grpc4bmi

Purpose

This software allows you to wrap your Basic Model Interface (BMI) implementation in a server process and communicate with it via the included Python client. The communication is serialized to protocol buffers by GRPC and occurs over network ports. Can run models in isolated containers using Docker or Apptainer.

Installation

Optionally, create your virtual environment and activate it, Then, run

pip install grpc4bmi

on the client (Python) side. If your server model is implemented in Python, do the same in the server environment (e.g. docker container). If the model is implemented in R, run instead

pip install grpc4bmi[R]

If the model is implemented in Julia, run instead

pip install grpc4bmi[julia]

in the server environment. For bleeding edge version from GitHub use

pip install git+https://github.com/eWaterCycle/grpc4bmi.git#egg=grpc4bmi

Finally if the model is implemented in C or C++, clone this git repo and run

make
make install

in the cpp folder.

Usage

Model written in Python

A model should be a subclass of the Bmi class from the bmipy package.

For inspiration look at the example in the test directory.

To start a server process that allows calls to your BMI implementation, type

run-bmi-server --name <PACKAGE>.<MODULE>.<CLASS> --port <PORT> --path <PATH>

where <PACKAGE>, <MODULE> are the python package and module containing your implementation, <CLASS> is your bmi model class name, <PORT> is any available port on the host system, and optionally <PATH> denotes an additional path that should be added to the system path to make your implementation work. The name option above is optional, and if not provided the script will look at the environment variables BMI_PACKAGE, BMI_MODULE and BMI_CLASS. Similarly, the port can be defined by the environment variable BMI_PORT. This software assumes that your implementation constructor has no parameters.

Model written in C/C++ (beta)

Create an executable along the lines of cpp/run-bmi-server.cc. You can copy the file and replace the function

Bmi* create_model_instance()
{
    /* Return your new BMI instance pointer here... */
}

with the instantiation of your model BMI. The model needs to implement the csdms BMI for C, but you may also implement our more object-oriented C++ interface BmiCppExtension.

Model written in R

The grpc4bmi Python package can also run BMI models written in R if the model is a subclass of AbstractBmi See https://github.com/eWaterCycle/bmi-r for instruction on R and Docker.

Run the R model a server with

run-bmi-server --lang R [--path <R file with BMI model>] --name [<PACKAGE>::]<CLASS> --port <PORT>

For example with WALRUS use

run-bmi-server --lang R --path ~/git/eWaterCycle/grpc4bmi-examples/walrus/walrus-bmi.r --name WalrusBmi --port 55555

Models written in Julia

from grpc4bmi.bmi_julia_model import BmiJulia

mymodel = BmiJulia.from_name('<package>.<model>', 'BasicModelInterface')

# Install Wflow.jl package in the Julia environment managed by the juliacall Python package.
from juliacall import Main as jl
jl.Pkg.add("Wflow")
# Create the model
from grpc4bmi.bmi_julia_model import BmiJulia
mymodel = BmiJulia.from_name('Wflow.Model', 'Wflow.bmi.BMI')

The client side

The client side has only a Python implementation. The default BMI client assumes a running server process on a given port.

from grpc4bmi.bmi_grpc_client import BmiClient
import grpc
mymodel = BmiClient(grpc.insecure_channel("localhost:<PORT>"))
print mymodel.get_component_name()
mymodel.initialize(<FILEPATH>)
...further BMI calls...

The package contains also client implementation that own the server process, either as a Python subprocess or a Docker container or a Singularity container or a Apptainer container running the run-bmi-server script. For instance

from grpc4bmi.bmi_client_subproc import BmiClientSubProcess
mymodel = BmiClientSubProcess(<PACKAGE>.<MODULE>.<CLASS>)

will automatically launch the server in a sub-process and

from grpc4bmi.bmi_client_docker import BmiClientDocker
mymodel = BmiClientDocker(<IMAGE>, <WORK DIR TO MOUNT>, input_dirs=[<INPUT DIRECTORIES TO MOUNT>])

will launch a Docker container based on supplied Docker image and will mount supplied directories to share files between the container and host.

from grpc4bmi.bmi_client_singularity import BmiClientSingularity
mymodel = BmiClientSingularity(<IMAGE>, <WORK DIR TO MOUNT>, input_dirs=[<INPUT DIRECTORIES TO MOUNT>])

will launch a singularity container on based supplied Singularity image and will mount supplied directories to share files between the container and host.

from grpc4bmi.bmi_client_apptainer import BmiClientApptainer
mymodel = BmiClientApptainer(<IMAGE>, <WORK DIR TO MOUNT>, input_dirs=[<INPUT DIRECTORIES TO MOUNT>])

will launch a Apptainer container on based supplied Apptainer image and will mount supplied directories to share files between the container and host.

For more documentation see https://grpc4bmi.readthedocs.io/.

Development: generating the gRPC code

When developers change the proto-file, it is necessary to install gRPC tools Python packages in your Python environment:

# Create virtual env
python3 -m venv .venv
. venv/bin/activate
# Make sure latest pip and wheel are install
pip install -U pip wheel
pip install -r dev-requirements.txt
# For R integration also install the R extras with
pip install -e .[R]
# For building docs (cd docs && make html) also install the docs extras with
pip install -e .[docs]

and install the C++ runtime and protoc command as described in https://github.com/google/protobuf/blob/master/src/README.md. After this, simply executing the proto_gen.sh script should do the job.