BptkServer

BPTK API Documentation for the BptkServer class
Keywords

system dynamics, agent-based modeling, flask, REST, bptk, bptk-py, python, business prototyping

BptkServer

The REST API server is essentially a wrapper around the regular bptkclass. The server is implemented using Flask and provides all the REST Endpoints / Routes that you need to interact with a simulation scenario.

Overview

The BptkServer class provides a REST-API using the Flask framework.

It is essentially a wrapper around the bptk class that forwards REST API calls to bptk.

You will typically start the framework by instantiating the bptk class within a Jupyer notebook, as follows:

from BPTK_Py.server import BptkServer

from model import bptk # assuming your model is in a file called model.py that sets up bptk

# Calling the BptkServer class
application = BptkServer(__name__, bptk)

if __name__ == "__main__":
   application.run()

Assuming you save that code in a file called application.py, you can then start the server fro the command line as follows:

export FLASK_ENV=development

BptkServer Constructor

BptkServer(import_name, bptk_factory=None, external_state_adapter=None, bearer_token=None, externalize_state_completely=False)

This class provides a Flask-based server that provides a REST-API for running bptk scenarios. The class inherts the properties and methods of Flask and doesn’t expose any further public methods.

  • Parameters

    • import_name – String. Flask’s own first argument; pass __name__.

    • bptk_factory – Callable. Called once per instance to build a bptk object with the scenario managers this server should serve. It is a factory rather than an instance because every simulation instance needs its own: two clients must not share one model’s state. This is also where a server-wide execution engine is set - see Execution Backends.

    • external_state_adapter – ExternalStateAdapter. Where instance state is kept when it is not in this process’s memory. Without one, state lives and dies with the process.

    • bearer_token – String. When set, every request must carry it as Authorization: Bearer ….

    • externalize_state_completely – Boolean (Default=False). Makes the server stateless. With False an instance stays in memory and is written to the adapter, so the adapter is a safety net. With True the instance is deleted from memory after every use and read back from the adapter on the next request, so this process holds nothing between requests - which is what lets several server processes share one workload behind a load balancer, and what lets any of them restart without losing a session. It needs an external_state_adapter; without one there is nowhere to externalise to.

agents

POST /agents

For an agent-based or hybrid model, this endpoint returns all the agents in the model with their corresponding states and properties.

begin-session

POST /{instance_uuid}/begin-session

This endpoint starts a session for single step simulation. There can only be one session per instance at a time.

Besides the scenarios to run, the body takes two optional fields:

  • backend – "python" or "rust", and it decides the engine for the whole session: a session keeps the engine it started on for its lifetime. Omit it and the server’s own default applies. An unknown value is rejected with 400 rather than quietly falling back, so a typo is visible.
  • seed – pins the Rust engine’s random number generator. It matters for a stochastic session that has to survive a restart: with a seed the resumed session replays identically, without one the numbers after the restart are different ones. The Python engine ignores it.
{
  "scenario_managers": ["smSirModel"],
  "scenarios": ["base"],
  "equations": ["infected"],
  "backend": "rust",
  "seed": 42
}

end-session

POST /{instance_uuid}/end-session

This endpoint ends a session for single step simulation and resets the internal cache.

execute

POST /execute

Runs a model that arrives in the request body, rather than one registered on the server. Where /run picks from the scenario managers the server was started with, /execute is fully self-contained: it touches neither the server’s own bptk instance, nor the instance manager, nor the external state adapter. Its first consumer is the visual modeler, where a model is designed in a browser and posted here to be run.

The body carries the model in the same JSON schema that Model.to_json() produces, a scenarios dict of overrides and the equations to return:

{
  "model": { "...": "JSON model, same schema as Model.to_json()" },
  "scenarios": {
    "baseline":     { "constants": { "transmission_prob": 0.001 } },
    "high_contact": { "constants": { "transmission_prob": 0.01 } }
  },
  "equations": ["susceptible", "infected", "recovered"]
}

scenarios is optional — without it the model runs once under the name default. Each scenario gets a fresh model, so overrides do not leak between them.

The response is {scenario_name: {equation: {t: value}}}.

This endpoint always runs on the Rust engine. There is no Python alternative for a raw JSON model: the Python engine needs the element graph, not the engine’s flat node format. A server without the compiled engine answers 500; a malformed body, a missing model or equations, or a model the engine cannot load answers 400 naming the field or the problem.

equations

POST /equations

This endpoint returns all available equations given the name of a scenario manager and of a scenario.

flat-session-results

GET /{instance_uuid}/flat-session-results

Returns the accumulated results of a session, from the first step to the last step that was run in a flat format.

full-metrics

GET /full-metrics

Returns metrics in JSON format. This endpoint is unprotected.

The following metrics are returned:

  • Instance count
  • Creation time und current timestep of each instance

healthy

GET /healthy

Unprotected endpoint useful for health checks, e.g. when running as a Kubernetes pod. It simply returns a HTTP 200 response and doesn’t interact with the bptk factory.

keep-alive

POST /{instance_uuid}/keep-alive This endpoint sets the “last accessed time” of the instance to the current time to prevent the instance from timeing out.

Arguments: None

load-state

GET /load-state

Loads all instances using the external state adapter

root

GET /

The root endpoint returns a simple html page that can be used for test purposes. The root endpoint is unprotected even if bearer token authorization is turned on.

run-step

POST /{instance_uuid}/run-step

This endpoint advances the relevant scenarios by one timestep and returns the data for that timestep.

Arguments:

instance_uuid: string

    The id of the instance to advance.

run-steps

POST /{instance_uuid}/run-steps

This endpoint advances the relevant scenarios by one timestep and returns the data for that timestep.

Arguments:

instance_uuid: string

    The id of the instance to advance.

session-results

GET /{instance_uuid}/session-results

Returns the accumulated results of a session, from the first step to the last step that was run.

metrics

GET /metrics

Returns metrics in a Prometheus compatible format. This endpoint is unprotected.

run

POST /run

Given a JSON dictionary that defines the relevant simulation scenarios and equations, this endpoint runs those scenarios and returns the data generated by the simulations.

save-state

GET /save-state

Save all instances with the provided external state adapter.

scenarios

GET /scenarios

The endpoint returns all available scenarios for the current simulation.

start-instance

POST /start-instance

This endpoint starts a new instance of BPTK on the server side, so that simulations can run in a “private” session. The endpoint returns an instance_id, which is needed to identify the instance in later calls.

  • Arguments:

    • timeout (dict,optional). The timeout period after which the instance is delete if it is not accessed in the meantime. The timer is reset every time the instance is accessed. The timeout dictionary can have the following keys: weeks, days, hours, minutes, seconds, milliseconds, microseconds. Values must be integers.

start-instances

POST /start-instances

This endpoint starts a number of new instances of BPTK on the server side, so that simulations can run in “private” sessions. The endpoint returns a list of instance_id’s, which are needed to identify the instances in later calls.

  • Arguments:

    • timeout (dict,optional). The timeout period after which the instance is delete if it is not accessed in the meantime. The timer is reset every time the instance is accessed. The timeout dictionary can have the following keys: weeks, days, hours, minutes, seconds, milliseconds, microseconds. Values must be integers.
    • instances (Integer). Number of instances to create

  • Response:

    • instance_uuids. List of instance UUIDs.
    • timeout:_ The timeouts the instances where started with.

stop-instance

POST /{instance_uuid}/stop-instance

Deletes the instance and, when the server runs with an external state adapter, its stored state as well. Use it to release an instance you are done with rather than waiting for it to time out.

stream-steps

POST /{instance_uuid}/stream-steps

This endpoint is used to stream a simulation. This is useful for long-running simulations, the result of each simulation step is streamed accross the API:

Arguments:

instance_uuid: string

The id of the instance to stream.

Usage

The following illustrates how to use the BPTK REST API. To get started you first need to start the server for the customer acquisition model from a Terminal console by running the run_server.sh script in this directory. The server should then be running on port 5000.

# find documentation for the requests library on https://docs.python-requests.org/
import requests
import json

List scenarios

Get a list of scenarios that the server knows about:

response = requests.get("http://localhost:5000/scenarios")
 
response
 
response.json()

List equations

Get the equations known to the dashboard scenario:

response_1 = requests.post(
    url='http://localhost:5000/equations',
    json={'scenarioManager': 'sddsl_customer_acquisition', 'scenario': 'interactive_scenario'},
)

Check the response

response_1

Read the response content

response_1.json()

Run some scenarios

Get some data for a scenario and some equations:

response_2 = requests.post(
    url='http://localhost:5000/run',
    json={
        'scenario_managers': ['sddsl_customer_acquisition'],
        'scenarios': ['base', 'low_word_of_mouth', 'high_word_of_mouth', 'interactive_scenario'],
        'equations': ['customers', 'customer_acquisition', 'market_saturation'],
        'settings': {'sddsl_customer_acquisition': {'interactive_scenario': {'constants': {'word_of_mouth_success': 0.5}}}},
    },
)
 
response_2
 
response_2.json()['sddsl_customer_acquisition']['base']['equations']['market_saturation']
 
response_3 = requests.post(url='http://localhost:5000/start-instance')
 
response_3
 
instance_uuid = response_3.json()['instance_uuid']
 
instance_uuid
 
response_4 = requests.post(
    url=f'http://localhost:5000/{instance_uuid}/begin-session',
    json={
        'scenario_managers': ['sddsl_customer_acquisition'],
        'scenarios': ['interactive_scenario'],
        'equations': ['customers', 'word_of_mouth_success'],
    },
)
 
response_4
 
response_4.json()
 
response_5 = requests.post(url=f'http://localhost:5000/{instance_uuid}/run-step')
 
response_5
 
response_5.json()
 
response_6 = requests.post(
    url=f'http://localhost:5000/{instance_uuid}/run-step',
    json={'settings': {'sddsl_customer_acquisition': {'interactive_scenario': {'constants': {'word_of_mouth_success': 0.7}}}}},
)
 
response_6.json()

You can start as many instances as you like - typically one for every user that starts an interactive session via a front-end. Let’s sart another session to illustrate this.

response_7 = requests.post(url='http://localhost:5000/start-instance')
 
another_instance_uuid = response_7.json()['instance_uuid']
 
another_instance_uuid
 
response_8 = requests.post(
    url=f'http://localhost:5000/{another_instance_uuid}/begin-session',
    json={
        'scenario_managers': ['sddsl_customer_acquisition'],
        'scenarios': ['interactive_scenario'],
        'equations': ['customers', 'word_of_mouth_success'],
    },
)
 
response_9 = requests.post(
    url=f'http://localhost:5000/{another_instance_uuid}/run-step',
    json={'settings': {'sddsl_customer_acquisition': {'interactive_scenario': {'constants': {'word_of_mouth_success': 0.1}}}}},
)
 
response_9.json()
 
response_10 = requests.post(
    url=f'http://localhost:5000/{instance_uuid}/run-step',
    json={'settings': {'sddsl_customer_acquisition': {'interactive_scenario': {'constants': {'word_of_mouth_success': 0.07}}}}},
)
 
response_10.json()