Files
gh-epieczko-betty/skills/telemetry.capture/INTEGRATION_EXAMPLES.md
2025-11-29 18:26:08 +08:00

449 lines
12 KiB
Markdown

# Telemetry Integration Examples
This document provides practical examples of integrating telemetry tracking into Betty Framework CLI entrypoints, workflows, and agents.
## Quick Start
The Betty Framework provides two main approaches for telemetry integration:
1. **Decorator Pattern** - For CLI entrypoints with standard main() functions
2. **Manual Capture** - For workflows, agents, and custom integrations
## 1. Decorator Pattern (Recommended for CLI)
### Standard CLI with Return Code
For CLI entrypoints that return an exit code:
```python
#!/usr/bin/env python3
import sys
import os
from typing import Optional, List
# Ensure project root on path
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../..")))
from betty.logging_utils import setup_logger
from betty.telemetry_integration import telemetry_tracked
logger = setup_logger(__name__)
@telemetry_tracked(skill_name="my.skill", caller="cli")
def main(argv: Optional[List[str]] = None) -> int:
"""Entry point for CLI execution."""
argv = argv or sys.argv[1:]
# Your skill logic here
if not argv:
logger.error("Missing required arguments")
return 1
# Process and return success
return 0
if __name__ == "__main__":
sys.exit(main(sys.argv[1:]))
```
**What the decorator does:**
- Automatically measures execution time
- Captures success/failure based on return code (0 = success, non-zero = failed)
- Logs telemetry with sanitized inputs
- Non-blocking - telemetry failures don't affect the main function
### Example: workflow.validate Integration
The `workflow.validate` skill has been updated with telemetry tracking:
```python
# skills/workflow.validate/workflow_validate.py
from betty.telemetry_integration import telemetry_tracked
@telemetry_tracked(skill_name="workflow.validate", caller="cli")
def main(argv: Optional[List[str]] = None) -> int:
"""Entry point for CLI execution."""
# ... existing validation logic ...
return 0 if response["ok"] else 1
```
**Test it:**
```bash
python3 skills/workflow.validate/workflow_validate.py example.yaml
cat registry/telemetry.json
```
## 2. Manual Capture Pattern
### Direct Function Call
For programmatic telemetry capture:
```python
from skills.telemetry.capture.telemetry_capture import create_telemetry_entry, capture_telemetry
import time
# Track execution manually
start_time = time.time()
try:
# Your logic here
result = execute_my_skill(param1, param2)
status = "success"
except Exception as e:
status = "failed"
raise
finally:
duration_ms = int((time.time() - start_time) * 1000)
entry = create_telemetry_entry(
skill_name="my.skill",
inputs={"param1": param1, "param2": param2},
status=status,
duration_ms=duration_ms,
caller="api"
)
capture_telemetry(entry)
```
### Workflow Execution Tracking
For capturing telemetry within workflow execution:
```python
# In workflow.compose or similar workflow executor
from betty.telemetry_integration import track_skill_execution
def execute_workflow_step(step_config: dict, workflow_name: str):
"""Execute a workflow step with telemetry tracking."""
skill_name = step_config["skill"]
skill_args = step_config["args"]
result = track_skill_execution(
skill_name=skill_name,
func=lambda: run_skill(skill_name, skill_args),
inputs={"args": skill_args},
workflow=workflow_name,
caller="workflow"
)
return result
```
### Agent Skill Invocation
For tracking skills invoked by agents:
```python
# In agent skill runner
from betty.telemetry_integration import track_skill_execution
def run_agent_skill(agent_name: str, skill_name: str, **kwargs):
"""Run a skill on behalf of an agent with telemetry."""
result = track_skill_execution(
skill_name=skill_name,
func=lambda: execute_skill(skill_name, **kwargs),
inputs=kwargs,
agent=agent_name,
caller="agent"
)
return result
```
## 3. CLI Helper Function
For simple CLI telemetry without decorators:
```python
from betty.telemetry_integration import capture_cli_telemetry
import time
def main():
start_time = time.time()
status = "failed"
try:
# Your CLI logic
process_command()
status = "success"
return 0
except Exception as e:
logger.error(f"Error: {e}")
return 1
finally:
duration_ms = int((time.time() - start_time) * 1000)
capture_cli_telemetry(
skill_name="my.skill",
inputs={"cli_args": sys.argv[1:]},
status=status,
duration_ms=duration_ms,
caller="cli"
)
```
## 4. Context-Rich Telemetry
### Full Context Example
Capture comprehensive context for deep analytics:
```python
entry = create_telemetry_entry(
skill_name="agent.define",
inputs={
"name": "my-agent",
"mode": "iterative",
"capabilities": ["code-gen", "testing"]
},
status="success",
duration_ms=2500,
agent="meta-orchestrator", # Which agent invoked this
workflow="agent-creation-pipeline", # Which workflow this is part of
caller="api", # How it was invoked
# Custom fields for advanced analytics
user_id="dev-123",
environment="staging",
version="1.0.0"
)
capture_telemetry(entry)
```
## 5. Batch Operations
For tracking multiple skill executions in batch:
```python
from skills.telemetry.capture.telemetry_capture import create_telemetry_entry, capture_telemetry
import time
def process_batch(items: list):
"""Process items in batch with per-item telemetry."""
for item in items:
start_time = time.time()
try:
process_item(item)
status = "success"
except Exception as e:
status = "failed"
duration_ms = int((time.time() - start_time) * 1000)
# Capture telemetry for each item
entry = create_telemetry_entry(
skill_name="batch.processor",
inputs={"item_id": item.id},
status=status,
duration_ms=duration_ms,
caller="batch"
)
capture_telemetry(entry)
```
## 6. Error Handling Best Practices
### Graceful Telemetry Failures
Always wrap telemetry in try-except to prevent failures from affecting main logic:
```python
def my_critical_function():
"""Critical function that must not fail due to telemetry issues."""
start_time = time.time()
result = None
try:
# Critical business logic
result = perform_critical_operation()
status = "success"
except Exception as e:
status = "failed"
raise # Re-raise the original error
finally:
# Capture telemetry (failures are logged but don't interrupt)
try:
duration_ms = int((time.time() - start_time) * 1000)
capture_cli_telemetry(
skill_name="critical.operation",
inputs={},
status=status,
duration_ms=duration_ms
)
except Exception as telemetry_error:
logger.warning(f"Telemetry capture failed: {telemetry_error}")
# Don't raise - telemetry is not critical
return result
```
## 7. Sensitive Data Protection
### Sanitize Inputs
Never capture sensitive data like passwords, tokens, or PII:
```python
def sanitize_inputs(inputs: dict) -> dict:
"""Remove sensitive fields from inputs before logging."""
sensitive_keys = ["password", "token", "api_key", "secret", "ssn", "credit_card"]
sanitized = {}
for key, value in inputs.items():
if any(sensitive in key.lower() for sensitive in sensitive_keys):
sanitized[key] = "***REDACTED***"
else:
sanitized[key] = value
return sanitized
# Use it
entry = create_telemetry_entry(
skill_name="auth.login",
inputs=sanitize_inputs({"username": "john", "password": "secret123"}),
status="success",
duration_ms=150
)
```
## 8. Testing with Telemetry
### Unit Test Example
```python
import unittest
from unittest.mock import patch, MagicMock
class TestMySkillWithTelemetry(unittest.TestCase):
@patch('betty.telemetry_integration.capture_cli_telemetry')
def test_skill_execution_captures_telemetry(self, mock_capture):
"""Test that telemetry is captured on skill execution."""
# Execute the skill
from skills.my.skill.my_skill import main
result = main(["arg1", "arg2"])
# Verify telemetry was captured
self.assertEqual(result, 0)
mock_capture.assert_called_once()
# Verify telemetry parameters
call_args = mock_capture.call_args
self.assertEqual(call_args.kwargs['skill_name'], "my.skill")
self.assertEqual(call_args.kwargs['status'], "success")
self.assertGreater(call_args.kwargs['duration_ms'], 0)
```
## 9. Migration Checklist
To add telemetry to an existing skill:
1. **Import the telemetry integration:**
```python
from betty.telemetry_integration import telemetry_tracked
```
2. **Apply the decorator to main():**
```python
@telemetry_tracked(skill_name="your.skill", caller="cli")
def main(argv: Optional[List[str]] = None) -> int:
```
3. **Ensure main() returns an int:**
- Return 0 for success
- Return non-zero for failure
4. **Test the integration:**
```bash
python3 skills/your.skill/your_skill.py test-args
cat registry/telemetry.json | jq '.[-1]' # View latest entry
```
## 10. Querying Telemetry Data
### Basic Queries
```bash
# Count total telemetry entries
jq 'length' registry/telemetry.json
# Find all failed executions
jq '.[] | select(.status == "failed")' registry/telemetry.json
# Get average duration for a skill
jq '[.[] | select(.skill == "workflow.validate") | .duration_ms] | add / length' registry/telemetry.json
# Top 10 slowest executions
jq 'sort_by(.duration_ms) | reverse | .[0:10]' registry/telemetry.json
# Executions by caller
jq 'group_by(.caller) | map({caller: .[0].caller, count: length})' registry/telemetry.json
```
### Advanced Analytics
```bash
# Skills executed in last hour
jq --arg cutoff "$(date -u -d '1 hour ago' +%Y-%m-%dT%H:%M:%S)" \
'.[] | select(.timestamp > $cutoff)' registry/telemetry.json
# Success rate by skill
jq 'group_by(.skill) | map({
skill: .[0].skill,
total: length,
successful: ([.[] | select(.status == "success")] | length),
success_rate: (([.[] | select(.status == "success")] | length) / length * 100)
})' registry/telemetry.json
```
## 11. Future: Prometheus Export
Example of how telemetry data could be exported to Prometheus (future implementation):
```python
# Future: skills/telemetry.capture/exporters/prometheus.py
from prometheus_client import Counter, Histogram, Gauge
skill_executions = Counter(
'betty_skill_executions_total',
'Total skill executions',
['skill', 'status', 'caller']
)
skill_duration = Histogram(
'betty_skill_duration_seconds',
'Skill execution duration',
['skill', 'caller']
)
def export_to_prometheus(telemetry_entry: dict):
"""Export telemetry entry to Prometheus metrics."""
skill_executions.labels(
skill=telemetry_entry['skill'],
status=telemetry_entry['status'],
caller=telemetry_entry.get('caller', 'unknown')
).inc()
skill_duration.labels(
skill=telemetry_entry['skill'],
caller=telemetry_entry.get('caller', 'unknown')
).observe(telemetry_entry['duration_ms'] / 1000.0)
```
## Summary
- **CLI Skills**: Use `@telemetry_tracked` decorator
- **Workflows**: Use `track_skill_execution()` helper
- **Custom Code**: Use `create_telemetry_entry()` + `capture_telemetry()`
- **Always**: Handle telemetry failures gracefully
- **Never**: Capture sensitive data in inputs
For more details, see the [SKILL.md](SKILL.md) documentation.