The Kieker language pack for Python is in an early stage. Thus, it may not operate as expected in all scenarios. We applied it to the Spyder IDE and other programs successfully and to parts of Tensorflow with limited results.
Prerequisites:
- python >=3.9
- build package (pip install build)
- clone this repository
- cd ./kieker-lang-pack-python
- run python3 -m build
- pip install dist/kieker-monitoring-for-python-0.0.1.tar.gz
This is the most basic and invasive approach. Currently, it also offers the most control over the instrumentation. Let's assume that you want to log following function every time it is called.
example.py
def some_function():
print('Hello World!')
example.py
import tools.const as con
def some_function():
print('Hello World!')
Now we have everything we need to log the function execution. At the current stage of development there are three record types with following entries.
BeforeOperationEvent
: timestamp in ms, traceId, traceOrder, function_name, fully qualified name of caller classAfterOperationEvent
: timestamp in ms, traceId, traceOrder, function_name, fully qualified name of caller classAfterOPerationFailedEvent
: timestamp in ms, traceId, traceOrder, function_name, fully qualified name of caller class, error name
Create a record object and pass it to the monitoring controler via new_monitoring_record()
. In our example we use BeforeExecutionRecord
and AfterExecutionRecord
.
example.py
import const as con
from monitoring.record.trace.operation.operationevent import (BeforeOperationEvent,
AfterOperationEvent,
AfterOperationFailedEvent,
)
ctrl = SingleMonitoringController('/path/to/config.ini')
def some_function():
trace_reg = con.trace_reg
trace = trace_reg.get_trace()
if(trace is None):
trace = trace_reg.register_trace()
monitoring_controller.new_monitoring_record(trace)
trace_id = trace.trace_id
timestamp = ctrl.time_source_controller.get_time()
before_record = BeforeOperationevent(timestamp, trace_id, trace.get_next_order_id(), 'some_function','example.some_function')
ctrl.new_monitoring_record(before_record)
print('Hello World!')
timestamp = ctrl.time_source_controller.get_time()
after_record = AfterOperationEvent(timestamp, trace_id, trace.get_next_order_id(), 'some_function','example.some_function')
ctrl.new_monitoring_record(after_record)
The above method gives us full control over program instrumentation, but it also requieres a lot of manual work. But this approach alters your own code logic, which might not be a problem for small projects, but for larger code basis it becomes a tedious job to keep an overview. To minimize the grade of source code manipulation you can use an Instrument
decorator. By anotation a function or method with @Instrument
you can save a lot of typing. Consider the following example.
example.py
from tools.aspect import instrument
from monitoring.controller import SingleMonitoringController
monitoring_controller = SingleMonitoringController('/path/to/config.ini') # Always instatiate a controller
@instrument
def some_function():
print('Hello World!')
We also provide two import Hooks. They are used in similar manner. All you have to do is to add either PostImportFinder or InstrumentOnImportFinder to the sys.meta_path in the module which is a an entry point for your python program. Bellow we show all the necessary API to do that.
from tools.importhook import PostImportFinder
from monitoring.controller import SingleMonitoringController
pattern_object = re.compile('YOUR REGEX') # Some regex to describe which modules should be instrumented
exclude_modules = list() # List with explicit strings of modules which should be excluded from the instrumentations
some_var = SingleMonitoringController('/path/to/config.ini')
sys.meta_path.insert(0, PostImportFinder(pattern_object, exclude_modules))
import re
from tools.aspect import decorate_find_spec
from tools.importhookast import InstrumentOnImportFinder
ex=list()
sys.meta_path.insert(0, InstrumentOnImportFinder(ignore_list=ex, empty=False, debug_on=True))