1) Pipeline prototyping
This is Yadu's original pipeline prototype, developed to show how to tasks from a Redis queue.
Installing
- Redis: Install Redis on your machine following instructions from here
- Python3.6
- Install the pipeline_prototype::
- cd pipeline_prototype pip install .
Running the pipeline
To run the pipeline:
pipeline-main
For more info on arguments for the main pipeline runner:
pipeline-main -h
The default behavior is for the pipeline to kick off a 4 chains of tasks, launch all tasks dispatched to it via the Redis queues and stop listening on the queue. This allows the pipeline to not block indefinitely on the redis queue.
Send params
Once the pipeline is running in blocking mode, you may send as many params as you wish via the pump.py utility. Separate from the pipeline running, you may pipe a simulated parameter into the pipeline from the commandline utility
pipeline-pump -i -p 100
Note
The parameter passed to the pump should be an integer or None.
Stop pipeline
To stop a pipeline that is running in blocking mode run
python3 pump.py -p None
2) Multi-server implementation prototyping
This version extends the first version to output results to another Redis queue.
Overview
An instance of the MpiMethodServer class (mpi_method_server.py) is created with input and output queues. The main_loop method implements the persistent method server. It repeatedly:
- receives a request on the input queue
- requests creation of an "MPI computation" (currently a dummy operation) to process the request
- sends the result of the computation on the output queue
Notes:
- The output queue might be piped directly to a value server. (Or, alternatively, the value server functionality could be integrated with that of the method server?)
- The implementation does not handle failures
- The implementation does not throttle the number of MPI computation requests made
Installing
See above.
Running the system
To run the system
python3 main.py [-m]
The program waits for requests (integers) submitted via a Redis "input" queue, runs a task for each request, and places the output on a Redis "output" queue. The -m flag allows for running on a Mac.
Send params
Once the system is running, you may send as many requests as you wish via the pump.py utility, using the -i flag to indicate that these requests are for the input queue. E.g.
python3 pump.py -i -p 100
Note
The parameter passed to the pump should be an integer.
Get results
Once the system is running, you may retrieve results from the output queue via the pull.py utility
python3 pull.py
By default, this is a blocking request. To impose a timeout, use the -t parameter, e.g.
python3 pull.py -t 2
Stop pipeline
To stop a pipeline that is running in blocking mode run
python3 pump.py -p None