVars.MAX_CONCURRENT_ACTIVITY_POLLS;
+export const maxWorkflowPolls = envVars.MAX_CONCURRENT_WORKFLOW_POLLS;
 export const namespace = envVars.TEMPORAL_NAMESPACE;
 export const taskQueue = envVars.CATALOG_ID;
 export const catalogId = envVars.CATALOG_ID;

package/src/worker/index.js

@@ -1,7 +1,10 @@
 import { Worker, NativeConnection } from '@temporalio/worker';
 import { Client } from '@temporalio/client';
 import { WorkflowIdConflictPolicy } from '@temporalio/common';
-import { address, apiKey, maxActivities, maxWorkflows, namespace, taskQueue, catalogId } from './configs.js';
+import {
+  address, apiKey, maxActivities, maxWorkflows, maxCachedWorkflows,
+  maxActivityPolls, maxWorkflowPolls, namespace, taskQueue, catalogId
+} from './configs.js';
 import { loadActivities, loadWorkflows, createWorkflowsEntryPoint } from './loader.js';
 import { sinks } from './sinks.js';
 import { createCatalog } from './catalog_workflow/index.js';
@@ -47,6 +50,9 @@ const callerDir = process.argv[2];
     interceptors: initInterceptors( { activities } ),
     maxConcurrentWorkflowTaskExecutions: maxWorkflows,
     maxConcurrentActivityTaskExecutions: maxActivities,
+    maxCachedWorkflows,
+    maxConcurrentActivityTaskPolls: maxActivityPolls,
+    maxConcurrentWorkflowTaskPolls: maxWorkflowPolls,
     bundlerOptions: { webpackConfigHook }
   } );