Remote BMI

The Basic Model Interface (BMI) is a standard interface for models. The interface is available in different languages and a language agnosting version in SIDL.

To have a consumer of the model and the provider of the model seperated you can use grpc4bmi, but this only on languages that have a grpc implementation. This repo replaced the gRPC protocol with an REST API. The REST API specification is in the OpenAPI format.

Usage

Python consumer

Installation

pip install remotebmi

A client can connect to a running server with the following code.

from remotebmi.client import BmiClient

model = RemoteBmiClient('http://localhost:50051')
# Now you can use the BMI methods on model
# for example
model.initialize('config.file')
model.update()
model.get_value('var_name')

A client can also start a Apptainer container containing the model and the server.

from remotebmi.client import BmiClientApptainer

model = BmiClientApptainer('my_model.sif', work_dir='/tmp')

The client picks a random port and expects the container to run the BMI web server on that port. The port is passed to the container using the BMI_PORT environment variable.

A client can also start a Docker container containing the model and the server.

from remotebmi.client import BmiClientDocker

model = BmiClientDocker('ewatercycle/wflowjl:0.7.3', work_dir='/tmp')

The BMI web server inside the Docker container should be running on port 50051. If the port is different, you can pass the port as the image_port argument to the BmiClientDocker constructor.

Python provider

Given you have a model class called MyModel in a package mypackage then the web service can be started with the following command.

BMI_MODULE=mypackage BMI_CLASS=MyModel run-bmi-server 

Julia provider

Given you have a model class called MyModel and a BMI called BMI inside the MyPackage package.

using MyPackage.MyModel
using RemoteBMI

port = parse(Int, get(ENV, "BMI_PORT", 50051))
RemoteBMI.run(MyModel, port=port, host="localhost")

R provider

Given you have a model called ModelBmi that has a BMI inside a MyModel R library.

library(remotebmi)
library(MyModel)

port = as.integer(Sys.getenv("BMI_PORT", 50051))
server <- RemoteBmiServer(MyModel$ModelBmi, port=port, host="localhost")
server$run()

Other languages

Any language that can run a HTTP server and parse/load JSON can be used as a provider.

The server should implement the openapi.yaml specification.