dbos-inc/agent-skills/dbos-python

DBOS Python Best Practices

Guide for building reliable, fault-tolerant Python applications with DBOS durable workflows. Targets DBOS Python 3.x.

When to Apply

Reference these guidelines when:

• Adding DBOS to existing Python code
• Creating workflows and steps
• Using queues for concurrency control
• Implementing workflow communication (events, messages, streams)
• Configuring and launching DBOS applications
• Using DBOSClient from external applications
• Testing DBOS applications
• Upgrading DBOS Python 2.x code to 3.x

Rule Categories by Priority

Critical Rules

DBOS Configuration and Launch

A DBOS application MUST configure and launch DBOS inside its main function:

import os
from dbos import DBOS, DBOSConfig
@DBOS.workflow()
def my_workflow():
    pass
if __name__ == "__main__":
    config: DBOSConfig = {
        "name": "my-app",
        "application_version": "0.1.0",
        "system_database_url": os.environ.get("DBOS_SYSTEM_DATABASE_URL"),
    }
    DBOS(config=config)
    DBOS.launch()

When creating a new application, set application_version to "0.1.0". If omitted, DBOS derives an opaque hash from workflow source code. When editing an existing application, leave its configured version alone — changing it is a deployment decision (see references/advanced-versioning.md).

Workflow and Step Structure

Workflows are comprised of steps. Any function performing complex operations or accessing external services must be a step:

@DBOS.step()
def call_external_api():
    return requests.get("https://api.example.com").json()
@DBOS.workflow()
def my_workflow():
    result = call_external_api()
    return result

Key Constraints

• Do NOT call DBOS.start_workflow or DBOS.recv from a step
• Do NOT use threads to start workflows - use DBOS.start_workflow or queues
• Workflows MUST be deterministic - non-deterministic operations go in steps
• Do NOT create/update global variables from workflows or steps
• In async def code, use the _async variants of DBOS methods (await DBOS.start_workflow_async(...), send_async, recv_async, sleep_async, register_queue_async, ...): many synchronous DBOS methods (such as DBOS.sleep, DBOS.recv, DBOS.send, DBOS.set_event, DBOS.get_event, and DBOS.register_queue) raise RuntimeError when called while an event loop is running
• Register queues and create schedules AFTER DBOS.launch(); create datasources BEFORE it
• Connecting to DBOS Conductor (conductor_key, or deploying to DBOS Cloud), and auto-upgrade (auto_upgrade) require pip install "dbos[enterprise]"; auto-upgrade also needs an entitled key (conductor_key, or transact_enterprise_key in development). See references/lifecycle-config.md
• Do NOT set enable_patching (deprecated in 3.4): DBOS.patch and DBOS.deprecate_patch work without it. To let patched code recover older workflows, pin application_version or use auto-upgrade (see references/advanced-patching.md)

Removed in DBOS 3.0 (never generate these)

@DBOS.transaction / DBOS.sql_session / application_database_url (use datasources), Queue(...) (use DBOS.register_queue), partition_queue / priority_enabled, @DBOS.scheduled (use DBOS.apply_schedules), and DBOS(fastapi=...) / DBOS(flask=...). To migrate existing 2.x code, see references/advanced-upgrading-v3.md.

How to Use

Read individual rule files for detailed explanations and examples:

references/lifecycle-config.md
references/workflow-determinism.md
references/queue-concurrency.md
references/advanced-upgrading-v3.md
references/advanced-shared-database.md

If multiple applications share one system database, see references/advanced-shared-database.md.

References

• https://docs.dbos.dev/
• https://github.com/dbos-inc/dbos-transact-py