data-annalist 0.3.3

Audit trail generator for data processing scripts.

Annalist

Audit trail generator for data processing scripts.

• Free software: GNU General Public License v3

• Documentation: https://annalist.readthedocs.io.

Usage

Create an Annalist object at the base of the module you’d like to audit. use the @Annalist.annalize decorator on any function you would like to annalize

from annalist.annalist import Annalist
from annalist.decorators import function_logger

ann = Annalist()
ann.configure()

@function_logger
def example_function():
    ...

Annalist also works on class methods, with the help of the ClassLogger decorator.

from annalist.decorators import ClassLogger

class ExampleClass():

    # Initializers can be annalized just fine
    @function_logger
    __init__(self, arg1, arg2):
        self.arg1 = arg1
        self._arg2 = arg2
        ...

    # DO NOT put an annalizer on a property definition.
    # The annalizer calls the property itself, creating infinite recursion.
    @property
    def arg2(self):
        return self._arg2

    # Putting an annalizer on a setter is fine though.
    # Just make sure you put it after the setter decorator.
    @function_logger
    @arg2.setter
    def arg2(self, value):
        self._arg2 = value

    # DO NOT put it on the __repr__ either.
    # Same as before, this creates infinite recursion.
    def __repr__(self):
        return f"{str(arg1)}: {str(arg2)}"

In the main script, the Annalist object must be called again. This will point to the singleton object initialized in the dependency. The annalist must be configured before usage.

>>> ann = Annalist()
>>> ann.configure(logger_name="Example Logger", analyst_name="Speve")

Now the annalized code can be run like normal, and will be audited.

>>> example_function()
2023/11/2 09:42:13 | INFO | example_function called by Speve as part of Example Logger session

Formatters

Annalist is built on the standard python logging library. Formatters can be specified in the same syntax as is documented in the logging docs. The available fields can be found in Fields.

Annalist supports two formatters. The File formatters formats the output to the logfile, and Stream formatter formats the console output.

from annalist.annalist import Annalist

ann = Annalist()
ann.configure(...)

ann.set_file_formatter(
    "%(asctime)s, %(analyst_name)s, example_funtion "
    "| %(message)s",
)

ann.set_stream_formatter(
    "%(asctime)s, %(function_name)s "
)

In this example, the console output might be

>>> example_function()
2023/11/2 09:42:13, example_function

whereas the contents of the logfile might be:

2023/11/2 09:42:13, example_function, Speve | This is an example.

Fields

Annalist collects information about a decorated function and makes those available as fields. Additionally, the fields from the logging library are also available, although they are generally less useful. Below are all the useful features that are available. See all the logging fields here.The reason for their limited usefulness are that most of the code references made there point to the annalist library, and not the decorated code.

All the fields that we consider useful are listed below:

Field	Description	Source
analyst_name	Name of the analyst writing the script	User configured
function_name	Function Name	Function Inspection
function_doc	Function Docstring	Function Inspection
ret_val	Return value	Function Inspection
ret_val_type	Return value type	Function Inspection
ret_annotation	Annotation of return value	Function Inspection
params	Input parameters	Function Inspection
asctime	Time of function call	Logging Library
levelname	Logging level name	Logging Library
levelno	Logging level number	Logging Library
message*	Needs to be passed as extra param	Logging Library
name	Logger name	Logging Library

The message field is an optional parameter that can be passed directly to the decorator. This is the simplest way to add more information to a function log.

@function_logger(message="this is a message")
def example_function():
    ...

You can also specify the level of the logger in the same way, as a decorator keword argument.

@function_logger(level="DEBUG")
def example_function():
    ...

Unfortunately, Annalist does not yet offer support of passing these fields into the @ClassLogger. However, we can still get information to the logger by inspecting the method arguments, and the attributes on the class instance. Consider the following setup:

from annalist.decorators import ClassLogger

class MyClass:
    @ClassLogger
    def __init__(attr, prop):
        self.attr = attr
        self._prop = prop

    @property
    def prop(self):
        return prop

    @ClassLogger
    @prop.setter
    def prop(self, value):
        self._prop = value

    @ClassLogger
    def square_attr(self):
        return self.attr ** 2

    @ClassLogger
    def add_prop_to_attr(self):
        return attr + prop

    @ClassLogger
    @staticmethod
    def increment_value(attr):
        return attr += 1

Note the two class attributes named attr and prop. We can track these properties based on their variable names by passing it into the formatter:

>>> from annalist.annalist import Annalist
>>> ann = Annalist()
>>> ann.configure(...)
>>> ann.add_stream_formatter("%(function_name)s | prop: %(prop)s | attr: %(attr)s")

The ClassLogger decorator activates upon runtime and inspects the namespace. First, it looks for the attribute in the names of the input arguments of the decorated function. If found, it sends it to the formatter (See “Custom Fields”):

>>> mc = MyClass(7, 2)
>>> mc.prop = 3
prop | prop: 3 | attr: 7

Notice that the setter of prop caused ClassLogger to look for the values of prop and attr on the mc instance.

>>> mc.square_attr()
49
square_attr | prop: 3 | attr: 7

Notice how the function square_attr did not alter the value of attr.

Because this logger is sensitive to the state of the logger, it is important to be weary of variable names.

>>> mc.increment_value(5)
6
square_attr | prop: 3 | attr: 5

Notice how, despite having no real reference to the attribute attr on the namespace, the logger found the input argument named attr, and associated this with the attribute it is logging. I believe this to be a useful feature, but care should be taken when using it like this.

Custom Fields

Annalist accepts any number of arbitrary fields in the formatter. If these fields are not one of the fields available by default, the fields is dynamically added and processed. However, this field must then be passed to the decorator in the extra_info argument.

For example, you might set the formatter as follows. In this example, the fields site and hts_file are custom, and are not available by default.

annalizer.set_file_formatter(
    "%(asctime)s, %(analyst_name)s, %(site)s, %(hts_file)s "
    "| %(message)s",
)

Then, passing those parameters into the example function looks like this:

hts_file = "file.hts"

@ann.annalize(
    level="INFO",
    message="This decorator passes extra parameters",
    extra_info={
        "site_name": "Site one",
        "hts_file": hts_file,
    }
)
def example_function():
    ...

If the custom fields are not included in a function decorator, they will simply default to None.

When using Annalist in a class method, you might want to log class properties. Unfortunately, the following syntax will not work, since the decorator has no knowledge of the class instance (self).

class ExampleClass:
    ...

    @ann.annalize(
        level="INFO",
        message="This decorator passes extra parameters",
        extra_info={
            "site_name": self.site_name, # THIS DOES NOT WORK!
            "hts_file": self.hts_file, # THIS DOES NOT WORK!
        }
    )
    def example_method(self):
        ...

In this case, you would need to wrap your method as a function in a method that passes the instance context to the decorator.

class ExampleClass:
    ...


    def example_function(self):
        @ann.annalize(
            level="INFO",
            message="This decorator passes extra parameters",
            extra_info={
                "site_name": self.site_name,
                "hts_file": self.hts_file,
            }
        )
        def example_function():
            ...

        example_function() # OR return example_function()

Notice that I gave the same function name to the outer and inner functions. This seems to work consistently by my testing since the two functions are in different name-spaces. I’m not sure if this is good practice though. But it keeps the logs nice and clean and non-confusing.

Levels

Annalist uses the levels as defined in the logging library. Upon configuration, the default level can be set, which is the level at which all logs are logged unless overridden. The default value for default level is “INFO”.

ann.configure(
    analyst_name="Speve",
    stream_format_str=format_str,
    level_filter="WARNING",
)

A annalized method can be logged at a raised or lowered level by specifying the logging level explicitely in the decorator:

@ann.annalize(level="DEBUG")
def unimportant_function():
    ...

Feature Roadmap

This roadmap outlines the planned features and milestones for the development of our deterministic and reproducible process auditing system.

Milestone 1: Audit Logging Framework

x Develop a custom audit logging framework or class. x Capture function names, input parameters, return values, data types, and timestamps. x Implement basic logging mechanisms for integration.

Milestone 1.5: Hilltop Auditing Parity

x Define custom fields and formatters x Manage logger levels correctly

Milestone 2: Standardized Logging Format

• Define a standardized logging format for comprehensive auditing.

• Ensure consistency and machine-readability of the logging format.

Milestone 3: Serialization and Deserialization

• Implement serialization and deserialization mechanisms.

• Store and retrieve complex data structures and objects.

• Test serialization for data integrity.

Milestone 4: Versioning and Dependency Tracking

• Capture and log codebase version (Git commit hash) and dependencies.

• Ensure accurate logging of version and dependency information.

Milestone 5: Integration Testing

• Create integration tests using the audit logging framework.

• Log information during the execution of key processes.

• Begin development of process recreation capability.

Milestone 6: Reproduction Tool (Partial)

• Develop a tool or script to read and reproduce processes from the audit trail.

• Focus on recreating the environment and loading serialized data.

Milestone 7: Documentation (Partial)

• Create initial documentation.

• Explain how to use the audit logging framework and the audit trail format.

• Document basic project functionalities.

Milestone 8: Error Handling

• Implement robust error handling for auditing and reproduction code.

• Gracefully handle potential issues.

• Provide informative and actionable error messages.

Milestone 9: MVP Testing

• Conduct testing of the MVP.

• Reproduce processes from the audit trail and verify correctness.

• Gather feedback from initial users within the organization.

Milestone 10: MVP Deployment

• Deploy the MVP within the organization.

• Make it available to relevant team members.

• Encourage usage and collect user feedback.

Milestone 11: Feedback and Iteration

• Gather feedback from MVP users.

• Identify shortcomings, usability issues, or missing features.

• Prioritize and plan improvements based on user feedback.

Milestone 12: Scaling and Extending

• Explore scaling the solution to cover more processes.

• Add additional features and capabilities to enhance usability.

Please note that milestones may overlap, and the order can be adjusted based on project-specific needs. We aim to remain flexible and responsive to feedback during development.

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

History

0.1.0 (2023-09-13)

• First release on PyPI.

0.1.1 (2023-10-27)

• Basic logging functionality.

• Only supports logging to console.

0.2.0 (2023-11-2)

• Implemented Annalist as a Singleton.

• Usage now includes configuration step.

0.3.0 (2023-11-20)

• Now takes arbitrary input paramaters.

• Able to support Hilltop audit trail parity.

• User can control logging levels

0.3.3 (2023-11-24)

• I’m not sure what happened to 0.3.1 and 0.3.2

• Now REALLLY able to support Hilltop audit trail parity.

• Improved support for class method logging