Greptile Overview

Summary

This PR significantly enhances the telemetry infrastructure by migrating from a basic implementation to a comprehensive OpenTelemetry-based system with proper semantic conventions for GenAI operations.

Key changes:

• Refactored telemetry from lib/telemetry.ts into a dedicated lib/telemetry/tracer.ts with GenAI semantic conventions
• Implemented client-side event batching (10s intervals or 50 events) with nested sanitization to reduce overhead
• Added OpenTelemetry trace span generation from workflow execution logs with proper parent-child relationships
• Enhanced trace spans UI with new modular components (trace-spans.tsx, trace-span-item.tsx) for better visualization
• Improved token normalization to handle multiple formats (object with input/output/prompt/completion fields)
• Added platform event tracking for workflow deployment, undeployment, and execution outcomes
• Updated batch processor settings for better performance (2048 queue size, 512 batch size)

Technical improvements:

• Streaming duration calculation now correctly extends to block end time
• Trace span building handles nested workflows and iteration blocks properly
• Client-side telemetry uses navigator.sendBeacon for reliable delivery before page unload

Confidence Score: 4/5

• Safe to merge with minor style improvements recommended
• The PR implements a comprehensive telemetry refactoring with proper error handling and privacy considerations. The code is well-structured with good separation of concerns. However, there are several type assertions to any that violate the project's custom instruction (c63aedff-69e6-48d8-81cf-9763416ee01c), and the nested sanitization could be more robust by scanning string values for sensitive patterns. These are non-critical style issues that don't affect functionality.
• Pay attention to apps/sim/lib/telemetry/tracer.ts and apps/sim/instrumentation-client.ts for type safety and sanitization improvements

Important Files Changed

File Analysis

Sequence Diagram

sequenceDiagram
    participant Client as Browser Client
    participant API as Telemetry API
    participant Collector as OTel Collector
    participant Executor as Workflow Executor
    participant Logger as Logging Session
    participant Tracer as OTel Tracer

    Note over Client: Client-side Telemetry
    Client->>Client: Batch events (10s or 50 events)
    Client->>Client: Sanitize events (remove sensitive data)
    Client->>API: POST /api/telemetry (batch)
    API->>API: Validate & sanitize data
    API->>Collector: Forward to OTel endpoint
    
    Note over Executor,Tracer: Server-side Workflow Execution
    Executor->>Logger: Start workflow execution
    Logger->>Logger: Create execution log entry
    Executor->>Executor: Execute workflow blocks
    Executor->>Logger: Log block executions
    Logger->>Logger: Build trace spans from logs
    Executor->>Logger: Complete workflow execution
    Logger->>Tracer: createOTelSpansForWorkflowExecution()
    Tracer->>Tracer: Convert TraceSpans to OTel spans
    Tracer->>Collector: Export spans via BatchSpanProcessor
    Logger->>Tracer: trackPlatformEvent('workflow.executed')
    Tracer->>Collector: Send platform event span

_{28 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}