mona-client 0.1.15

Client code for python Mona instrumentation

Python Client

Installation

$ pip install mona-client

Environment variables

Mona uses several environment variables you need to set in order for things to work as intended:

MONA_HOST - the Mona server's IP address you get from Mona. MONA_PORT - The port for the above address. MONA_USER_ID - A unique user id we provide for you.

Quick Start and Example

• Install as mentioned above
• Set environment variables as mentioned above
• Instrument code with client as seen below

from mona_client import client

# Only call once,
# Add parameteres (host, port, ...) on this call if not already set through env vars.
client.init_client()

...

with client.new_mona_context(ARC_class="MY_COOL_ALGORITHM_NAME"):
    ...
    client.export({"foo": "bar"})
    ...

ARCs, sub-ARCs, KAPIs and Dimensions

All the information you send Mona is sent using the "export()" function. You can send any json-serializable dict via this function. All the fields sent via export are either considered (raw) KAPIs or (raw) dimensions.

KAPI stands for Key Algorithmic Performance Indicator. These are all data relevant for how your algorithm is performing. Examples could be a sub-component's confidence score, or the amount of results received from a specific algorithm.

Dimensions are all metadata relevant for segmenting KAPI values. For example, end-user demographics, a customer/user-id or a data source id.

We call the exported data "raw", as sometimes we don't use this data directly, but generate actual KAPIs/dimensions from it (e.g., when looking at an average value over time as a KAPI).

As you export data to Mona throughout your code, Mona aggregates it on its servers to a struct called an ARC. ARC stands for Algorithm Run Context. Each ARC describes an entire algorithm run. You can create as many ARCs as you'd like in your code, but in general, a single run of your code should correspond to a single ARC. In order to tell Mona all exports should belong to the same ARC, you must init a new context. This is usually done in one place in your code, where a new run is started, using:

with client.new_mona_context(arc_class="MY_COOL_ALGORITHM_NAME"):

You assign "classes" to contexts. This tells Mona which ARCs describe runs of the same algorithm, thus allowing Mona to compare them, segment them and find relevant insights.

If you have logically separable parts to your algorithm, you can init sub-contexts using:

with client.new_mona_sub_context(arc_class="MY_COOL_SUB_COMPONENT_NAME"):

Override Export Timestamp

When using the export function, an "export timestamp" is automatically added to the exported message. In some cases, there's a need to override this value with a given timestamp (For example, when exporting historical data in batch). When exporting with a given timestamp, use the export function and send the export timestamp (in seconds) as a parameter. For example:

client.export({"foo": "bar"}, 1552903200.0)

Concurrency

Mona saves the ARC's id as a special variable, which is local to the thread and to greenlets. This means that if you start a new thread/greenlet, by default the new thread will have an empty context.

This is usually the preffered behavior, as a new thread usually means a new received request (on servers) or a completely new run of an algorithm.

If by any chance you'd like to continue with the same context on a new thread, just use the child class MonaThread under mona_thread.py. This class takes care of transferring the full context id to the newly started thread.

from mona_client import client
from mona_client.mona_thread import MonaThread

def threaded_function():
    print(client.get_full_context_id() == main_context)

with client.new_mona_context(arc_class="threaded_algorithm_name"):
    global main_context
    main_context = client.get_full_context_id()
    MonaThread(target=threaded_function).start()  # Prints "True"

Testing the client code

The client's tests are written using the pytest framework, so in order to run the tests (assuming you have pytest and pytest-mock installed on your environment), you just need to type "pytest" to your shell.

Uploading new version to PyPI

The main reference to follow to do that is on: https://packaging.python.org/tutorials/packaging-projects/

• Register on PyPI with your mona email: https://pypi.org/
• Ask [e-mail] to add you as collaborator
• If not installed, install twine: $ python3 -m pip install --user --upgrade twine
• Change version number under setup.py
• If a new dependency is required, add it under setup.py under "install_requires"
• If not installed, install build tools: $ python3 -m pip install --user --upgrade setuptools wheel
• Build new version: $ python3 setup.py sdist bdist_wheel
• Upload new version (can change '*' to actual version): $ python -m twine upload dist/*

Configuration and Big Red Button

If you'd like to set up configuration for mona, you can fill a simple json configuration file. See mona_client_config.json for the default configuration file used. There is currently only one configuration option, which is the "Big Red Button" - the "disable_all" configuration, which, when set to true, completely disables all mona activity (no more exporting and context inits).

Mona listens to changes on the configuartion file under the environment variable "MONA_CLIENT_CONFIG_FILE". So if you want to disable mona while running, just change "disable_all" from "false" to "true" in your config file and mona will update automatically.

If you don't set MONA_CLIENT_CONFIG_FILE yourself, Mona will use a default configuration file located on mona_client/config/mona_client_config.json. If logging is enabled, Mona will log out the full address of this file on startup. You can then make changes to that file to update configuration while mona is running.

Another option to quickly disable all Mona activity, is to set the MONA_DISABLE_ALL environment variable to a truthy value.

Logging

Mona is using it's own logger named "mona-logger". You can configure it in your code by just calling ''' logging.getLogger("mona-logger") ''' and then setting handlers and formatters as you please.

You can also configure Mona's logging using two different environment variables:

1. MONA_LOGGING_LEVEL - set this to a level according to python's logging constants (e.g., 10 = debug)
2. MONA_PRINT_LOGS - set this to true if you want Mona to print logs to stdout.

Special field names

Don't use field names with "MONA_" as their prefix. Mona uses this pattern internally. If you do that, these fields will be discarded before being emitted to Mona.