DPF threading and RSPLIT APDL Command

Mike.Thompson

I was curious if anyone knows the best methods to do post processing on lots of different mesh entities. Example: You have a model with lots of bolts and you want to do time-consuming post processing on each one. I am thinking of creating multiple smaller results files for each bolt with RSPLIT command. Then use threading module with mech_dpf to do post processing on each bolt-rst in separate threads.

Or, should I keep it in a single results file, and use a results extraction operator on a single thread in sequence, and send that raw data out for further processing on separate thread?

Would this be an efficient way to do this, or am I missing an easier, more obvious way?

Pernelle Marone-Hitz

I think @Pierre Thieffry should be able to provide some useful inputs on this one.

Pierre Thieffry

@Mike.Thompson I recently worked on a multi-thread dpf example. I don't think you need to split the rst file.

You could either:

• load your entire result field for the full model and perform operations on each bolt on different dpf servers using multi-threading
• perform all operations including the retrieval of the result field for each bolt on different dpf servers.

Second option would be probably a bit more memory efficient. First option is if you need to operate on more parts of the model than just the bolt.

Here's a skeleton script for multi-threading:

import threading


import ansys.dpf.core as dpf
import time,sys
import os


class returnValueThread(threading.Thread):
    """
        Class defined so the threads will return custom values
    """
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.result = None

    def run(self):
        if self._target is None:
            return  # could alternatively raise an exception, depends on the use case
        try:
            self.result = self._target(*self._args, **self._kwargs)
        except Exception as exc:
            print(f'{type(exc).__name__}: {exc}', file=sys.stderr)  # properly handle the exception

    def join(self, *args, **kwargs):
        super().join(*args, **kwargs)
        return self.result


def worker_function_for_bolt(worker parameters)

    return bolt_data

def getGlobalData(rst_file,server):    

    return global_data


if __name__ == '__main__':

    nservers=4

    # Results file 
    rst_file= r'...'

    # Create dpf servers
    servers_list=[]
    for proc in range(nproc):
        server = dpf.start_local_server(
            as_global=False, config=dpf.AvailableServerConfigs.InProcessServer)
        servers_list.append(server)

    # Get global fields to be used in susbsequent operations. 
    # Done only once and on first server
    global_data=getGlobalData(rst_file,servers_list[0])


    bolt_results={}

    sidx=0   
    # We cycle over server ids to spread the work over multiple servers
    threads_list=[]
    for boltid in bolt_list:            
        server=servers_list[sidx]

        x = returnValueThread(target=worker_function_for_bolt, args=[worker args])
        x.start()
        threads_list.append(x)
        sidx += 1
        if sidx ==nproc:
            sidx = 0

    # Gather output from threads
    for th in threads_list:
        bolt_results.update(th.join()) # bolt_resutls will gather the outputs from the worker function

    dpf.server.shutdown_all_session_servers()