rebrandly-otel 0.3.4

Python OTEL wrapper by Rebrandly

Rebrandly OpenTelemetry SDK for Python

A comprehensive OpenTelemetry instrumentation SDK designed specifically for Rebrandly services, with built-in support for AWS Lambda functions and message processing.

Overview

The Rebrandly OpenTelemetry SDK provides a unified interface for distributed tracing, metrics collection, and structured logging across Python applications. It offers automatic instrumentation for AWS Lambda functions, simplified span management, and seamless integration with OTLP-compatible backends.

Table of Contents

• Installation
• Quick Start
• Getting Started
  • Step 1: Install the Package
  • Step 2: Configure Environment Variables
  • Step 3: Choose Your Integration Pattern
  • Step 4: Verify It's Working
  • Step 5: Add Custom Instrumentation
• Configuration
• Core Components
• Built-in Metrics
• Tracing Features
• Automatic Span Attributes
• Logging Integration
• AWS Lambda Support
• Performance Considerations
• Export Formats
• Thread Safety
• Resource Attributes
• Error Handling
• Compatibility
• Best Practices
  • Span Management
  • Error Handling
  • Metric Cardinality
  • Lambda Functions
  • Context Propagation
  • Logging
  • Configuration
  • Performance
  • Security
  • Testing
  • Async/Await Support
  • Database Instrumentation
• Examples
  • Lambda - Send SNS / SQS
  • Lambda - Receive SQS
  • Lambda - Receive SNS
  • Flask
  • FastAPI
  • PyMySQL Database Instrumentation
• Troubleshooting
• Testing
• License
• Build and Deploy

Installation

pip install rebrandly-otel

Note: The SDK automatically initializes when you import it. No manual setup or initialization calls are needed - just from rebrandly_otel import otel and start using it!

Dependencies

• opentelemetry-api
• opentelemetry-sdk
• opentelemetry-exporter-otlp-proto-grpc
• opentelemetry-semantic-conventions
• psutil (for system metrics)

Quick Start

Get started with the Rebrandly OpenTelemetry SDK in under 5 minutes:

from rebrandly_otel import otel, logger
from opentelemetry.trace import SpanKind, Status, StatusCode

# 1. SDK auto-initializes when imported
# Optional: Configure via environment variables (see Configuration section)

# 2. Create a traced operation using context manager
def process_order(order_id):
    with otel.span("process-order", attributes={"order.id": order_id}) as span:
        logger.info(f"Processing order {order_id}")

        # Your business logic here
        save_to_database(order_id)

        # Exceptions are automatically recorded
        # No need to manually call span.end()

# 3. For AWS Lambda functions
from rebrandly_otel import lambda_handler

@lambda_handler(name="order-processor")
def handler(event, context):
    logger.info("Lambda invoked", extra={"event": event})
    order_id = event.get('orderId')
    process_order(order_id)
    return {'statusCode': 200, 'body': 'Success'}

# 4. For Flask applications
from flask import Flask
from rebrandly_otel import app_before_request, app_after_request, flask_error_handler

app = Flask(__name__)
app.before_request(app_before_request)
app.after_request(app_after_request)
app.register_error_handler(Exception, flask_error_handler)

@app.route('/orders/<order_id>')
def get_order(order_id):
    with otel.span("fetch-order"):
        logger.info(f"Fetching order {order_id}")
        # Your logic here
        return {"order_id": order_id, "status": "shipped"}

Next Steps:

• See Getting Started for detailed integration guide
• Check Configuration to set up environment variables
• Explore Examples for framework-specific patterns (Flask, FastAPI, Lambda)

Getting Started

Step 1: Install the Package

pip install rebrandly-otel

Step 2: Configure Environment Variables

Create a .env file or set environment variables:

# Required
export OTEL_SERVICE_NAME=my-service
export OTEL_SERVICE_VERSION=1.0.0

# Optional - for sending data to an OTLP collector
export OTEL_EXPORTER_OTLP_ENDPOINT=https://your-collector:4317

# Optional - for debugging locally
export OTEL_DEBUG=true

Step 3: Choose Your Integration Pattern

For Flask Applications

from flask import Flask
from rebrandly_otel import otel, logger, app_before_request, app_after_request, flask_error_handler

app = Flask(__name__)

# Register OTEL handlers - handles ALL telemetry automatically
app.before_request(app_before_request)
app.after_request(app_after_request)
app.register_error_handler(Exception, flask_error_handler)

# Your routes - no telemetry code needed!
@app.route('/api/users')
def get_users():
    logger.info("Fetching users")
    # Business logic only
    return {"users": []}

@app.route('/api/users/<user_id>')
def get_user(user_id):
    # Add custom spans when needed
    with otel.span("fetch-user-details"):
        logger.info(f"Fetching user {user_id}")
        # Your business logic
        return {"user_id": user_id, "name": "John Doe"}

if __name__ == '__main__':
    app.run(debug=True)

For FastAPI Applications

from fastapi import FastAPI, Depends
from contextlib import asynccontextmanager
from rebrandly_otel import otel, logger, force_flush
from rebrandly_otel.fastapi_support import setup_fastapi, get_current_span

@asynccontextmanager
async def lifespan(app: FastAPI):
    logger.info("Application starting up")
    yield
    logger.info("Application shutting down")
    force_flush()

app = FastAPI(lifespan=lifespan)

# Setup OTEL integration - handles ALL telemetry automatically
setup_fastapi(otel, app)

# Your routes
@app.get("/api/users")
async def get_users():
    logger.info("Fetching users")
    return {"users": []}

@app.get("/api/users/{user_id}")
async def get_user(user_id: int, span=Depends(get_current_span)):
    # Add custom spans when needed
    with otel.span("fetch-user-details", attributes={"user.id": user_id}):
        logger.info(f"Fetching user {user_id}")
        # Your business logic
        return {"user_id": user_id, "name": "John Doe"}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)

For AWS Lambda Functions

from rebrandly_otel import lambda_handler, logger, otel

@lambda_handler(name="user-processor")
def handler(event, context):
    logger.info("Processing event", extra={"event_type": event.get("eventType")})

    # Your business logic
    user_id = event.get('userId')

    # Add custom spans if needed
    with otel.span("process-user", attributes={"user.id": user_id}):
        result = process_user(user_id)

    return {
        'statusCode': 200,
        'body': result
    }

def process_user(user_id):
    logger.info(f"Processing user {user_id}")
    # Your business logic here
    return {"processed": True}

For Standalone Scripts

from rebrandly_otel import otel, logger, force_flush, shutdown

def main():
    # Create traced operation
    with otel.span("main-operation"):
        logger.info("Starting operation")

        # Your business logic
        process_data()

        logger.info("Operation completed")

def process_data():
    with otel.span("process-data"):
        # Nested spans are automatically linked
        logger.info("Processing data")
        # Your logic here

if __name__ == "__main__":
    try:
        main()
    except Exception as e:
        logger.error(f"Operation failed: {e}", exc_info=True)
    finally:
        # Ensure telemetry is flushed before exit
        force_flush(timeout_millis=5000)
        shutdown()

Step 4: Verify It's Working

Local Debugging

Set OTEL_DEBUG=true to see telemetry output in your console:

OTEL_DEBUG=true python app.py

You should see trace and metric data logged to the console.

Production Setup

Configure OTEL_EXPORTER_OTLP_ENDPOINT to point to your OpenTelemetry Collector or backend:

export OTEL_EXPORTER_OTLP_ENDPOINT=https://your-collector.example.com:4317
export OTEL_EXPORTER_OTLP_HEADERS="x-api-key=your-api-key"

Common backends:

• Honeycomb: https://api.honeycomb.io:443
• Lightstep: https://ingest.lightstep.com:443
• Jaeger: http://jaeger-collector:4317
• Self-hosted Collector: http://localhost:4317

Step 5: Add Custom Instrumentation (Optional)

Add custom spans, metrics, and logs as needed:

from rebrandly_otel import otel, logger, meter
from opentelemetry.trace import Status, StatusCode

# Custom span with attributes
with otel.span("custom-operation", attributes={"user.id": user_id}) as span:
    # Add events to the span
    span.add_event("processing_started", {"timestamp": datetime.now().isoformat()})

    # Your code
    result = do_work()

    # Add more attributes dynamically
    span.set_attribute("result.count", len(result))

# Custom metric
order_counter = meter.meter.create_counter(
    name="orders.created",
    description="Number of orders created",
    unit="1"
)
order_counter.add(1, {"order.type": "standard", "region": "us-east-1"})

# Custom histogram for measuring durations
duration_histogram = meter.meter.create_histogram(
    name="order.processing.duration",
    description="Order processing duration",
    unit="ms"
)
duration_histogram.record(123.45, {"order.type": "standard"})

# Structured logging with trace correlation
logger.info("Order processed", extra={
    "order_id": 12345,
    "user_id": 67890,
    "amount": 99.99
})

Configuration

The SDK is configured through environment variables:

Variable	Description	Default
OTEL_SERVICE_NAME	Service identifier	default-service-python
OTEL_SERVICE_VERSION	Service version	1.0.0
OTEL_SERVICE_APPLICATION	Application namespace (groups multiple services under one application)	Fallback to OTEL_SERVICE_NAME
OTEL_EXPORTER_OTLP_ENDPOINT	OTLP collector endpoint	None
OTEL_DEBUG	Enable console debugging	false
OTEL_CAPTURE_REQUEST_BODY	Enable HTTP request body capture for Flask and FastAPI (default: true). Set to false to disable. Only captures JSON content with automatic sensitive data redaction.	true
OTEL_SPAN_ATTRIBUTES	Attributes automatically added to all spans (format: key1=value1,key2=value2)	None
BATCH_EXPORT_TIME_MILLIS	Batch export interval	100
ENV or ENVIRONMENT or NODE_ENV	Deployment environment	local

Core Components

RebrandlyOTEL Class

The main entry point for all telemetry operations. Implements a singleton pattern to ensure consistent instrumentation across your application.

Properties

• tracer: Returns the RebrandlyTracer instance for distributed tracing
• meter: Returns the RebrandlyMeter instance for metrics collection
• logger: Returns the configured Python logger with OpenTelemetry integration

Initialization

The SDK auto-initializes as soon as you embed it.

Key Methods

span(name, attributes=None, kind=SpanKind.INTERNAL, message=None)

Context manager for creating traced spans with automatic error handling and status management.

lambda_handler(name=None, attributes=None, kind=SpanKind.CONSUMER, auto_flush=True, skip_aws_link=True)

Decorator for AWS Lambda functions with automatic instrumentation, metrics collection, and telemetry flushing.

aws_message_handler(name=None, attributes=None, kind=SpanKind.CONSUMER, auto_flush=True)

Decorator for processing individual AWS messages (SQS/SNS) with context propagation.

aws_message_span(name, message=None, attributes=None, kind=SpanKind.CONSUMER)

Context manager for creating spans from AWS messages with automatic context extraction.

force_flush(start_datetime=None, timeout_millis=1000)

Forces all pending telemetry data to be exported. Critical for serverless environments.

shutdown()

Gracefully shuts down all OpenTelemetry components.

Built-in Metrics

The SDK automatically registers and tracks the following metrics:

Standard Metrics

• cpu_usage_percentage (Gauge): CPU utilization percentage
• memory_usage_bytes (Gauge): Memory usage in bytes

Custom Metrics

You can create the custom metrics you need using the default open telemetry metrics

from src.rebrandly_otel import meter

sqs_counter = meter.meter.create_counter(
    name="sqs_sender_counter",
    description="Number of messages sent",
    unit="1"
)
sqs_counter.add(1)

Tracing Features

Automatic Context Propagation

The SDK automatically extracts and propagates trace context from:

• AWS SQS message attributes
• AWS SNS message attributes
• HTTP headers
• Custom carriers

Span Attributes

Lambda spans automatically include:

• faas.trigger: Detected trigger type (sqs, sns, api_gateway, etc.)
• faas.execution: AWS request ID
• faas.id: Function ARN
• cloud.provider: Always "aws" for Lambda
• cloud.platform: Always "aws_lambda" for Lambda

Automatic Span Attributes

The SDK supports automatically adding custom attributes to all spans via the OTEL_SPAN_ATTRIBUTES environment variable. This is useful for adding metadata that applies to all telemetry in a service, such as team ownership, deployment environment, or version information.

Configuration

Set the OTEL_SPAN_ATTRIBUTES environment variable with a comma-separated list of key-value pairs:

export OTEL_SPAN_ATTRIBUTES="team=backend,environment=production,version=1.2.3"

Behavior

• Universal Application: Attributes are added to ALL spans, including:

  • Manually created spans (tracer.start_span(), tracer.start_as_current_span())
  • Lambda handler spans (@lambda_handler)
  • AWS message handler spans (@aws_message_handler)
  • Flask/FastAPI middleware spans
  • Auto-instrumented spans (database queries, HTTP requests, etc.)

• Format: Same as OTEL_RESOURCE_ATTRIBUTES - comma-separated key=value pairs

• Value Handling: Supports values containing = characters (e.g., URLs)

• Whitespace: Leading/trailing whitespace is automatically trimmed

Example

import os

# Set environment variable
os.environ['OTEL_SPAN_ATTRIBUTES'] = "team=backend,service.owner=platform-team,deployment.region=us-east-1"

# Initialize SDK
from rebrandly_otel import otel, logger

# Create any span - attributes are added automatically
with otel.span('my-operation'):
    logger.info('Processing request')
    # The span will include:
    # - team: "backend"
    # - service.owner: "platform-team"
    # - deployment.region: "us-east-1"
    # ... plus any other attributes you set manually

Use Cases

• Team/Ownership Tagging: team=backend,owner=[e-mail]
• Environment Metadata: environment=production,region=us-east-1,availability_zone=us-east-1a
• Version Tracking: version=1.2.3,build=12345,commit=abc123def
• Cost Attribution: cost_center=engineering,project=customer-api
• Multi-Tenancy: tenant=acme-corp,customer_tier=enterprise

Difference from OTEL_RESOURCE_ATTRIBUTES

• OTEL_RESOURCE_ATTRIBUTES: Service-level metadata (set once, applies to the entire service instance)
• OTEL_SPAN_ATTRIBUTES: Span-level metadata (added to each individual span at creation time)

Both use the same format but serve different purposes in the OpenTelemetry data model.

Span Filtering for Cost Optimization

For high-volume services that generate massive amounts of telemetry data, the SDK supports span filtering to reduce ingestion costs while maintaining visibility into critical errors. This feature works with the OpenTelemetry Collector's tail sampling policies.

Use Case: Errors-Only Filtering

High-traffic services (APIs, event processors, background workers) often generate thousands of successful spans that provide little debugging value. By enabling errors-only filtering, you can:

• Reduce ingestion costs by 90-99% (only error spans are stored)
• Maintain error visibility - all errors are still captured and traced
• Preserve spanmetrics - metrics are generated at the agent level before filtering

Configuration

Set the span.filter=errors-only attribute using OTEL_SPAN_ATTRIBUTES:

export OTEL_SPAN_ATTRIBUTES="span.filter=errors-only"

In serverless.yml (for Lambda functions):

provider:
  environment:
    OTEL_SERVICE_NAME: high-volume-api
    OTEL_EXPORTER_OTLP_ENDPOINT: http://otel-collector:4317
    OTEL_SPAN_ATTRIBUTES: "span.filter=errors-only"

How It Works

1. Application Level: The OTEL_SPAN_ATTRIBUTES environment variable adds span.filter=errors-only to all spans created by your service

2. Agent Level: The OTEL Agent receives all traces and generates spanmetrics (request counts, latencies, etc.) from 100% of traces

3. Gateway Level - Two-Stage Filtering:

   Stage 1 - Filter Processor (runs first):

   • Drops spans with span.filter=errors-only AND status is NOT ERROR
   • Passes all other spans to the next stage

   Stage 2 - Tail Sampling (runs second):

   • Applies policy-based sampling to remaining spans
   • Errors-only-filter policy: Samples error spans from errors-only services
   • General errors-policy: Samples all other error spans
   • Randomized-policy: Samples 30% of remaining successful spans (from services WITHOUT errors-only filter)

4. Backend: Only error traces from errors-only services are stored. Services without the filter get 30% of successful spans + all errors sampled.

Example

import os
from rebrandly_otel import otel, logger

# Enable errors-only filtering
os.environ['OTEL_SPAN_ATTRIBUTES'] = "span.filter=errors-only"

from rebrandly_otel import lambda_handler

@lambda_handler(name="high-volume-processor")
def handler(event, context):
    logger.info("Processing event")

    # Successful execution - span will be dropped at gateway
    if process_successfully(event):
        return {'statusCode': 200, 'body': 'Success'}

    # Error case - span will be sampled and stored
    raise Exception("Processing failed")  # This trace will be captured!

When to Use Errors-Only Filtering

✅ Good candidates for errors-only filtering:

• High-traffic APIs (> 1000 requests/second)
• Background job processors with high success rates
• Event streaming consumers (SQS, Kinesis, Kafka)
• Scheduled tasks that run frequently
• Health check endpoints

❌ Not recommended for:

• Low-traffic services (< 100 requests/minute)
• Services in active development or debugging
• Critical payment/transaction processing (you may want to sample successful traces too)
• Services where latency analysis of successful requests is important

Combining with Other Filters

You can combine span.filter with other span attributes:

export OTEL_SPAN_ATTRIBUTES="span.filter=errors-only,team=backend,environment=production"

All attributes are added to every span, enabling rich filtering and querying capabilities.

Monitoring and Validation

To verify errors-only filtering is working:

1. Check spanmetrics in your metrics backend - should show 100% of requests
2. Check traces in your tracing backend - should only show error traces
3. Calculate sampling rate: trace_count / metric_request_count should be very low (< 1%)

# Add custom metric to track sampling
from rebrandly_otel import meter

sampling_counter = meter.meter.create_counter(
    name="spans.filtered",
    description="Spans filtered by errors-only policy",
    unit="1"
)

# Increment on successful operations (these will be dropped)
sampling_counter.add(1, {"filter": "errors-only", "outcome": "success"})

Cost Savings Example

Before (all traces sampled):

• Service: 10,000 requests/minute
• Error rate: 0.5%
• Traces stored: 14.4 million/day
• Cost: ~$500/month (at $0.10 per 100K spans)

After (errors-only filtering):

• Service: 10,000 requests/minute
• Error rate: 0.5%
• Traces stored: 72,000/day (only errors)
• Cost: ~$2.50/month (at $0.10 per 100K spans)

Savings: 99.5% reduction, ~$497/month saved 💰

Troubleshooting

Issue: No traces appearing in backend after enabling errors-only filtering

Solution: Verify errors are actually occurring. Test by intentionally raising an exception:

# Temporary test endpoint
@app.route('/test-error')
def test_error():
    raise Exception("Test error for trace validation")

Issue: Successful traces still appearing in backend

Solution:

1. Verify OTEL_SPAN_ATTRIBUTES is set correctly
2. Check OTEL Collector gateway configuration has the errors-only policy
3. Ensure collector version supports tail sampling and policies

Exception Handling

Spans automatically capture exceptions with:

• Full exception details and stack traces
• Automatic status code setting
• Exception events in the span timeline

Logging Integration

The SDK integrates with Python's standard logging module:

from rebrandly_otel import logger

# Use as a standard Python logger
logger.info("Processing started", extra={"request_id": "123"})
logger.error("Processing failed", exc_info=True)

Features:

• Automatic trace context injection
• Structured logging support
• Console and OTLP export
• Log level configuration via environment

AWS Lambda Support

Trigger Detection

Automatically detects and labels Lambda triggers:

• API Gateway (v1 and v2)
• SQS
• SNS
• S3
• Kinesis
• DynamoDB
• EventBridge
• Batch

Automatic Metrics

For Lambda functions, the SDK automatically captures:

• Memory usage
• CPU utilization

Context Extraction

Automatically extracts trace context from:

• SQS MessageAttributes
• SNS MessageAttributes (including nested format)
• Custom message attributes

Performance Considerations

Batch Processing

The SDK uses batch processing to optimize network usage and reduce overhead:

import os

# Configure batch export interval (milliseconds)
os.environ['BATCH_EXPORT_TIME_MILLIS'] = '100'  # Default: 100ms

# Faster flushing for Lambda (reduce cold start impact)
os.environ['BATCH_EXPORT_TIME_MILLIS'] = '50'   # Flush every 50ms

# Slower flushing for high-throughput apps (better batching)
os.environ['BATCH_EXPORT_TIME_MILLIS'] = '200'  # Flush every 200ms

Trade-offs:

• Lower values (50ms): Faster data delivery, higher network overhead, better for serverless
• Higher values (200ms): Better batching, lower overhead, risk of data loss on crashes

Lambda Optimization

For AWS Lambda functions, the SDK is specifically optimized:

Cold Start Impact: < 50ms

• Lazy initialization of exporters
• Minimal import overhead
• Efficient resource allocation

Memory Usage: ~20-30 MB additional

• Efficient span buffering
• Automatic cleanup on function freeze
• No memory leaks in long-running containers

Best Practices for Lambda:

from rebrandly_otel import lambda_handler, force_flush

@lambda_handler(name="my-function", auto_flush=True)
def handler(event, context):
    # auto_flush=True ensures telemetry is exported
    # Add 2-3 seconds to timeout for flush buffer
    return process_event(event)

# For manual control
@lambda_handler(name="my-function", auto_flush=False)
def handler(event, context):
    result = process_event(event)
    force_flush(timeout_millis=2000)  # Explicit flush with 2s timeout
    return result

Sampling Strategies

For high-traffic applications, implement sampling to reduce overhead:

from opentelemetry.sdk.trace.sampling import ParentBasedTraceIdRatioBased, ALWAYS_ON

# Sample 10% of traces
sampler = ParentBasedTraceIdRatioBased(0.1)

# Use ALWAYS_ON for low-traffic or critical services
sampler = ALWAYS_ON

Metric Cardinality Management

Avoid high-cardinality attributes that create too many metric series:

from rebrandly_otel import meter

# ❌ Bad: Creates millions of unique metric series
order_counter = meter.meter.create_counter("orders.processed")
order_counter.add(1, {"user_id": "12345", "order_id": "98765"})  # Too many combinations!

# ✅ Good: Limited cardinality
order_counter = meter.meter.create_counter("orders.processed")
order_counter.add(1, {
    "order.type": "standard",  # Only a few types
    "region": "us-east-1",     # Limited regions
    "tier": "premium"           # Few tier values
})

Cardinality Guidelines:

• Keep attribute combinations under 1000 per metric
• Use aggregations in your application layer for high-cardinality data
• Monitor metric series count in your backend

Span Attributes Best Practices

Optimize span attributes for performance and cost:

from rebrandly_otel import otel

# ✅ Good: Reasonable attribute size
with otel.span("process-order", attributes={
    "order.id": "12345",
    "user.id": "67890",
    "order.total": 99.99
}) as span:
    process_order()

# ❌ Bad: Large payloads in attributes
with otel.span("process-order", attributes={
    "order.full_details": json.dumps(order),  # Could be huge!
    "request.body": request_body               # Potentially large
}) as span:
    process_order()

Guidelines:

• Keep individual attributes under 1KB
• Avoid storing full payloads in attributes
• Use events for detailed debugging data
• Leverage OTEL_CAPTURE_REQUEST_BODY=false to disable body capture

Database Query Optimization

Minimize overhead from database instrumentation:

from rebrandly_otel import instrument_pymysql

# Configure slow query threshold to reduce span volume
connection = instrument_pymysql(otel, connection, options={
    'slow_query_threshold_ms': 1000,   # Only flag queries > 1s
    'capture_bindings': False          # Disable parameter capture (faster)
})

Thread Pool and Async Considerations

The SDK is thread-safe and works with async code:

import asyncio
from rebrandly_otel import otel, logger

async def async_operation():
    # Context is automatically propagated in async functions
    with otel.span("async-work"):
        await asyncio.sleep(0.1)
        logger.info("Async work completed")

# Multiple concurrent operations
async def main():
    await asyncio.gather(
        async_operation(),
        async_operation(),
        async_operation()
    )

Memory Management

Monitor and optimize memory usage:

import psutil
from rebrandly_otel import otel, meter, logger

# Create memory gauge for monitoring
memory_gauge = meter.meter.create_observable_gauge(
    name="process.memory.used",
    callbacks=[lambda: psutil.Process().memory_info().rss],
    description="Process memory usage",
    unit="bytes"
)

# Check span buffer size
def check_telemetry_overhead():
    process = psutil.Process()
    mem_before = process.memory_info().rss

    # Create 1000 spans
    for i in range(1000):
        with otel.span(f"test-{i}"):
            pass

    mem_after = process.memory_info().rss
    overhead = (mem_after - mem_before) / 1000
    logger.info(f"Per-span memory overhead: {overhead} bytes")

Production Optimization Checklist

• Configure appropriate BATCH_EXPORT_TIME_MILLIS for your workload
• Implement sampling for high-traffic services (> 100 req/s)
• Keep metric cardinality under 1000 combinations per metric
• Monitor memory usage and adjust batch settings if needed
• Use OTEL_DEBUG=false in production (significant performance impact)
• Set appropriate timeout buffers for Lambda functions (add 2-3s)
• Review and limit span attribute sizes (< 1KB per attribute)
• Disable request body capture if not needed (OTEL_CAPTURE_REQUEST_BODY=false)
• Use connection pooling for database instrumentation
• Monitor OTLP exporter queue depth and adjust batch settings

Export Formats

Supported Exporters

• OTLP/gRPC: Primary export format for production
• Console: Available for local development and debugging

Thread Safety

All components are thread-safe and can be used in multi-threaded applications:

• Singleton pattern ensures single initialization
• Thread-safe metric recording
• Concurrent span creation support

Resource Attributes

Automatically includes:

• Service name and version
• Python runtime version
• Deployment environment
• Custom resource attributes via environment

Error Handling

• Graceful degradation when OTLP endpoint unavailable
• Non-blocking telemetry operations
• Automatic retry with exponential backoff
• Comprehensive error logging

Compatibility

• Python 3.7+
• AWS Lambda runtime support
• Compatible with OpenTelemetry Collector
• Works with any OTLP-compatible backend

Best Practices

Span Management

1. Use Context Managers for Automatic Cleanup

from rebrandly_otel import otel, logger

# ✅ Good: Context manager automatically ends span
with otel.span("process-order", attributes={"order.id": order_id}):
    process_order(order_id)
    # Span automatically ended, even if exception occurs

# ⚠️ Manual span management (less preferred)
span = otel.tracer.start_span("process-order")
try:
    process_order(order_id)
finally:
    span.end()  # Must remember to end

2. Use Meaningful Span Names

# ✅ Good: Descriptive, operation-focused names
with otel.span("fetch-user-profile"):
    pass
with otel.span("validate-payment"):
    pass
with otel.span("send-email-notification"):
    pass

# ❌ Bad: Vague or implementation-focused names
with otel.span("function1"):
    pass
with otel.span("handler"):
    pass
with otel.span("process"):
    pass

3. Add Contextual Attributes

with otel.span("create-order", attributes={
    # Business context
    "order.id": order_id,
    "user.id": user_id,
    "order.total": total_amount,
    "payment.method": payment_method,
    # Technical context
    "db.system": "postgresql",
    "http.method": "POST"
}) as span:
    # Can add more attributes dynamically
    span.set_attribute("order.items_count", len(items))

4. Record Exceptions Properly

from opentelemetry.trace import Status, StatusCode

with otel.span("risky-operation") as span:
    try:
        risky_operation()
    except Exception as e:
        span.record_exception(e)
        span.set_status(Status(StatusCode.ERROR, str(e)))
        raise  # Re-raise after recording

Error Handling

1. Distinguish Error Types

from rebrandly_otel import otel, logger
from opentelemetry.trace import Status, StatusCode

with otel.span("process-payment") as span:
    try:
        process_payment(amount)
        span.set_status(Status(StatusCode.OK))
    except ValidationError as e:
        # Client errors (4xx) - not span errors
        span.set_attribute("error.validation", str(e))
        span.set_status(Status(StatusCode.OK))  # Business logic, not system error
        raise
    except Exception as e:
        # Server errors (5xx) - mark span as error
        span.record_exception(e)
        span.set_status(Status(StatusCode.ERROR, str(e)))
        logger.error(f"Payment processing failed: {e}")
        raise

Metric Cardinality

1. Limit Attribute Values

from rebrandly_otel import meter

# ❌ Bad: Unbounded cardinality
request_counter.add(1, {
    "user.id": user_id,         # Millions of users!
    "request.id": request_id,   # Every request unique!
    "timestamp": time.time()    # Always unique!
})

# ✅ Good: Bounded cardinality
request_counter.add(1, {
    "http.method": "GET",       # ~10 values
    "http.route": "/api/users", # Hundreds of routes
    "http.response.status_code": 200     # ~50 status codes
})

2. Aggregate High-Cardinality Data

# Store detailed data in spans, aggregate in metrics
with otel.span("process-order", attributes={
    "user.id": user_id,    # Detailed (high cardinality OK in spans)
    "order.id": order_id
}) as span:
    # Aggregated attributes in metrics (low cardinality)
    order_counter.add(1, {
        "user.tier": get_user_tier(user_id),    # bronze/silver/gold
        "order.category": get_category(order)    # electronics/clothing/food
    })

Lambda Functions

1. Always Flush Before Exit

from rebrandly_otel import lambda_handler, force_flush

# Using decorator (auto-flush enabled by default)
@lambda_handler(name="my-function")
def handler(event, context):
    # Your code
    return response

# Manual flush if needed
@lambda_handler(name="my-function", auto_flush=False)
def handler(event, context):
    result = process(event)
    force_flush(timeout_millis=5000)  # Flush with 5s timeout
    return result

2. Add Buffer to Timeout

# If Lambda timeout is 30s, set function timeout to 27s
# Reserve 3s for telemetry flush

import time

LAMBDA_TIMEOUT_SEC = 30
FLUSH_BUFFER_SEC = 3
FUNCTION_TIMEOUT_SEC = LAMBDA_TIMEOUT_SEC - FLUSH_BUFFER_SEC

@lambda_handler(name="my-function")
def handler(event, context):
    deadline = time.time() + FUNCTION_TIMEOUT_SEC

    # Check timeout during processing
    if time.time() > deadline:
        raise TimeoutError("Function timeout approaching")

    return process_with_timeout(event, deadline)

Context Propagation

1. Propagate Context in HTTP Calls

import requests
from opentelemetry.propagate import inject

def call_downstream(url, data):
    # Extract current context and inject into headers
    headers = {}
    inject(headers)  # Automatically adds traceparent header

    # Make request with trace headers
    response = requests.post(url, json=data, headers=headers)
    return response.json()

2. Propagate Context in AWS Messages

import boto3
import json
from rebrandly_otel import otel

sqs = boto3.client('sqs')

# Get trace context for message attributes
trace_attrs = otel.tracer.get_attributes_for_aws_from_context()

# Send message with trace context
sqs.send_message(
    QueueUrl=queue_url,
    MessageBody=json.dumps(data),
    MessageAttributes=trace_attrs  # Automatic context injection
)

Logging

1. Use Structured Logging

from rebrandly_otel import logger

# ✅ Good: Structured with context
logger.info("Order processed", extra={
    "order_id": order.id,
    "user_id": user.id,
    "amount": order.total,
    "duration": processing_time
})

# ❌ Bad: String formatting
logger.info(f"Order {order.id} processed for user {user.id} with amount {order.total}")

2. Log at Appropriate Levels

logger.debug("Entering function", extra={"function": "process_order"})  # Development only
logger.info("Order created", extra={"order_id": 123})                   # Normal operations
logger.warning("Slow query detected", extra={"duration": 2.0})          # Performance issues
logger.error("Payment failed", extra={"error": str(e)}, exc_info=True) # Errors with traceback

Configuration

1. Use Environment-Specific Settings

# .env.development
export OTEL_DEBUG=true
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
export BATCH_EXPORT_TIME_MILLIS=50
export LOG_LEVEL=DEBUG

# .env.production
export OTEL_DEBUG=false
export OTEL_EXPORTER_OTLP_ENDPOINT=https://collector.prod.example.com:4317
export BATCH_EXPORT_TIME_MILLIS=100
export LOG_LEVEL=INFO
export OTEL_SPAN_ATTRIBUTES=environment=production,team=backend

2. Don't Hardcode Service Names

# ❌ Bad: Hardcoded in code
os.environ['OTEL_SERVICE_NAME'] = 'my-service'

# ✅ Good: Set via environment
# In Dockerfile or deployment config:
# ENV OTEL_SERVICE_NAME=my-service
# ENV OTEL_SERVICE_VERSION=1.2.3

Performance

1. Implement Sampling for High-Traffic Services

from opentelemetry.sdk.trace.sampling import ParentBasedTraceIdRatioBased, ALWAYS_ON

# Sample 10% of traces for high-traffic services (> 1000 req/s)
sampler = ParentBasedTraceIdRatioBased(0.1)

# Use ALWAYS_ON for low-traffic or critical services
sampler = ALWAYS_ON

2. Disable Debug Mode in Production

# Significant performance impact!
export OTEL_DEBUG=false  # In production

# Only enable for troubleshooting specific issues

3. Monitor Telemetry Overhead

from rebrandly_otel import meter
import time

# Track telemetry overhead
telemetry_duration = meter.meter.create_histogram(
    name="telemetry.overhead",
    description="Time spent on telemetry operations",
    unit="ms"
)

start = time.time()
# Your telemetry operation
overhead = (time.time() - start) * 1000
telemetry_duration.record(overhead)

Security

1. Sanitize Sensitive Data

from rebrandly_otel import logger

# ❌ Bad: Logging sensitive data
logger.info("User login", extra={"username": username, "password": password})

# ✅ Good: Exclude sensitive data
logger.info("User login", extra={
    "username": username,
    "password_provided": bool(password)
})

# SDK automatically redacts sensitive fields when OTEL_CAPTURE_REQUEST_BODY=true

2. Use Secure Connections

# ✅ Good: TLS endpoint
export OTEL_EXPORTER_OTLP_ENDPOINT=https://collector.example.com:4317

# ⚠️ Caution: Only use HTTP for local development
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317

Testing

1. Disable Telemetry in Tests

# conftest.py or test setup
import os
import pytest

@pytest.fixture(autouse=True)
def disable_telemetry():
    os.environ['OTEL_EXPORTER_OTLP_ENDPOINT'] = ''
    os.environ['OTEL_DEBUG'] = 'false'

# Or mock the SDK
from unittest.mock import Mock, patch

@patch('rebrandly_otel.otel')
@patch('rebrandly_otel.logger')
def test_my_function(mock_logger, mock_otel):
    mock_otel.span.return_value.__enter__ = Mock()
    mock_otel.span.return_value.__exit__ = Mock()
    # Your test

2. Test Telemetry Integration Separately

# test_telemetry_integration.py
from rebrandly_otel import otel
import os

def test_span_creation():
    os.environ['OTEL_DEBUG'] = 'true'
    with otel.span("test-span"):
        pass
    # Verify span was created

Async/Await Support

1. Use Spans with Async Functions

import asyncio
from rebrandly_otel import otel, logger

async def async_operation():
    # Context is automatically propagated in async functions
    with otel.span("async-work"):
        await asyncio.sleep(0.1)
        logger.info("Async work completed")

# Multiple concurrent operations
async def main():
    await asyncio.gather(
        async_operation(),
        async_operation(),
        async_operation()
    )

asyncio.run(main())

Database Instrumentation

1. Always Instrument Connections

import pymysql
from rebrandly_otel import otel, instrument_pymysql

# Create connection
connection = pymysql.connect(
    host='localhost',
    user='user',
    password='password',
    database='mydb'
)

# Instrument the connection
connection = instrument_pymysql(otel, connection, options={
    'slow_query_threshold_ms': 1000,  # Flag slow queries
    'capture_bindings': False         # Disable for performance/security
})

# All queries now automatically traced
with connection.cursor() as cursor:
    cursor.execute("SELECT * FROM users WHERE id = %s", (user_id,))

Examples

Lambda - Send SNS / SQS message

import os
import json
import boto3
from rebrandly_otel import otel, lambda_handler, logger

sqs = boto3.client('sqs')
QUEUE_URL = os.environ.get('SQS_URL')

@lambda_handler("sqs_sender")
def handler(event, context):
    logger.info("Starting SQS message send")

    # Get trace context for propagation
    trace_attrs = otel.tracer.get_attributes_for_aws_from_context()

    # Send message with trace context
    response = sqs.send_message(
        QueueUrl=QUEUE_URL,
        MessageBody=json.dumps({"data": "test message"}),
        MessageAttributes=trace_attrs
    )

    logger.info(f"Sent SQS message: {response['MessageId']}")

    return {
        'statusCode': 200,
        'body': json.dumps({'messageId': response['MessageId']})
    }

Lambda Receive SQS message

import json
from rebrandly_otel import lambda_handler, logger, aws_message_span

@lambda_handler(name="sqs_receiver")
def handler(event, context):
    for record in event['Records']:
        # Process each message with trace context
        process_message(record)

def process_message(record):
    with aws_message_span("process_message_sqs_receiver", message=record) as s:
        logger.info(f"Processing message: {record['messageId']}")

        # Parse message body
        body = json.loads(record['body'])
        logger.info(f"Message data: {body}")

Lambda Receive SNS message (record specific event)

import json
from rebrandly_otel import lambda_handler, logger, aws_message_span

@lambda_handler(name="sns_receiver")
def handler(event, context):
    for record in event['Records']:
        # Process each message with trace context
        process_message(record)

def process_message(record):
    message = json.loads(record['Sns']['Message'])
    if message['event'] == 'whitelisted-event':
        with aws_message_span("process_message_sns_receiver", message=record) as s:
            logger.info(f"Processing message: {record['messageId']}")
    
            # Parse message body
            body = json.loads(record['body'])
            logger.info(f"Message data: {body}")

Flask

from flask import Flask, jsonify
from src.rebrandly_otel import otel, logger, app_before_request, app_after_request, flask_error_handler
from datetime import datetime

app = Flask(__name__)

# Register the centralized OTEL handlers
app.before_request(app_before_request)
app.after_request(app_after_request)
app.register_error_handler(Exception, flask_error_handler)

@app.route('/health')
def health():
    logger.info("Health check requested")
    return jsonify({"status": "healthy"}), 200

@app.route('/process', methods=['POST', 'GET'])
def process():
    with otel.span("process_request"):
        logger.info("Processing POST request")

        # Simulate processing
        result = {"processed": True, "timestamp": datetime.now().isoformat()}

        logger.info(f"Returning result: {result}")
        return jsonify(result), 200

@app.route('/error')
def error():
    logger.error("Error endpoint called")
    raise Exception("Simulated error")

if __name__ == '__main__':
    app.run(debug=True)

FastAPI

# main_fastapi.py
from fastapi import FastAPI, HTTPException, Depends
from contextlib import asynccontextmanager
from src.rebrandly_otel import otel, logger, force_flush
from src.fastapi_support import setup_fastapi, get_current_span
from datetime import datetime
from typing import Optional
import uvicorn

@asynccontextmanager
async def lifespan(app: FastAPI):
    # Startup
    logger.info("FastAPI application starting up")
    yield
    # Shutdown
    logger.info("FastAPI application shutting down")
    force_flush()

app = FastAPI(title="FastAPI OTEL Example", lifespan=lifespan)

# Setup FastAPI with OTEL
setup_fastapi(otel, app)

@app.get("/health")
async def health():
    """Health check endpoint."""
    logger.info("Health check requested")
    return {"status": "healthy"}

@app.post("/process")
@app.get("/process")
async def process(span = Depends(get_current_span)):
    """Process endpoint with custom span."""
    with otel.span("process_request"):
        logger.info("Processing request")

        # You can also use the injected span directly
        if span:
            span.add_event("custom_processing_event", {
                "timestamp": datetime.now().isoformat()
            })

        # Simulate some processing
        result = {
            "processed": True,
            "timestamp": datetime.now().isoformat()
        }

        logger.info(f"Returning result: {result}")
        return result

@app.get("/error")
async def error():
    """Endpoint that raises an error."""
    logger.error("Error endpoint called")
    raise HTTPException(status_code=400, detail="Simulated error")

@app.get("/exception")
async def exception():
    """Endpoint that raises an unhandled exception."""
    logger.error("Exception endpoint called")
    raise ValueError("Simulated unhandled exception")

@app.get("/items/{item_id}")
async def get_item(item_id: int, q: Optional[str] = None):
    """Example endpoint with path and query parameters."""
    with otel.span("fetch_item", attributes={"item_id": item_id, "query": q}):
        logger.info(f"Fetching item {item_id} with query: {q}")

        if item_id == 999:
            raise HTTPException(status_code=404, detail="Item not found")

        return {
            "item_id": item_id,
            "name": f"Item {item_id}",
            "query": q
        }

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)

PyMySQL Database Instrumentation

The SDK provides connection-level instrumentation for PyMySQL that automatically traces all queries without requiring you to instrument each query individually.

import pymysql
from rebrandly_otel import otel, logger, instrument_pymysql

# SDK auto-initializes on import

# Create and instrument your connection
connection = pymysql.connect(
    host='localhost',
    user='your_user',
    password='your_password',
    database='your_database'
)

# Instrument the connection - all queries are now automatically traced
connection = instrument_pymysql(otel, connection, options={
    'slow_query_threshold_ms': 1000,  # Queries over 1s flagged as slow
    'capture_bindings': False  # Set True to capture query parameters
})

# Use normally - all queries automatically traced
with connection.cursor() as cursor:
    cursor.execute("SELECT * FROM users WHERE id = %s", (123,))
    result = cursor.fetchone()
    logger.info(f"Found user: {result}")

connection.close()
otel.force_flush()

Features:

• Automatic span creation for all queries
• Query operation detection (SELECT, INSERT, UPDATE, etc.)
• Slow query detection and flagging
• Duration tracking
• Error recording with exception details
• Optional query parameter capture (disabled by default for security)

Environment configuration:

• PYMYSQL_SLOW_QUERY_THRESHOLD_MS: Threshold for slow query detection (default: 1500ms)

More examples

You can find More examples here

Troubleshooting

Common Issues

No Data Exported

Symptoms: Telemetry data not appearing in your observability backend.

Solutions:

1. Verify OTEL_EXPORTER_OTLP_ENDPOINT is correctly set:

   echo $OTEL_EXPORTER_OTLP_ENDPOINT
   

2. Check network connectivity to the collector:

   curl -v $OTEL_EXPORTER_OTLP_ENDPOINT
   

3. Enable debug mode to see console output:

   export OTEL_DEBUG=true
   python app.py
   

4. Verify the collector is running and accepting connections
5. Check for firewall rules blocking outbound gRPC traffic (port 4317)

Missing Traces in Lambda

Symptoms: Lambda function executes but no traces appear in backend.

Solutions:

1. Ensure force_flush() is called before handler returns:

   from rebrandly_otel import lambda_handler, force_flush
   
   @lambda_handler(name="my-function")
   def handler(event, context):
       # Your code
       force_flush(timeout_millis=5000)  # Explicitly flush
       return response
   

2. Verify Lambda timeout allows enough time for flush (add 2-3 seconds buffer)
3. Check Lambda execution role has network access to OTLP endpoint
4. Verify environment variables are set in Lambda configuration
5. Check CloudWatch Logs for error messages

Trace Context Not Propagating

Symptoms: Distributed traces appear as disconnected spans instead of a unified trace.

Solutions:

1. Verify message attributes are being sent:

   # For SQS
   trace_attrs = otel.tracer.get_attributes_for_aws_from_context()
   response = sqs.send_message(
       QueueUrl=queue_url,
       MessageBody=json.dumps(data),
       MessageAttributes=trace_attrs  # Don't forget this!
   )
   

2. Check that receiving end uses aws_message_span() or @aws_message_handler:

   with aws_message_span("process-message", message=record):
       # Processing logic
       pass
   

3. For HTTP services, verify headers are being propagated:

   import requests
   from opentelemetry.propagate import inject
   
   headers = {}
   inject(headers)  # Injects traceparent header
   response = requests.get(url, headers=headers)
   

High Memory Usage

Symptoms: Application memory usage grows over time.

Solutions:

1. Reduce batch export interval:

   export BATCH_EXPORT_TIME_MILLIS=50  # Flush more frequently (default: 100)
   

2. Implement sampling for high-traffic applications:

   from opentelemetry.sdk.trace.sampling import ParentBasedTraceIdRatioBased
   
   # Sample 10% of traces
   sampler = ParentBasedTraceIdRatioBased(0.1)
   

3. Monitor metric cardinality - avoid high-cardinality attributes:

   # Bad: user_id has millions of possible values
   counter.add(1, {"user_id": user_id})
   
   # Good: user_tier has limited values
   counter.add(1, {"user_tier": "premium"})
   

4. Check for span leaks - ensure all spans are properly closed

Import Errors

Symptoms: ModuleNotFoundError or ImportError when importing rebrandly_otel.

Solutions:

1. Verify installation:

   pip show rebrandly-otel
   

2. Reinstall if necessary:

   pip uninstall rebrandly-otel
   pip install rebrandly-otel
   

3. Check Python version compatibility (requires Python 3.7+):

   python --version
   

4. For Lambda, ensure package is included in deployment package or layer

Database Instrumentation Not Working

Symptoms: Database queries not appearing as spans.

Solutions:

1. Verify connection is instrumented:

   from rebrandly_otel import instrument_pymysql
   
   connection = pymysql.connect(...)
   connection = instrument_pymysql(otel, connection)  # Don't forget this!
   

2. Check that you're using the instrumented connection object
3. Verify slow query threshold settings:

   export PYMYSQL_SLOW_QUERY_THRESHOLD_MS=1000
   

Flask/FastAPI Routes Not Traced

Symptoms: HTTP requests not creating spans.

Solutions:

1. Verify middleware is registered before route definitions:

   # Flask
   app.before_request(app_before_request)
   app.after_request(app_after_request)
   app.register_error_handler(Exception, flask_error_handler)
   
   # Then define routes
   @app.route('/api/users')
   def get_users():
       ...
   

2. For FastAPI, ensure setup_fastapi() is called:

   from rebrandly_otel.fastapi_support import setup_fastapi
   setup_fastapi(otel, app)
   

3. Check for middleware conflicts with other libraries

Debugging Tips

Enable Debug Mode

See all telemetry data in console:

export OTEL_DEBUG=true
python app.py

Check Span Attributes

Print span attributes during development:

with otel.span("test-span") as span:
    print(f"Trace ID: {span.get_span_context().trace_id}")
    print(f"Span ID: {span.get_span_context().span_id}")
    # Your code

Verify Environment Variables

Check all OTEL configuration:

import os
print("OTEL_SERVICE_NAME:", os.getenv("OTEL_SERVICE_NAME"))
print("OTEL_EXPORTER_OTLP_ENDPOINT:", os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT"))
print("OTEL_DEBUG:", os.getenv("OTEL_DEBUG"))

Test Trace Context Propagation

Manually test context extraction:

from rebrandly_otel import otel

# Simulate SQS message
test_message = {
    'messageAttributes': {
        'traceparent': {'stringValue': '00-trace-id-span-id-01'}
    }
}

context = otel.tracer.extract_context_from_aws_message(test_message)
print(f"Extracted context: {context}")

Testing

Running Tests

The test suite uses pytest.

Run all tests:

pytest

Run specific test file:

pytest tests/test_flask_support.py -v
pytest tests/test_fastapi_support.py -v
pytest tests/test_usage.py -v
pytest tests/test_pymysql_instrumentation.py -v
pytest tests/test_metrics_and_logs.py -v
pytest tests/test_decorators.py -v
pytest tests/test_span_attributes_processor.py -v

Run with coverage:

pytest --cov=src --cov-report=html

Test Coverage

The test suite includes:

• Integration tests (test_usage.py): Core OTEL functionality, Lambda handlers, message processing
• Flask integration tests (test_flask_support.py): Flask setup and hooks
• FastAPI integration tests (test_fastapi_support.py): FastAPI setup and middleware
• PyMySQL instrumentation tests (test_pymysql_instrumentation.py): Database connection instrumentation, query tracing, helper functions
• Metrics and logs tests (test_metrics_and_logs.py): Custom metrics creation (counter, histogram, gauge), logging levels (info, warning, debug, error)
• Decorators tests (test_decorators.py): Lambda handler decorator, AWS message handler decorator, traces decorator, aws_message_span context manager
• Span attributes processor tests (test_span_attributes_processor.py): Automatic span attributes from OTEL_SPAN_ATTRIBUTES (31 tests)

License

Rebrandly Python SDK is released under the MIT License.

Build and Deploy

brew install pipx
pipx ensurepath
pipx install build
pipx install twine

build

twine upload dist/*

If build gives you an error, try:

pyproject-build

twine upload dist/*