tendrl 0.1.1

A Python SDK for the Tendrl data collection platform with cross-platform UNIX socket support, offline storage, and dynamic batching. Licensed for use with Tendrl services only.

Tendrl Python SDK

A Python SDK for the Tendrl data collection platform with cross-platform UNIX socket support, offline storage, and dynamic batching.

⚠️ License Notice

This software is licensed for use with Tendrl services only.

✅ Allowed

• Use the software with Tendrl services
• Inspect and learn from the code for educational purposes
• Modify or extend the software for personal or Tendrl-related use

❌ Not Allowed

• Use in any competing product or service
• Connect to any backend not operated by Tendrl, Inc.
• Package into any commercial or hosted product (e.g., SaaS, PaaS)
• Copy design patterns or protocol logic for another system without permission

For licensing questions, contact: [e-mail]

See the LICENSE file for complete terms and restrictions.

Features

• 🔌 Cross-Platform Communication: Windows 10 1803+, Linux, macOS
• 🔄 Dual Operating Modes: HTTP API and Tendrl Nano Agent (Unix Socket)
• 💾 Offline Message Storage: SQLite-based persistence with TTL
• ⚡ Dynamic Batching: CPU/memory-aware batch sizing
• 🎯 Resource Monitoring: Automatic system resource adaptation
• 🔐 Secure Communication: AF_UNIX sockets + HTTPS API
• 📊 Performance Metrics: Built-in system monitoring utilities

Platform Support

Windows

• Requirements: Windows 10 version 1803+ or Windows Server 2019+ (Agent Mode)
• Recommended: Use Tendrl Nano Agent for optimal performance
• Agent Installation: Download and run tendrl-agent.exe with your API key
• Connection: Python SDK connects automatically to the local agent

Unix/Linux/macOS

• Native Support: Works on all modern versions
• Recommended: Use Tendrl Nano Agent for optimal performance
• Direct API: Can also connect directly to Tendrl API without local agent

Operating Modes

The Python SDK supports two operating modes optimized for different use cases:

📡 Direct API Mode (Recommended for Simplicity)

How it works: Python SDK → HTTP/2 → Tendrl Server

client = Client(mode="api", api_key="your_key")  # Direct to server

Benefits:

• Simple setup: No additional components required
• Direct control: Full visibility into HTTP requests and responses
• Dynamic batching: CPU/memory-aware batching (10-500 messages)
• Offline storage: SQLite persistence during network outages
• Connection pooling: httpx HTTP/2 connection reuse
• Automatic retries: Built-in retry mechanisms and error handling
• Resource monitoring: Adaptive behavior based on system load

Performance Characteristics:

• Light Load (< 10 msg/sec): ~2-5ms per message
• Heavy Load (100+ msg/sec): ~0.5-1ms per message (true HTTP batch requests)
• Per-message latency: ~2-10ms individual, ~0.5-1ms batched
• Batching: True HTTP batching - multiple messages per HTTP request
• Resource usage: Higher CPU/memory due to Python interpreted overhead

🚀 Nano Agent Mode (Recommended for Performance)

How it works: Python SDK → Unix Socket → Go Nano Agent → HTTP/2 → Tendrl Server

client = Client(mode="agent")  # Connects to local Go agent

Benefits:

• Superior performance: 5-20x faster due to Go efficiency + Unix socket IPC
• Single point of egress: One agent serves multiple applications/languages on same host
• Enhanced batching: Go agent optimizes HTTP request batching better than Python
• Shared efficiency: Single Go process serves multiple Python applications
• Centralized configuration: Manage API keys and settings in one place
• Optimized connection management: Go's superior HTTP/2 implementation
• Advanced resource adaptation: More sophisticated CPU/memory monitoring
• Production reliability: Battle-tested Go networking stack
• Lower system overhead: Compiled efficiency vs Python interpreted overhead

Performance Characteristics:

• Light Load (< 10 msg/sec): ~0.5ms per message
• Heavy Load (100+ msg/sec): ~0.1ms per message (optimized batching)
• Per-message latency: ~0.1-0.5ms (Unix socket + Go efficiency)
• Batching: More intelligent batching algorithms in Go
• Resource usage: Significantly lower CPU/memory per message

📊 Feature & Performance Comparison

Feature	Agent Mode	Direct API Mode
Performance (Light Load)	~0.5ms/msg	~2-5ms/msg
Performance (Heavy Load)	~0.1ms/msg (batched)	~0.5-1ms/msg (batched)
Message Batching	✅ Intelligent (10-500 msgs)	✅ True HTTP batching (10-500 msgs)
Offline Storage	✅ SQLite persistence	✅ SQLite persistence
Connection Pooling	✅ Optimized Go HTTP/2 pools	✅ httpx connection pooling
Automatic Retries	✅ Built-in agent logic	✅ SDK retry mechanisms
Resource Usage	✅ Low (shared Go process)	⚠️ Higher (Python overhead)
CPU/Memory Adaptation	✅ Dynamic batching	❌ Fixed behavior
Multi-App/Language Support	✅ Single agent serves all	❌ Each app manages own connection
Setup Complexity	⚠️ Requires agent install	✅ Simple (SDK only)
Network Resilience	✅ Agent handles outages	✅ SDK handles outages
Debugging	⚠️ Two-component system	✅ Direct HTTP visibility
Deployment	⚠️ Two processes to manage	✅ Single process

🏆 Key Differentiators

Agent Mode Advantages:

• Intelligent Batching: Combines multiple messages into single HTTP requests (major performance gain)
• Go Performance: Compiled efficiency vs Python interpreted overhead
• Resource Efficiency: Single optimized process serves multiple Python applications
• Adaptive Behavior: Automatically adjusts batching based on system load

Direct API Mode Advantages:

• Simplicity: No additional components to install or manage
• Direct Control: Full visibility into HTTP requests and responses
• Single Process: Easier debugging and deployment
• Immediate Feedback: Each publish() call gets direct server response

💡 Choosing the Right Mode

Use Direct API Mode when:

• Simplicity is priority: Quick setup, no additional components
• Development/Testing: Prototyping, debugging, local development
• Low to moderate volume: < 50 messages per second
• Deployment constraints: Can't install additional services
• Direct feedback needed: Want immediate HTTP responses per message

Use Nano Agent Mode when:

• Performance is priority: Need maximum throughput and lowest latency
• High volume: > 50 messages per second
• Production environments: Need optimized resource usage
• Multiple applications: Sharing agent across several Python processes
• Multi-language environment: Different programming languages on same host
• Centralized management: Want single point for configuration and monitoring
• Resource constrained: Every CPU cycle and MB matters

Installation

pip install tendrl

Basic Usage

from tendrl import Client

# Initialize client with direct API key
client = Client(mode="api", api_key="your_key")

# Or use environment variable
# export TENDRL_KEY=your_key
client = Client(mode="api")

# One-time data collection using decorator
@client.tether(tags=["metrics"])
def collect_metrics():
    return {
        "cpu_usage": 42.0,
        "memory": 84.0
    }

# Periodic data collection
@client.tether(tags=["system"], interval=60)
def system_stats():
    return {
        "uptime": 3600,
        "load": 1.5
    }

# Start the client
client.start()

## API Reference

### Client Initialization

```python
Client(
    mode: str = "api",                 # "api" or "agent"
    api_key: str = None,               # API key for authentication
    check_msg_rate: float = 3,         # Check for new messages rate (Seconds)
    debug: bool = False,               # Enable debug logging
    target_cpu_percent: float = 70,    # Target CPU usage
    target_mem_percent: float = 80,    # Target memory usage
    min_batch_size: int = 10,          # Minimum batch size
    max_batch_size: int = 500,         # Maximum batch size
    offline_storage: bool = False,     # Enable offline storage
    db_path: str = "tendrl_offline.db" # Sqlite3 file path
)

Message Publishing

# Direct message publishing
client.publish(
    msg: dict,                   # Message data
    tags: List[str] = None,      # Message tags
    entity: str = "",            # Send to another entity
    wait_response: bool = False, # Wait for response
    timeout: int = 5             # Response timeout
) -> str:                        # Returns message ID if wait_response=True

Tether Decorator

@client.tether(
    tags: List[str] = None,      # Message tags
    write_offline: bool = False,  # Enable offline storage
    db_ttl: int = 200,           # Offline storage TTL
    entity: str = "",            # Send to another entity
)

Advanced Usage

Resource Monitoring

# Get current system metrics
metrics = client.get_system_metrics()
print(f"CPU: {metrics.cpu_usage}%")
print(f"Memory: {metrics.memory_usage}%")
print(f"Queue Load: {metrics.queue_load}%")

# Configure resource limits
client = Client(
    target_cpu_percent=60.0,
    target_mem_percent=70.0
)

Batch Processing Configuration

client = Client(
    min_batch_size=10,
    max_batch_size=500,
    min_batch_interval=0.1,
    max_batch_interval=1.0
)

Offline Storage

# Enable offline storage
client = Client(
    offline_storage=True,
    db_path="/path/to/storage.db"
)

# Tether with offline storage
@client.tether(tags=["metrics"], write_offline=True, db_ttl=3600)
def collect_metrics():
    return {"data": "value"}

Logging Configuration

import logging

logging.basicConfig(
    level=logging.DEBUG,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)

client = Client(debug=True)

Best Practices

1. Resource Management

• Use appropriate batch sizes
• Monitor system metrics
• Implement proper cleanup

2. Error Handling

• Use retries for transient failures
• Log errors appropriately
• Handle offline scenarios

3. Performance

• Use batch processing
• Monitor queue size
• Configure appropriate intervals

4. Security

• Secure API keys
• Use HTTPS
• Validate input data

Troubleshooting

Common issues and solutions:

Windows Issues

1. Agent Connection Error

# Ensure Tendrl Nano Agent is running
# Check if tendrl-agent.exe process is active in Task Manager
# Or run: tasklist /FI "IMAGENAME eq tendrl-agent.exe"

# Start the agent if not running
# tendrl-agent.exe -apiKey=YOUR_API_KEY

2. Agent Not Found

# Verify Windows version compatibility
import platform
print(f"Windows version: {platform.version()}")
# Requires Windows 10 1803+ or Windows Server 2019+

# Ensure agent directory exists
import os
os.makedirs("C:\\ProgramData\\tendrl", exist_ok=True)

General Issues

1. Connection Issues

# Increase timeout
client = Client(timeout=10)

2. Queue Full

# Increase batch processing
client = Client(
    max_batch_size=1000,
    min_batch_interval=0.05
)

3. High Memory Usage

# Adjust batch size
client = Client(
    max_batch_size=100,
    target_mem_percent=60.0
)

4. Message Loss

# Enable offline storage
client = Client(
    offline_storage=True,
    db_ttl=3600
)

Offline Message Flow

The following shows how messages with tags are handled during offline periods:

@tether(tags=['sensor', 'prod'])
         ↓
   Message Created
         ↓
   Connection Check
         ↓
    ┌─────────────┐
    │  Online?    │
    └─────────────┘
         ↓
    ┌────┴────┐
    │         │
   Yes       No
    │         │
    ↓         ↓
Send with   Store in SQLite
  Tags      WITH TAGS
    │         │
    │    Connection Restored
    │         ↓
    │    Batch Processing (50 msgs)
    │         ↓
    │    Parse data + tags
    │         ↓
    │    Reconstruct message
    │         ↓
    └─────────┘
         ↓
   Send to Server
         ↓
Server processes webhook
    with tags

Key Features

• Tags Preservation: Tags are stored with offline messages and restored when sent
• Batched Processing: Large offline backlogs are processed in manageable batches (50 messages)
• Fault Tolerance: Failed batches don't affect successfully sent messages
• Webhook Compatibility: Server receives messages with proper tags for processing

Note: The client uses queue-based processing, making it effectively non-blocking. If you need async APIs in an async application, you can wrap calls using asyncio.to_thread(client.publish, data).

Using with Tendrl Nano Agent

For optimal performance (see Operating Modes comparison), use the Tendrl Nano Agent:

1. Start the Tendrl Nano Agent

Windows:

# Download tendrl-agent.exe and run
tendrl-agent.exe -apiKey=YOUR_API_KEY

Unix/Linux:

# Download tendrl-agent and run
export TENDRL_API_KEY=your_key
./tendrl-agent

2. Connect Python SDK to Agent

from tendrl import Client

# Agent mode - connects to local Tendrl Nano Agent
client = Client(mode="agent")
client.start()

# Publish data through the agent
client.publish({"sensor": "temperature", "value": 23.5})

Alternative: Direct API Mode

# Direct API mode - connects directly to Tendrl servers
client = Client(mode="api", api_key="your_key")
client.start()

Performance Note: Agent mode provides 5-20x better performance than direct API mode. See the Operating Modes section for detailed comparison.