Design OpenM++ MPI job controller using GO

[chuckbelisle]

Epic: #2
Relates to: #12

[KrisWilliamson]

A branch of this project was created: openmppIss19
This branch contains an example template in a style that kubeflow mpi-operator can use and example golang code that uses it.

code in the openmpp/UI project related to the teplates and the ability to launch mpi jobs are:

• runModel.go
• runCatalog.go
  This is what has been found so far, and may not be a complete list.

[KrisWilliamson]

The Run Model button in the OpemM++ UI Gui sends the command

• POST /api/run to the openM++ oms service (web server daemon), which calls
• apiRunModelRoutes::omsAPI.go which calls
• the function runModelHandler in handleRunModel.go which calls
• the function runModel in runModel.go
  • makeCommand :: runModel.go - creates the command line argument that launches the job.

The oms daemon has a routing table set up in omsApi.go, which makes it possible to change functionality.

There are some extra functionality branching off this path.

Full path:
/openmpp/docker/ompp-build-debian/ompp/ompp-go
It is unknown why this code is in the docker directory, and duplicated.

• /ompp/ompp-go/oms

[KrisWilliamson]

Trace in golang code

• runModelHandler::handlerRunModel.go
  • runModel::runModel.go
    • cmd line constructed in this function, and executed in a goroutine (async) in lines 434-479
      • makeCommand::runModel.go - builds the command line string

[KrisWilliamson]

A branch was created to hold the updated mpi template. https://github.com/StatCan/openmpp/tree/openmppiss19

[KrisWilliamson]

Link to open M++ API documentation on how to API call to run the model
https://github.com/openmpp/openmpp.github.io/wiki/Oms-API-POST-run-the-model

[KrisWilliamson]

OMS is started via shell script start-oms.sh from the Jupyterlab dockerfile.
https://github.com/StatCan/aaw-kubeflow-containers/blob/master/output/jupyterlab-cpu/Dockerfile

[KrisWilliamson]

Code changes would be limited to the runModel.go file.
Either:

• The runModel and makeCommand functions are modified, or
• A new version of runModel and makeCommad function (runModelkf and makeCommandkf) are created and runModelHandler function is modified to point to the new runModelkf, probably stored in a new file (runModelkf.go)

These two options are almost identical except for tracking changes. The second option (create a separate file with separate functions) is probably the best option to understand the changes.