PSA/docs/workflow/distributed-implementation-plan.md

Asynchronous Workflow Implementation Update

Overview

This document provides an update to our asynchronous workflow implementation plan based on an analysis of the current codebase. While many of the foundational components exist, there are critical integration gaps that need to be addressed to achieve a fully asynchronous workflow system.

Current Implementation Status

Components Already Implemented

1. Redis Stream Client (redisStreamClient.ts)

   • Full implementation for publishing/consuming events via Redis Streams
   • Support for consumer groups and pending message claiming
   • Message acknowledgment and error handling

2. Distributed Lock Mechanism (distributedLock.ts)

   • Complete Redis-based distributed lock implementation
   • Functions for acquiring, releasing, and extending locks
   • Helper functions for executing code with locks

3. Distributed Transaction Support (distributedTransaction.ts)

   • Integration of database transactions with distributed locks
   • Error handling and retry mechanisms

4. Worker Framework (workflowWorker.ts)

   • Robust implementation with health monitoring and metrics
   • Support for concurrent event processing and pending message claiming
   • Graceful shutdown handling

Critical Integration Gaps

1. Missing Workflow Runtime Integration

   • The TypeScriptWorkflowRuntime class in workflowRuntime.ts still uses in-memory state tracking
   • No enqueueEvent method for "fire and forget" event submission
   • No processQueuedEvent method referenced by the worker
   • No proper event sourcing pattern for state derivation

2. Incomplete Database Integration

   • The event processing record handling only partially implemented
   • No transaction management for ensuring atomic operations

3. Missing Runtime-Worker Connection

   • The worker is implemented but not properly connected to the workflow runtime
   • No service startup/initialization for workers

Implementation Steps

1. Enhance TypeScriptWorkflowRuntime

Update server/src/lib/workflow/core/workflowRuntime.ts to add support for asynchronous processing:

interface ProcessQueuedEventParams {
  eventId: string;
  executionId: string;
  processingId: string;
  workerId: string;
  tenant: string;
}

interface ProcessQueuedEventResult {
  success: boolean;
  errorMessage?: string;
  previousState?: string;
  currentState?: string;
  actionsExecuted: Array<{
    actionName: string;
    success: boolean;
    error?: string;
  }>;
}

export class TypeScriptWorkflowRuntime {
  // ... existing code ...

  /**
   * Enqueues an event for asynchronous processing
   * Returns quickly after persisting the event and publishing to Redis
   */
  async enqueueEvent(options: EventSubmissionOptions): Promise<{ eventId: string }> {
    const { execution_id, event_name, payload, user_id, tenant } = options;
    
    // Generate a unique event ID
    const eventId = `evt-${Date.now()}-${Math.random().toString(36).substring(2, 9)}`;
    
    // Get Redis stream client
    const redisStreamClient = getRedisStreamClient();
    
    // Verify execution exists and is valid
    const executionState = this.executionStates.get(execution_id);
    if (!executionState) {
      throw new Error(`Workflow execution "${execution_id}" not found`);
    }
    
    // Verify tenant
    if (executionState.tenant !== tenant) {
      throw new Error(`Tenant mismatch for workflow execution "${execution_id}"`);
    }
    
    // Create event
    const event: WorkflowEvent = {
      id: eventId,
      name: event_name,
      payload: payload || {},
      user_id,
      timestamp: new Date().toISOString()
    };
    
    // Use distributed transaction to persist event and publish to Redis
    await executeDistributedTransaction(knex, `workflow:${execution_id}`, async (trx) => {
      // Persist event to database
      await this.persistEventToDatabase(execution_id, tenant, event, trx);
      
      // Create processing record
      await this.createEventProcessingRecord(eventId, tenant, trx);
      
      // Publish to Redis stream
      await redisStreamClient.publishEvent({
        id: eventId,
        executionId: execution_id,
        eventName: event_name,
        tenant,
        userId: user_id,
        payload: payload || {},
        timestamp: event.timestamp
      });
    });
    
    return { eventId };
  }
  
  /**
   * Process a queued event from Redis stream
   * Called by worker processes to handle events asynchronously
   */
  async processQueuedEvent(
    params: ProcessQueuedEventParams
  ): Promise<ProcessQueuedEventResult> {
    const { eventId, executionId, processingId, workerId, tenant } = params;
    
    // Acquire distributed lock to ensure only one worker processes this event
    const lockKey = `event:${eventId}:processing`;
    const lockOwner = `worker:${workerId}`;
    
    const lockAcquired = await acquireDistributedLock(lockKey, lockOwner, {
      waitTimeMs: 5000,
      ttlMs: 60000
    });
    
    if (!lockAcquired) {
      return {
        success: false,
        errorMessage: 'Failed to acquire lock for event processing',
        actionsExecuted: []
      };
    }
    
    try {
      // Mark event as processing
      await this.updateEventProcessingStatus(processingId, 'processing', {
        workerId,
        startedAt: new Date().toISOString()
      });
      
      // Load execution data and events
      const workflowDef = this.workflowDefinitions.get(executionId);
      if (!workflowDef) {
        throw new Error(`Workflow "${executionId}" not found`);
      }
      
      // Load all events from database
      const eventLog = await this.loadEventsFromDatabase(executionId, tenant);
      
      // Find the event being processed
      const currentEvent = eventLog.find(e => e.id === eventId);
      if (!currentEvent) {
        throw new Error(`Event ${eventId} not found`);
      }
      
      // Derive state by replaying events
      const replayState = this.replayEvents(workflowDef, eventLog.filter(e => e.id !== eventId));
      const previousState = replayState.currentState;
      
      // Execute actions
      const actionResults = await this.executeActionsForEvent(
        workflowDef,
        executionId,
        currentEvent,
        tenant
      );
      
      // Update workflow state
      const currentState = this.getExecutionState(executionId, tenant).currentState;
      
      // Mark event as completed
      await this.updateEventProcessingStatus(processingId, 'completed', {
        workerId,
        completedAt: new Date().toISOString(),
        result: {
          previousState,
          currentState,
          actionsExecuted: actionResults.map(a => a.actionName)
        }
      });
      
      return {
        success: true,
        previousState,
        currentState,
        actionsExecuted: actionResults
      };
    } catch (error) {
      // Mark event as failed
      await this.updateEventProcessingStatus(processingId, 'failed', {
        workerId,
        completedAt: new Date().toISOString(),
        error: error instanceof Error ? error.message : String(error)
      });
      
      return {
        success: false,
        errorMessage: error instanceof Error ? error.message : String(error),
        actionsExecuted: []
      };
    } finally {
      // Release lock
      await releaseDistributedLock(lockKey, lockOwner);
    }
  }
  
  /**
   * Helper methods
   */
  private async persistEventToDatabase(
    executionId: string,
    tenant: string,
    event: WorkflowEvent,
    trx: Knex.Transaction
  ): Promise<void> {
    // Implementation to persist event to database
  }
  
  private async createEventProcessingRecord(
    eventId: string,
    tenant: string, 
    trx: Knex.Transaction
  ): Promise<void> {
    // Implementation to create processing record
  }
  
  private async updateEventProcessingStatus(
    processingId: string,
    status: 'pending' | 'processing' | 'completed' | 'failed',
    metadata: Record<string, any>
  ): Promise<void> {
    // Implementation to update processing status
  }
  
  private async loadEventsFromDatabase(
    executionId: string,
    tenant: string
  ): Promise<WorkflowEvent[]> {
    // Implementation to load events from database
  }
  
  private replayEvents(
    workflowDef: WorkflowDefinition,
    events: WorkflowEvent[]
  ): { currentState: string; data: Record<string, any> } {
    // Implementation to replay events and derive state
  }
  
  private async executeActionsForEvent(
    workflowDef: WorkflowDefinition,
    executionId: string,
    event: WorkflowEvent,
    tenant: string
  ): Promise<Array<{ actionName: string; success: boolean; error?: string }>> {
    // Implementation to execute actions for an event
  }
}

2. Implement Event Persistence Model

Create or enhance server/src/lib/workflow/persistence/workflowEventModel.ts:

import { Knex } from 'knex';
import { WorkflowEvent } from '../core/workflowContext';

export class WorkflowEventModel {
  static async persistEvent(
    event: WorkflowEvent,
    executionId: string,
    tenant: string,
    fromState: string,
    trx?: Knex.Transaction
  ): Promise<void> {
    const db = trx || knex;
    
    await db('workflow_events').insert({
      event_id: event.id,
      execution_id: executionId,
      event_name: event.name,
      from_state: fromState,
      to_state: '', // Will be determined during processing
      payload: JSON.stringify(event.payload),
      user_id: event.user_id,
      tenant,
      timestamp: event.timestamp || new Date().toISOString()
    });
  }
  
  static async loadEventLog(
    executionId: string,
    tenant: string
  ): Promise<WorkflowEvent[]> {
    const events = await knex('workflow_events')
      .where({
        execution_id: executionId,
        tenant
      })
      .orderBy('timestamp', 'asc');
    
    return events.map(e => ({
      id: e.event_id,
      name: e.event_name,
      payload: JSON.parse(e.payload),
      user_id: e.user_id,
      timestamp: e.timestamp,
      fromState: e.from_state,
      toState: e.to_state
    }));
  }
  
  static async updateEventStateTransition(
    eventId: string,
    tenant: string,
    toState: string,
    trx?: Knex.Transaction
  ): Promise<void> {
    const db = trx || knex;
    
    await db('workflow_events')
      .where({
        event_id: eventId,
        tenant
      })
      .update({
        to_state: toState,
        updated_at: new Date().toISOString()
      });
  }
}

3. Integrate Worker with Workflow Runtime

Enhance the existing WorkflowWorker class to properly integrate with the TypeScriptWorkflowRuntime:

export class WorkflowWorker {
  // ... existing code ...
  
  constructor(private workflowRuntime: TypeScriptWorkflowRuntime) {
    // Generate a unique worker ID
    this.workerId = `${os.hostname()}-${process.pid}-${generateUniqueId()}`;
    
    // Create Redis Stream client
    this.redisStreamClient = getRedisStreamClient();
  }
  
  // ... modify processEvent method ...
  
  private async processEvent(
    eventId: string,
    executionId: string,
    processingId: string,
    tenant: string
  ): Promise<void> {
    const startTime = Date.now();
    this.eventsProcessed++;
    
    try {
      // Create parameters for processing the queued event
      const params: ProcessQueuedEventParams = {
        eventId,
        executionId,
        processingId,
        workerId: this.workerId,
        tenant
      };
      
      // Use the workflow runtime to process the event
      const result = await this.workflowRuntime.processQueuedEvent(params);
      
      if (!result.success) {
        logger.error(`[WorkflowWorker] Failed to process event ${eventId}: ${result.errorMessage}`, {
          workerId: this.workerId,
          eventId,
          executionId,
          error: result.errorMessage
        });
        this.eventsFailed++;
      } else {
        logger.info(`[WorkflowWorker] Successfully processed event ${eventId}`, {
          workerId: this.workerId,
          eventId,
          executionId,
          previousState: result.previousState,
          currentState: result.currentState,
          actionsExecuted: result.actionsExecuted.map(a => a.actionName)
        });
        
        this.eventsSucceeded++;
      }
      
      // Record processing time
      const processingTime = Date.now() - startTime;
      this.processingTimes.push(processingTime);
      
    } catch (error) {
      // ... existing error handling ...
    }
  }
}

4. Create Worker Service for Application Startup

Create server/src/services/workflowWorkerService.ts to initialize and manage worker instances:

import { getWorkflowRuntime } from '@shared/workflow/core/workflowRuntime';
import { getActionRegistry } from '@shared/workflow/core/actionRegistry'; 
import { startWorkflowWorker, WorkflowWorker } from '../lib/workflow/workers/workflowWorker';
import logger from '../utils/logger';

export class WorkflowWorkerService {
  private workers: WorkflowWorker[] = [];
  private workerCount: number;
  
  constructor(workerCount: number = 2) {
    this.workerCount = workerCount;
  }
  
  async start(): Promise<void> {
    logger.info(`[WorkflowWorkerService] Starting ${this.workerCount} workflow workers`);
    
    try {
      // Initialize action registry and workflow runtime
      const actionRegistry = getActionRegistry();
      const workflowRuntime = getWorkflowRuntime(actionRegistry);
      
      // Start worker instances
      for (let i = 0; i < this.workerCount; i++) {
        const worker = await startWorkflowWorker(workflowRuntime);
        this.workers.push(worker);
        logger.info(`[WorkflowWorkerService] Started worker ${i + 1} of ${this.workerCount}`);
      }
      
      logger.info('[WorkflowWorkerService] All workflow workers started successfully');
    } catch (error) {
      logger.error('[WorkflowWorkerService] Failed to start workflow workers:', error);
      throw error;
    }
  }
  
  async stop(): Promise<void> {
    logger.info(`[WorkflowWorkerService] Stopping ${this.workers.length} workflow workers`);
    
    // Stop all workers
    await Promise.all(this.workers.map(async (worker) => {
      try {
        await worker.stop();
      } catch (error) {
        logger.error('[WorkflowWorkerService] Error stopping worker:', error);
      }
    }));
    
    this.workers = [];
    logger.info('[WorkflowWorkerService] All workflow workers stopped');
  }
}

// Singleton instance
let workerService: WorkflowWorkerService | null = null;

export function getWorkflowWorkerService(workerCount?: number): WorkflowWorkerService {
  if (!workerService) {
    workerService = new WorkflowWorkerService(workerCount);
  }
  return workerService;
}

5. Update Configuration

In server/src/config/index.ts, update configuration for workflow system:

export const config = {
  // ... existing config ...
  
  workflow: {
    // Number of worker instances to run
    workerCount: parseInt(process.env.WORKFLOW_WORKER_COUNT || '2', 10),
    
    // Redis stream configuration
    redis: {
      streamPrefix: process.env.WORKFLOW_REDIS_STREAM_PREFIX || 'workflow:events:',
      consumerGroup: process.env.WORKFLOW_REDIS_CONSUMER_GROUP || 'workflow-workers',
      batchSize: parseInt(process.env.WORKFLOW_REDIS_BATCH_SIZE || '10', 10),
      idleTimeoutMs: parseInt(process.env.WORKFLOW_REDIS_IDLE_TIMEOUT_MS || '60000', 10),
    }
  }
};

6. Update API Controllers to Use Enqueue

Update server/src/controllers/workflowController.ts:

import { config } from '../config';
import { getWorkflowRuntime } from '@shared/workflow/core/workflowRuntime';
import { Request, Response } from 'express';

export class WorkflowController {
  async submitEvent(req: Request, res: Response): Promise<void> {
    const { execution_id, event_name, payload, tenant } = req.body;
    const user_id = req.user?.id;
    
    try {
      const runtime = getWorkflowRuntime();
      
      // Enqueue event for asynchronous processing
      const result = await runtime.enqueueEvent({
        execution_id,
        event_name,
        payload,
        user_id,
        tenant
      });
      
      res.status(202).json({
        message: 'Event accepted for processing',
        eventId: result.eventId
      });
    } catch (error) {
      // Handle errors
      res.status(400).json({
        error: error instanceof Error ? error.message : 'Unknown error'
      });
    }
  }
}

Testing Strategy

1. Unit Tests for New Components

   • Test enqueueEvent with Redis Stream publishing
   • Test processQueuedEvent for event replay and processing
   • Test asynchronous event flow from submission to processing

2. Integration Tests with Worker

   • Test worker processing of Redis stream messages
   • Test worker coordination with distributed locks
   • Test failure recovery and message reclaiming

3. End-to-End Asynchronous Workflow Tests

   • Test complete workflows running with asynchronous processing
   • Test system resilience to worker restarts
   • Test workflow resumption after application restart

Implementation Roadmap

1. Phase 1: Core Implementation (2 weeks)

   • Implement enqueueEvent method in TypeScriptWorkflowRuntime
   • Implement processQueuedEvent method in TypeScriptWorkflowRuntime
   • Implement event persistence and loading from database

2. Phase 2: Integration (1 week)

   • Connect WorkflowWorker with TypeScriptWorkflowRuntime
   • Create WorkflowWorkerService for application startup
   • Update configuration for asynchronous processing

3. Phase 3: Testing and Validation (2 weeks)

   • Develop unit and integration tests
   • Test with multiple workers
   • Validate worker failover and recovery

4. Phase 4: Deployment and Monitoring (1 week)

   • Deploy to staging environment
   • Implement monitoring for worker health
   • Document deployment requirements

Conclusion

The implementation plan described here will transform our current workflow system into a fully asynchronous execution platform. By focusing on the specific integration points required between existing components, we can leverage much of the code that already exists while filling in the critical gaps.

Key success factors for this implementation:

1. Proper event sourcing implementation: Ensure that workflow state is derived by replaying events
2. Robust error handling: Implement thorough error handling and recovery mechanisms
3. Idempotent processing: Guarantee that event processing is idempotent for reliability
4. Comprehensive testing: Thoroughly test worker coordination and failure scenarios

With these enhancements, our workflow system will gain the resilience, scalability, and fault tolerance necessary for handling mission-critical business processes in an asynchronous environment.