How to customize Prefect's logging configuration

Prefect relies on the standard Python implementation of logging configuration. The full specification of the default logging configuration for any version of Prefect can always be inspected here. The default logging level is INFO.

Customize logging configuration

Prefect provides several settings to configure the logging level and individual loggers.

Any value in Prefect's logging configuration file can be overridden through a Prefect setting of the form PREFECT_LOGGING_[PATH]_[TO]_[KEY]=value corresponding to the nested address of the field you are configuring.

For example, to change the default logging level for flow runs but not task runs, update your profile with:

prefect config set PREFECT_LOGGING_LOGGERS_PREFECT_FLOW_RUNS_LEVEL="ERROR"

or set the corresponding environment variable:

export PREFECT_LOGGING_LOGGERS_PREFECT_FLOW_RUNS_LEVEL="ERROR"

You can also configure the "root" Python logger. The root logger receives logs from all loggers unless they explicitly opt out by disabling propagation. By default, the root logger is configured to output WARNING level logs to the console. As with other logging settings, you can override this from the environment or in the logging configuration file. For example, you can change the level with the PREFECT_LOGGING_ROOT_LEVEL environment variable.

In some situations you may want to completely overhaul the Prefect logging configuration by providing your own logging.yml file. You can create your own version of logging.yml in one of two ways:

1. Create a logging.yml file in your PREFECT_HOME directory (default is ~/.prefect).
2. Specify a custom path to your logging.yml file using the PREFECT_LOGGING_CONFIG_PATH setting.

If Prefect cannot find the logging.yml file at the specified location, it will fall back to using the default logging configuration.

See the Python Logging configuration documentation for more information about the configuration options and syntax used by logging.yml.

Formatters

Prefect log formatters specify the format of log messages. The default formatting for task and flow run records is "%(asctime)s.%(msecs)03d | %(levelname)-7s | Task run %(task_run_name)r - %(message)s" for tasks and similarly "%(asctime)s.%(msecs)03d | %(levelname)-7s | Flow run %(flow_run_name)r - %(message)s" for flows.

The variables available to interpolate in log messages vary by logger. In addition to the run context, message string, and any keyword arguments, flow and task run loggers have access to additional variables.

The flow run logger has the following variables available for formatting:

• flow_run_name
• flow_run_id
• flow_name
• deployment_name

The task run logger has the following variables available for formatting:

• task_run_id
• flow_run_id
• task_run_name
• task_name
• flow_run_name
• flow_name

You can specify custom formatting by setting the relevant environment variable or by modifying the formatter in a custom logging.yml file as described earlier. For example, the following changes the formatting for the flow runs formatter:

PREFECT_LOGGING_FORMATTERS_STANDARD_FLOW_RUN_FMT="%(asctime)s.%(msecs)03d | %(levelname)-7s | %(flow_run_id)s - %(message)s"

The resulting messages, using the flow run ID instead of name, look like this:

10:40:01.211 | INFO    | e43a5a80-417a-41c4-a39e-2ef7421ee1fc - Created task run
'othertask-1c085beb-3' for task 'othertask'

Styles

By default, Prefect highlights specific keywords in the console logs with a variety of colors. You can toggle highlighting on/off with the PREFECT_LOGGING_COLORS setting:

PREFECT_LOGGING_COLORS=False

You can also change what gets highlighted and even adjust the colors by updating the styles - see the styles section of the Prefect logging configuration file for available keys.

You can even build your own handler with a custom highlighter. For example, to additionally highlight emails:

1. Copy and paste the following code into my_package_or_module.py (rename as needed) in the same directory as the flow run script; or ideally as part of a Python package so it's available in site-packages and accessible anywhere within your environment.

import logging
from typing import Dict, Union

from rich.highlighter import Highlighter

from prefect.logging.handlers import PrefectConsoleHandler
from prefect.logging.highlighters import PrefectConsoleHighlighter

class CustomConsoleHighlighter(PrefectConsoleHighlighter):
    base_style = "log."
    highlights = PrefectConsoleHighlighter.highlights + [
        # ?P<email> is naming this expression as `email`
        r"(?P<email>[\w-]+@([\w-]+\.)+[\w-]+)",
    ]

class CustomConsoleHandler(PrefectConsoleHandler):
    def __init__(
        self,
        highlighter: Highlighter = CustomConsoleHighlighter,
        styles: Dict[str, str] = None,
        level: Union[int, str] = logging.NOTSET,
   ):
        super().__init__(highlighter=highlighter, styles=styles, level=level)

2. Update ~/.prefect/logging.yml to use my_package_or_module.CustomConsoleHandler and additionally reference the base_style and named expression: log.email.

    console_flow_runs:
        level: 0
        class: my_package_or_module.CustomConsoleHandler
        formatter: flow_runs
        styles:
            log.email: magenta
            # other styles can be appended here, e.g.
            # log.completed_state: green

3. On your next flow run, text that looks like an email is highlighted. For example, [email protected] is colored in magenta below:

from prefect import flow
from prefect.logging import get_run_logger

@flow
def log_email_flow():
    logger = get_run_logger()
    logger.info("[email protected]")

log_email_flow()

Apply markup in logs

To use Rich's markup in Prefect logs, first configure PREFECT_LOGGING_MARKUP:

PREFECT_LOGGING_MARKUP=True

The following will highlight "fancy" in red:

from prefect import flow
from prefect.logging import get_run_logger

@flow
def my_flow():
    logger = get_run_logger()
    logger.info("This is [bold red]fancy[/]")

my_flow()

If enabled, strings that contain square brackets may be inaccurately interpreted and lead to incomplete output. For example, DROP TABLE [dbo].[SomeTable];" outputs DROP TABLE .[SomeTable];. </Warning>

Include logs from other libraries

By default, Prefect won't capture log statements from libraries that your flows and tasks use. You can tell Prefect to include logs from these libraries with the PREFECT_LOGGING_EXTRA_LOGGERS setting.

To use this setting, specify one or more Python library names to include, separated by commas. For example, if you want Prefect to capture Dask and SciPy logging statements with your flow and task run logs, use:

PREFECT_LOGGING_EXTRA_LOGGERS=dask,scipy

Configure this setting as an environment variable or in a profile. See Settings for more details about how to use settings.

Disable logging

Prefect prints run logs to the console and sends them to the API. Use the settings in this section to turn off one or both outputs.

Stop sending logs to the API

Set PREFECT_LOGGING_TO_API_ENABLED to false to keep logs out of the API. This setting works with Prefect Cloud and with a self-hosted Prefect server.

prefect config set PREFECT_LOGGING_TO_API_ENABLED=false

[logging.to_api]
enabled = false

When this setting is false:

• Flow run logs, task run logs, Prefect Cloud worker logs, and logs from extra loggers are not sent to the API.
• Logs still print to the console.
• States, events, artifacts, and other run data are still sent to the API. Only log records are affected.

Where to set it

Prefect reads this setting in the process that runs your flow. Set it at the scope that matches how you run your flows:

Environment variables from these scopes are merged key by key, and the most specific scope wins. A flow run's env overrides the deployment's, the deployment's overrides the work pool's, and all of them override the worker's own settings.

Silence a logger

To stop a logger from writing to the console and to the API, raise its level or disable it.

To keep only WARNING and higher records from task runs, raise the level of the task run logger:

prefect config set PREFECT_LOGGING_LOGGERS_PREFECT_TASK_RUNS_LEVEL="WARNING"

To disable the flow run and task run loggers for a block of code, use disable_run_logger. Inside the block, get_run_logger() returns a logger that discards every record, even outside of a run. This is useful when you call a task's function directly in a test:

from prefect import task
from prefect.logging import disable_run_logger, get_run_logger


@task
def process_order(order_id: int) -> int:
    get_run_logger().info(f"Processing order {order_id}")
    return order_id


with disable_run_logger():
    assert process_order.fn(42) == 42

See Test workflows for more testing patterns.

Keep one message out of the API

To print a single log record to the console without sending it to the API, pass send_to_api=False in the extra argument of the log call:

from prefect import flow
from prefect.logging import get_run_logger


@flow
def sync_customers():
    logger = get_run_logger()
    logger.info("Starting customer sync")
    logger.info("Local debug detail", extra={"send_to_api": False})


sync_customers()

The second message prints to the console but is not sent to the API.

Silence the warning for logs outside of a run

Prefect can only send a log record to the API when the record belongs to a flow run. If a logger that is attached to the API handler is used outside of a run, Prefect drops the record and prints a warning. This happens, for example, when a library listed in PREFECT_LOGGING_EXTRA_LOGGERS logs outside of a flow run.

To turn the warning off:

prefect config set PREFECT_LOGGING_TO_API_WHEN_MISSING_FLOW=ignore

The default value is warn. Set the value to error to raise an exception instead.