Monitoring Dashboard

PlanAI includes a built-in web-based monitoring dashboard that provides real-time insights into your graph execution. This feature helps you understand workflow behavior, debug issues, and track performance.

Overview

The monitoring dashboard offers:

• Real-time Updates: Live task status updates using server-sent events (SSE)
• Task Organization: Tasks grouped into Queued, Active, and Completed categories
• Detailed Information: Click on any task to view provenance and input details
• Performance Metrics: Track execution times and identify bottlenecks

Enabling the Dashboard

Web Dashboard

To enable the web-based dashboard, set run_dashboard=True when calling graph.run():

graph.run(initial_tasks, run_dashboard=True)

By default, the dashboard will be accessible at http://localhost:5000. You can customize the port:

graph.run(
    initial_tasks,
    run_dashboard=True,
    dashboard_port=8080  # Custom port
)

Terminal Monitoring

For a lightweight terminal-based monitoring option, use the builtin terminal dashboard which is enabled automatically and shows when executing a graph via run or execute.

Dashboard Features

Task Status Tracking

The dashboard displays tasks in three categories:

1. Queued: Tasks waiting to be processed
2. Active: Tasks currently being processed
3. Completed: Tasks that have finished processing

Each task card shows:

• Task ID
• Task type
• Assigned worker
• Processing status

Task Details View

Click on any task to view detailed information:

# Example of information available in task details
{
    "task_id": "abc123",
    "type": "ProcessedData",
    "worker": "DataProcessor",
    "status": "completed",
    "provenance": [...],  # Full provenance chain
    "input_provenance": [...],  # Input task lineage
    "data": {...},  # Task data (if available)
    "processing_time": 1.23  # seconds
}

Real-time Updates

The dashboard uses server-sent events for live updates without page refreshes:

Using the Dashboard for Debugging

Identifying Bottlenecks

Look for patterns in the Active tasks section:

• Tasks stuck in Active state may indicate processing issues
• Large queues for specific workers suggest bottlenecks

Tracing Task Flow

Use the provenance information to understand task flow:

1. Click on a completed task
2. View its input provenance to see what led to this task
3. Trace back through the workflow to understand the full pipeline

Monitoring LLM Tasks

For LLMTaskWorker tasks, the dashboard shows:

• LLM processing status
• Token usage (when available)
• Response times

Production Considerations

Performance Impact

The dashboard has minimal performance impact:

• Uses asynchronous updates
• Lightweight SSE connections
• Optional feature that can be disabled

Security

For production deployments:

# Restrict dashboard access to localhost only
graph.run(
    initial_tasks,
    run_dashboard=True,
    dashboard_host="127.0.0.1"  # Local access only
)

Alternative Monitoring

PlanAI emits extensive logs during processing. The best way to capture them is to use the builtin setup_logging method:

from planai.utils import setup_logging

def main():
    setup_logging()

This will generate two different logs files. One general log that captures most of PlanAI’s operations and one llm log that shows the messages sent and received from LLMs.

Best Practices

1. Development: Always enable the dashboard during development
2. Testing: Use the dashboard to verify workflow behavior
3. Production: Consider security implications before exposing the dashboard
4. Debugging: Best done via the emitted logs in general although the dashboard shows failing tasks as well.
5. Performance: Monitor the Active, Queued and Failed tasks. Adjust parallelism for individual LLMTaskWorkers.