Temporal Python Pro

Temporal Python Pro

Master durable workflow orchestration with Temporal's Python SDK. This skill provides comprehensive guidance for building reliable, scalable distributed systems using workflows and activities.

When to Use This Skill

Use this skill proactively when:

• Designing or implementing Temporal workflows in Python
• Debugging workflow determinism issues or replay failures
• Implementing saga patterns for distributed transactions
• Setting up testing strategies for Temporal applications
• Deploying Temporal workers to production
• Optimizing workflow performance and resource utilization
• Troubleshooting common Temporal Python pitfalls
• Onboarding to Temporal Python SDK

Core Concepts

Workflows

Workflows define the orchestration logic. They MUST be deterministic - same inputs must produce same outputs.

from datetime import timedelta
from temporalio import workflow

@workflow.defn
class OrderWorkflow:
    @workflow.run
    async def run(self, order_id: str) -> str:
        # Orchestrates activities
        await workflow.execute_activity(
            reserve_inventory,
            order_id,
            start_to_close_timeout=timedelta(seconds=30)
        )

        await workflow.execute_activity(
            process_payment,
            order_id,
            start_to_close_timeout=timedelta(seconds=30)
        )

        return f"Order {order_id} completed"

Activities

Activities perform non-deterministic operations (I/O, API calls, database operations).

from temporalio import activity

@activity.defn
async def reserve_inventory(order_id: str) -> None:
    # Business logic and I/O operations
    inventory_system.reserve(order_id)

Determinism Rules

CRITICAL: Workflow code must be deterministic.

Forbidden in workflows:

• datetime.now(), time.time() → Use workflow.now()
• random.choice(), random.random() → Move to activity
• uuid.uuid4() → Use workflow.uuid() or activity
• asyncio.wait() with activities → Use asyncio.gather()
• File I/O, network calls → Move to activities
• Global mutable state → Use workflow instance variables

Common Pitfalls

AI-Generated Code Issues

AI often generates non-deterministic workflow code. Watch for:

1. System time in workflows

   # ❌ WRONG (AI-generated)
   timestamp = datetime.now()
   
   # ✅ CORRECT
   timestamp = workflow.now()
   

2. Random numbers in workflows

   # ❌ WRONG (AI-generated)
   choice = random.choice(options)
   
   # ✅ CORRECT
   choice = await workflow.execute_activity(
       random_choice, options
   )
   

3. Business logic in workflows

   # ❌ WRONG
   processed = complex_transform(data)  # In workflow
   
   # ✅ CORRECT
   processed = await workflow.execute_activity(
       transform, data
   )
   

Non-Idempotent Activities

Activities MUST be idempotent - retries should not cause side effects.

# ❌ NOT IDEMPOTENT
@activity.defn
async def charge_card(amount: float):
    payment_gateway.charge(amount)  # Charges twice on retry!

# ✅ IDEMPOTENT
@activity.defn
async def charge_card(charge_id: str, amount: float):
    payment_gateway.charge_with_idempotency_key(charge_id, amount)

Testing Strategy

Unit Tests with Time Skipping

import pytest
from temporalio.testing import WorkflowEnvironment

@pytest.mark.asyncio
async def test_order_workflow():
    async with await WorkflowEnvironment.start_time_skipping() as env:
        async with Worker(
            env.client,
            task_queue="test-queue",
            workflows=[OrderWorkflow],
            activities=[mock_activities]
        ):
            result = await env.client.execute_workflow(
                OrderWorkflow.run,
                "order-123",
                id="test-order",
                task_queue="test-queue"
            )

            assert result.success

Replay Testing (Required Before Deploying)

Always test workflow replay with production history to catch non-determinism.

async def test_replay_production():
    # Fetch history from production
    handle = prod_client.get_workflow_handle("prod-workflow-id")
    history = await handle.fetch_history()

    # Replay in test
    async with await WorkflowEnvironment.start_time_skipping() as env:
        async with Worker(
            env.client,
            task_queue="test-queue",
            workflows=[MyWorkflow],  # NEW code
            activities=[mock_activities]
        ):
            # Fails if non-deterministic!
            await env.client.execute_workflow(
                MyWorkflow.run,
                args,
                id="replay-test",
                task_queue="test-queue",
                replay_history=history
            )

Production Deployment

Worker Configuration

from temporalio.worker import Worker

worker = Worker(
    client,
    task_queue="production-queue",
    workflows=[MyWorkflows],
    activities=[MyActivities],
    max_concurrent_activities=50,
    max_concurrent_workflow_tasks=20
)

Graceful Shutdown

import signal

shutdown_event = asyncio.Event()

def handler(sig, frame):
    shutdown_event.set()

signal.signal(signal.SIGTERM, handler)

await worker.run_until(shutdown_event.is_set)
await worker.shutdown()  # Graceful shutdown

Quick Reference

When to Use Temporal

Use Temporal for:

• Long-running processes (minutes to months)
• Multi-step orchestrations with human approval
• Reliable retries and compensation
• Distributed transactions (saga pattern)
• Stateful workflows across failures

Use regular async for:

• Fast operations (< 10 seconds)
• Simple orchestration
• No durability required

Key Resources

• Core Concepts - Workflows, activities, determinism, signals, queries
• Best Practices - Workflow design, activity patterns, error handling
• Common Pitfalls - Determinism violations, AI mistakes, production issues
• Testing Guide - Unit tests, integration tests, replay testing, time skipping
• Production Deployment - Worker config, monitoring, scaling, security

Pre-Deployment Checklist

Before deploying workflow code:

• All time references use workflow.now()
• No random, uuid, datetime modules in workflows
• No asyncio.wait() with activities
• All activities are idempotent
• All activities have appropriate timeouts
• Business logic moved to activities
• Workflow replay tested with production history
• Search attributes configured for queries
• Unit tests pass with time skipping
• Integration tests with real activities pass