Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[MRG] Add multi-level security awareness to event formatter #30

Closed
wants to merge 8 commits into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions docs/index.rst
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@ Jupyter Telemetry

Telemetry provides a configurable traitlets object, EventLog, for structured event-logging in Python. It leverages Python's standard logging library for filtering, handling, and recording events. All events are validated (using jsonschema) against registered JSON schemas.

If you're looking for telemetry in Jupyter frontend applications (like JupyterLab), checkout the work happening in jupyterlab-telemetry_!
If you're looking for telemetry in Jupyter frontend applications (like JupyterLab), checkout the work happening in jupyterlab-telemetry_!

.. _jupyterlab-telemetry: https://github.com/jupyterlab/jupyterlab-telemetry

Expand Down Expand Up @@ -98,8 +98,8 @@ Schemas can be registered from a Python dict object, a file, or a URL. This exam
:maxdepth: 2
:caption: Table of Contents:

pages/schemas.rst

pages/schemas
pages/sensitive-data

Indices and tables
------------------
Expand Down
20 changes: 15 additions & 5 deletions docs/pages/schemas.rst
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@ Writing a Schema
================


Schemas should follow valid `JSON schema`_. These schemas can be written in valid YAML or JSON.
Schemas should follow valid `JSON schema`_. These schemas can be written in valid YAML or JSON.

At a minimum, valid schemas should have the following keys:

Expand All @@ -16,7 +16,13 @@ At a minimum, valid schemas should have the following keys:

+ ``title`` : name of the property
+ ``description``: documentation for this property.
+ ``pii``: (optional) boolean for whether this property is personally identifiable information or not.
+ ``level``: the level of sensitivity of this property.

Jupyter Telemetry provides three levels of sensitivity. The list of sensitivity level in increasing order:

+ ``'unrestricted'``
+ ``'user-identifier'``
+ ``'user-identifiable-information'``

- ``required``: list of required properties.

Expand All @@ -33,11 +39,15 @@ Here is a minimal example of a valid JSON schema for an event.
properties:
name:
title: Name
description: |
Name of event
type: string
level: unrestricted
description: Name of event
user:
title: User name
level: user-identifier
description: Name of user who initiated event
required:
- name
- user


.. _JSON schema: https://json-schema.org/
42 changes: 42 additions & 0 deletions docs/pages/sensitive-data.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
Logging sensitive data
======================

Jupyter Telemetry requires that all fields in an event schema explicitly state when personally-identifiable information is being collected. Each property in an event schema must have a ``sensitivity_level`` attribute.

Since events may include data with varying degrees of sensitivity, Jupyter Telemetry uses a multi-level security approach. It exposed three levels of sensitivity:

+ ``'unrestricted'``
+ ``'personal-identifier'``
+ ``'personally-identifiable-information'``

Each event property can be given one of these three levels:

.. code-block:: yaml

$id: example.schema
...
sensitivity_level:
properties:
name:
title: Name
sensitivity_level: unrestricted
description: Name of event
type: string
user:
title: User name
sensitivity_level: personal-identifier
description: Name of the user who initiated the event.
affiliation:
title: I

Jupyter Telemetry uses ``sensitivity_level`` to drop sensitive data when emitting events. By default, properties greater than "unclassifed" are dropped from recorded event data.

Each logging handler increase the level of sensitive data it emots. This can be configured by changing its ``.event_level`` attribute.
Zsailer marked this conversation as resolved.
Show resolved Hide resolved

.. code-block:: python

import logging

handler = logging.FileHandler('events.log')
handler.event_level = 'secret'

16 changes: 10 additions & 6 deletions jupyter_telemetry/eventlog.py
Original file line number Diff line number Diff line change
Expand Up @@ -6,13 +6,13 @@
from datetime import datetime

import jsonschema
from pythonjsonlogger import jsonlogger
from ruamel.yaml import YAML
from traitlets import List
from traitlets.config import Configurable, Config

from .traits import Handlers
from . import TELEMETRY_METADATA_VERSION
from .formatter import JsonEventFormatter

yaml = YAML(typ='safe')

Expand Down Expand Up @@ -64,19 +64,23 @@ def __init__(self, *args, **kwargs):
self.log.propagate = False
# We will use log.info to emit
self.log.setLevel(logging.INFO)
self.schemas = {}

if self.handlers:
formatter = jsonlogger.JsonFormatter(json_serializer=_skip_message)
for handler in self.handlers:
# Create a formatter for this handler.
formatter = JsonEventFormatter(
logger=self,
handler=handler,
json_serializer=_skip_message
)
# Set formatted for handler.
handler.setFormatter(formatter)
self.log.addHandler(handler)

self.schemas = {}

def _load_config(self, cfg, section_names=None, traits=None):
"""Load EventLog traits from a Config object, patching the
handlers trait in the Config object to avoid deepcopy errors.

"""
my_cfg = self._find_my_config(cfg)
handlers = my_cfg.pop("handlers", [])
Expand Down Expand Up @@ -112,7 +116,7 @@ def register_schema(self, schema):
jsonschema.validators.validator_for(schema).check_schema(schema)

# Check that the properties we require are present
required_schema_fields = {'$id', 'version'}
required_schema_fields = {'$id', 'version', 'properties'}
for rsf in required_schema_fields:
if rsf not in schema:
raise ValueError(
Expand Down
58 changes: 58 additions & 0 deletions jupyter_telemetry/formatter.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,58 @@
from traitlets import HasTraits, validate, Set
from pythonjsonlogger import jsonlogger


class JsonEventFormatter(jsonlogger.JsonFormatter):
"""Patch the jsonlogger formatter to include levels for telemetry.

Properties in a logged event that has a level less than
the handler's event_level will be dropped from the emitted event.
"""
def __init__(self, logger, handler, *args, **kwargs):
self.logger = logger
self.handler = handler
super(JsonEventFormatter, self).__init__(*args, **kwargs)

@property
def allowed_tags(self):
return getattr(self.handler, 'allowed_tags', {})

@property
def hashed_tags(self):
return getattr(self.handler, 'hashed_tags', {})

def process_log_record(self, log_record):
log_record = super(JsonEventFormatter, self).process_log_record(log_record)
return self.process_tags(log_record)

def drop_property(self, key, record):
del record[key]
return record

def hash_property(self, key, record):
hash_function = lambda x: x
record[key] = hash_function(record[key])
return record

def process_tags(self, log_record):
"""
"""
# Registered schemas are identified by their name and version.
key = (log_record['__schema__'], log_record['__schema_version__'])
schema = self.logger.schemas[key]['properties']
props = [key for key in log_record.keys()
if not key.startswith('__') and key != 'message']

# Walk through the recorded event and handle each key
# based on its tag/category.
for key in props:
tag = schema[key]['tag']
if tag in self.allowed_tags or tag == "unrestricted":
# If the tag is found in the allowed_tags trait, do nothing.
if tag in self.hashed_tags:
log_record = self.hash_property(key, log_record)
# Drop tags not listed in allowed_tags
else:
log_record = self.drop_property(key, log_record)

return log_record
3 changes: 1 addition & 2 deletions tests/test_eventlog.py
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,6 @@ def test_good_config_file(tmp_path):
# Pass config to EventLog
e = EventLog(config=cfg)

# Assert the
assert len(e.handlers) > 0
assert isinstance(e.handlers[0], logging.Handler)

Expand All @@ -49,4 +48,4 @@ def test_bad_config_file(tmp_path):
cfg = get_config_from_file(tmp_path, BAD_CONFIG)

with pytest.raises(TraitError):
e = EventLog(config=cfg)
EventLog(config=cfg)
18 changes: 12 additions & 6 deletions tests/test_register_schema.py
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,8 @@ def test_reserved_properties():
'version': 1,
'properties': {
'__fail__': {
'type': 'string'
'type': 'string',
'level': 'unclassified'
},
},
})
Expand Down Expand Up @@ -100,7 +101,8 @@ def test_record_event():
'version': 1,
'properties': {
'something': {
'type': 'string'
'type': 'string',
'level': 'unclassified'
},
},
}
Expand Down Expand Up @@ -138,7 +140,8 @@ def test_register_schema_file():
'version': 1,
'properties': {
'something': {
'type': 'string'
'type': 'string',
'level': 'unclassified'
},
},
}
Expand Down Expand Up @@ -166,7 +169,8 @@ def test_allowed_schemas():
'version': 1,
'properties': {
'something': {
'type': 'string'
'type': 'string',
'level': 'unclassified'
},
},
}
Expand Down Expand Up @@ -194,10 +198,12 @@ def test_record_event_badschema():
'version': 1,
'properties': {
'something': {
'type': 'string'
'type': 'string',
'level': 'unclassified'
},
'status': {
'enum': ['success', 'failure']
'enum': ['success', 'failure'],
'level': 'unclassified'
}
}
}
Expand Down
79 changes: 79 additions & 0 deletions tests/test_tags.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,79 @@
import io
import json
import logging

from jupyter_telemetry.eventlog import EventLog

import pytest


@pytest.fixture
def version(): return 1


@pytest.fixture()
def schema_id(): return 'test.event'


@pytest.fixture
def schema(schema_id, version):
return {
'$id': schema_id,
'title': 'Test Event',
'version': version,
'description': 'Test Event.',
'type': 'object',
'properties': {
'nothing-exciting': {
'description': 'a property with nothing exciting happening',
'tag': 'unrestricted',
'type': 'string'
},
'id': {
'description': 'user ID',
'tag': 'user-identifier',
'type': 'string'
},
'email': {
'description': 'email address',
'tag': 'user-identifiable-information',
'type': 'string'
},
}
}


@pytest.mark.parametrize(
'tags,expected_props',
[
({'unrestricted'}, {'nothing-exciting'}),
({'user-identifier'}, {'nothing-exciting', 'id'}),
({'user-identifiable-information'}, {'nothing-exciting', 'email'})
]
)
def test_properties_tags(schema, schema_id, version, tags, expected_props):
sink = io.StringIO()

# Create a handler that captures+records events with allowed tags.
handler = logging.StreamHandler(sink)
handler.allowed_tags = tags

e = EventLog(
handlers=[handler],
allowed_schemas=[schema_id]
)
e.register_schema(schema)

event = {
'nothing-exciting': 'hello, world',
'id': 'test id',
'email': 'test@testemail.com',
}

# Record event and read output
e.record_event(schema_id, version, event)
recorded_event = json.loads(sink.getvalue())
recorded_props = set([key for key in recorded_event if not key.startswith('__')])

assert expected_props == recorded_props