Monitoring Your Service¶

Monitoring is a very important aspect of software deployment. Deployed algorithms often deteriorate over time, in a process called domain shift, and being able to monitor your deployed algorithms can make all the difference in detecting this in the early stages.

Monitoring Variables¶

To monitor the value of a variable in the code you can use the store() function. It takes an arbitrary number of keyword arguments and saves their values in a time-series database, where they can be retrieved at a later time. store() supports integer, float and string data. Let’s use the hello service as an example:

import logging
from mvi.mvi import MviService, store

mvi = MviService()
logger = logging.getLogger(__name__)

mvi.add_parameter("greeting_phrase", "Hello")

@mvi.entrypoint
def hello(name: str) -> str:
    greeting_phrase = mvi.get_parameter("greeting_phrase")
    logger.info(f"Greeting someone with the name: {name}")

    # Now the name and greeting will be stored everytime this function is called
    store(name=name, greeting=greeting_phrase)

    return f"{greeting_phrase} {name}"

Monitoring an Entrypoint¶

It is also possible to automatically monitor the input and output of a service entrypoint. To do this we pass monitor=True to the entrypoint() decorator:

import logging
from mvi.mvi import MviService, store

mvi = MviService()
logger = logging.getLogger(__name__)

mvi.add_parameter("greeting_phrase", "Hello")

# Now the input and output of this entrypoint is stored after each call
@mvi.entrypoint(monitor=True)
def hello(name: str) -> str:
    greeting_phrase = mvi.get_parameter("greeting_phrase")
    logger.info(f"Greeting someone with the name: {name}")
    return f"{greeting_phrase} {name}"

In this case, the data is saved as json strings, in order to support more data formats.

Monitoring a Parameter¶

Parameters set with add_parameter() can also be monitored by passing monitor=True, which also supports integer, float and string data:

import logging
from mvi.mvi import MviService, store

mvi = MviService()
logger = logging.getLogger(__name__)

# Now the value of ``"greeting_phrase"`` is stored every time it's updated
mvi.add_parameter("greeting_phrase", "Hello", monitor=True)

@mvi.entrypoint
def hello(name: str) -> str:
    greeting_phrase = mvi.get_parameter("greeting_phrase")
    logger.info(f"Greeting someone with the name: {name}")
    return f"{greeting_phrase} {name}"

Getting the Data¶

The MultiViz Integrator provides three options for accessing the time-series data for your monitored variables.

Option 1: Json format

The time-series data in json format can be collected via the following entrypoint:

http://<host>/services/<servce_name>_<service_version>/~monitor

It is possible to specify start time and end time for the wanted time-series data by adding end and start query parameters to the url, like this:

http://<host>/services/<servce>_<service_version>/~monitor?end=<...>?start=<...>

The end and start query parameters needs to have the following format: YYYY-MM-DD[T]HH:MM[:SS[.ffffff]][Z or [±]HH[:]MM]]], so for instance: 2020-01-01 02:30

It it also possible to specify which variables to query time-series by specify the query parameter variables, like this:

http://<host>/services/<servce_name>_<service_version>/~monitor?end=<...>?start=<...>?variables=v1?variables=v2

This will query time-series data for variable a and b. The returned json time-series data will be on the following format:

{
    "a": {
        'timestamp': [t1,t2,...,tn]
        'value': [v1,v2,...,vn]
        }
    "b": {
        'timestamp': [t1,t2,...,tn]
        'value': [v1,v2,...,vn]
        }
}

Example, using the requests package in python:

response = requests.GET(
    "services/name_version/~monitor",
    headers={"Authorization": f"Bearer {TOKEN}"})
data = response.json()

Option 2: CSV files

The time-series data can also be returned in csv format, the entrypoint for this is:

http://<host>/services/<servce_name>_<service_version>/~csv

It is possible to specify, which variables and in during which time interval to query time-series data for by using the following query parameters: end, start and varaibles. For instance:

http://<host>/services/<servce_name>_<service_version>/~monitor/csv?end=<...>?start=<...>?variables=v1?variables=v2

The end and start query parameters needs to have the following format: YYYY-MM-DD[T]HH:MM[:SS[.ffffff]][Z or [±]HH[:]MM]]], so for instance: 2020-01-01 02:30

This will query time-series data for variable a and b.This entrypoint returns a zip file containing one csv file per requested variable. The csv files has the following format:

timestamp	value
t1	v1
t2	v2

Option 3: The whole service database

The time-series data is stored a in sqlite databases and the whole service database can be requested at the following entrypoint:

http://<host>/services/<servce_name>_<service_version>/~monitor/db