Python cron jobs without the server.

Where it breaks.

The server is the single point of failure, and everything around it drifts.

• The server has to stay up. A reboot, a disk failure, or a missed patch kills every job on the host.
• No retry. A failed run is a missed run. You find out when someone notices the data is stale.
• Logs go to stdout or a file. Finding what went wrong means SSH, grep, and hope.
• The crontab and the script drift apart. Someone edits one without updating the other.
• Dependencies rot. The Python version, the pip packages, and the OS underneath all age at different rates.
• No rollback. A broken change means restoring from backup or rewriting under pressure.

A scheduled task on a managed runtime.

Write the script, set the schedule, and the platform handles the server, the runtime, and the logs.

rows = qbash.integrations.postgres.execute_query(
    query="SELECT department, sum(amount) FROM expenses "
    "WHERE date >= current_date - interval '7 days' "
    "GROUP BY department"
)["result"]

report = qbash.ai.run_prompt(
    slug="weekly-expense-summary",
    variables={"rows": rows},
)

qbash.integrations.slack.create_message(
    channel="#finance", text=report["summary"],
)

qbash.integrations.outlook.send_message(
    to_recipients="cfo@example.com",
    subject="Weekly expense summary",
    body=report["body"],
)

Runs without a host.

The server and the crontab are handled by the platform, and every run is logged with what happened.

issues = qbash.integrations.postgres.execute_query(
    query="SELECT * FROM checks WHERE status = 'failing'"
)["result"]

if issues:
    qbash.integrations.slack.create_message(
        channel="#alerts",
        text=f"Found {len(issues)} failing checks",
    )

    for issue in issues:
        qbash.integrations.asana.create_task(
            name=issue["name"], project_gid="1200000000000001",
            notes=issue["details"],
        )

results = qbash.parallel(
    check_endpoint, endpoints,
    concurrency=10, max_per_minute=60,
    checkpoint="health-check", retry=2,
)

failures = [r for r in results.results if not r["healthy"]]

When a Python job outgrows the Lambda 15-minute timeout.

The Lambda timeout defaults to 3 seconds and goes up to 900 seconds in Configuration > General > Timeout. Past 15 minutes, the AWS path is Step Functions, which chains invocations with a state machine defined in JSON plus IAM roles. ECS removes the timeout but adds containerization. A task runs the same Python for up to 12 hours.

items = qbash.integrations.postgres.execute_query(
    query="SELECT * FROM items WHERE processed = false"
)["result"]

results = qbash.parallel(
    process_item, items,
    concurrency=10, max_per_minute=120,
    checkpoint="batch-process", retry=3,
)

qbash.integrations.slack.create_message(
    channel="#data", text=f"Processed {results.completed} items",
)

Related