一组执行计划任务的集群python工人

Question

I would like to build a set of cluster workers (ie droplets in digitalocean or so).

Each worker would be doing a periodic task and sending the results periodically to the main application.

Here is a pseudocode to demontrate the functionality:

Worker code

while True:
    resultFromLocationXY = calculate_my_local_task()
    send_task_to_the_main_application(resultFromLocationXY)
    time.sleep(5)

Main appplication code

In the main application I would like to asynchronously evaluate worker results:

while True:
    resultFromLocationXY = listen_to_results_from_location('xy') # not blocking, if new results are available, update the results variable

    process_the_results([resultFromLocationXY, resultFromLocationXX, ...])
    time.sleep(5)

I have been using the ipython ipcluster solution. I was able to create a remote worker and to execute a task by using an apply_async function AND arrange it all in a non blocking way.

BUT: I was not able to make periodic tasks, kind of streaming tasks. Moreover, I would like to have more nodes in one location and stream to the same variable into the main application.

I would prefer a non ipython solution.

Answer 1

To me one of the best solution could be something like

• Create a python daemon that will listen for workers. Once a worker send some data, the main daemon will create a thread and start again to listen for other workers
• Create the workers and use cron utility. In that way you could create workers on the fly and within a certain interval. Workers will call the main daemon. Once the main daemon respond with an "OK status" the worker could start