Improvement of Pandas data processing performance using Multi-threading with the Queue (Another crossover of Space Stone, Reality Stone & Power Stone)

Today, we’ll discuss how to improve your panda’s data processing power using Multi-threading. Note that, we are not going to use any third party python package. Also, we’ll be using a couple of python scripts, which we’ve already discussed in our previous posts. Hence, this time, I won’t post them here.

Please refer the following scripts –

a. callClient.py
b. callRunServer.py
c. clsConfigServer.py
d. clsEnDec.py
e. clsFlask.py
f. clsL.py
g. clsParam.py
h. clsSerial.py
i. clsWeb.py

Please find the above scripts described here with details.

So, today, we’ll be looking into how the multi-threading really helps the application to gain some performance over others.

Let’s go through our existing old sample files –

And, we’ve four columns that are applicable for encryption. This file contains 10K records. That means the application will make 40K calls to the server for a different kind of encryption for each column.

Now, if you are going with the serial approach, which I’ve already discussed here, will take significant time for data processing. However, if we could club a few rows as one block & in this way we can create multiple blocks out of our data csv like this –

As you can see that blocks are marked with a different color. So, now if you send each block of data in parallel & send the data for encryption. Ideally, you will be able to process data much faster than the usual serial process. And, this what we would be looking for with the help of python’s multi-threading & queue. Without the queue, this program won’t be possible as the queue maintains the data & process integrity.

One more thing we would like to explain here. Whenever this application is sending the block of data. It will be posting that packed into a (key, value) dictionary randomly. Key will be the thread name. The reason, we’re not expecting data after process might arrive in some random order wrapped with the dictionary as well. Once the application received all the dictionary with dataframe with encrypted/decrypted data, the data will be rearranged based on the key & then joined back with the rest of the data.

Let’s see one sample way of sending & receiving random thread –

The left-hand side, the application is splitting the recordset into small chunks of a group. Once, those group created, using python multi-threading the application is now pushing them into the queue for the producer to produce the encrypted/decrypted value. Similar way, after processing the application will push the final product into the queue for consuming the final output.

This is the pictorial representation of dictionary ordering based on the key-value & then the application will extract the entire data to form the target csv file.

Let’s explore the script –

1. clsParallel.py (This script will consume the split csv files & send the data blocks in the form of the dictionary using multi-threading to the API for encryption in parallel. Hence, the name comes into the picture.)

import pandas as p
import clsWeb as cw
import datetime
from clsParam import clsParam as cf
import threading
from queue import Queue
import gc
import signal
import time
import os

# Declaring Global Variable
q = Queue()
m = Queue()
tLock = threading.Lock()
threads = []

fin_dict = {}
fin_dict_1 = {}
stopping = threading.Event()

# Disbling Warnings
def warn(*args, **kwargs):
    pass
import warnings
warnings.warn = warn

class clsParallel(object):
    def __init__(self):
        self.path = cf.config['PATH']
        self.EncryptMode = str(cf.config['ENCRYPT_MODE'])
        self.DecryptMode = str(cf.config['DECRYPT_MODE'])
        self.num_worker_threads = int(cf.config['NUM_OF_THREAD'])
        

    # Lookup Methods for Encryption
    def encrypt_acctNbr(self, row):
        # Declaring Local Variable
        en_AcctNbr = ''
        json_source_str = ''

        # Capturing essential values
        EncryptMode = self.EncryptMode
        lkp_acctNbr = row['Acct_Nbr']
        str_acct_nbr = str(lkp_acctNbr)
        fil_acct_nbr = str_acct_nbr.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_acct_nbr + '","dataTemplate":"subGrAcct_Nbr"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_acct_nbr)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            en_AcctNbr = x.getResponse(EncryptMode)
        else:
            en_AcctNbr = ''

        return en_AcctNbr

    def encrypt_Name(self, row):
        # Declaring Local Variable
        en_AcctName = ''

        # Capturing essential values
        EncryptMode = self.EncryptMode
        lkp_acctName = row['Name']
        str_acct_name = str(lkp_acctName)
        fil_acct_name = str_acct_name.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_acct_name + '","dataTemplate":"subGrName"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_acct_name)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            en_AcctName = x.getResponse(EncryptMode)
        else:
            en_AcctName = ''

        return en_AcctName

    def encrypt_Phone(self, row):
        # Declaring Local Variable
        en_Phone = ''

        # Capturing essential values
        EncryptMode = self.EncryptMode
        lkp_phone = row['Phone']
        str_phone = str(lkp_phone)
        fil_phone = str_phone.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_phone + '","dataTemplate":"subGrPhone"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_phone)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            en_Phone = x.getResponse(EncryptMode)
        else:
            en_Phone = ''

        return en_Phone

    def encrypt_Email(self, row):
        # Declaring Local Variable
        en_Email = ''

        # Capturing essential values
        EncryptMode = self.EncryptMode
        lkp_email = row['Email']
        str_email = str(lkp_email)
        fil_email = str_email.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_email + '","dataTemplate":"subGrEmail"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_email)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            en_Email = x.getResponse(EncryptMode)
        else:
            en_Email = ''

        return en_Email

    # Lookup Methods for Decryption
    def decrypt_acctNbr(self, row):
        # Declaring Local Variable
        de_AcctNbr = ''
        json_source_str = ''

        # Capturing essential values
        EncryptMode = self.DecryptMode
        lkp_acctNbr = row['Acct_Nbr']
        str_acct_nbr = str(lkp_acctNbr)
        fil_acct_nbr = str_acct_nbr.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_acct_nbr + '","dataTemplate":"subGrAcct_Nbr"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_acct_nbr)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            de_AcctNbr = x.getResponse(EncryptMode)
        else:
            de_AcctNbr = ''

        return de_AcctNbr

    def decrypt_Name(self, row):
        # Declaring Local Variable
        de_AcctName = ''

        # Capturing essential values
        EncryptMode = self.DecryptMode
        lkp_acctName = row['Name']
        str_acct_name = str(lkp_acctName)
        fil_acct_name = str_acct_name.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_acct_name + '","dataTemplate":"subGrName"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_acct_name)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            de_AcctName = x.getResponse(EncryptMode)
        else:
            de_AcctName = ''

        return de_AcctName

    def decrypt_Phone(self, row):
        # Declaring Local Variable
        de_Phone = ''

        # Capturing essential values
        EncryptMode = self.DecryptMode
        lkp_phone = row['Phone']
        str_phone = str(lkp_phone)
        fil_phone = str_phone.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_phone + '","dataTemplate":"subGrPhone"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_phone)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            de_Phone = x.getResponse(EncryptMode)
        else:
            de_Phone = ''

        return de_Phone

    def decrypt_Email(self, row):
        # Declaring Local Variable
        de_Email = ''

        # Capturing essential values
        EncryptMode = self.DecryptMode
        lkp_email = row['Email']
        str_email = str(lkp_email)
        fil_email = str_email.strip()

        # Forming JSON String for this field
        json_source_str = '{"dataGroup":"GrDet","data":"' + fil_email + '","dataTemplate":"subGrEmail"}'

        # Identifying Length of the field
        len_acct_nbr = len(fil_email)

        # This will trigger the service if it has valid data
        if len_acct_nbr > 0:
            x = cw.clsWeb(json_source_str)
            de_Email = x.getResponse(EncryptMode)
        else:
            de_Email = ''

        return de_Email

    def getEncrypt(self, df_dict):
        try:
            df_input = p.DataFrame()
            df_fin = p.DataFrame()

            # Assigning Target File Basic Name
            for k, v in df_dict.items():
                Thread_Name = k
                df_input = v

            # Checking total count of rows
            count_row = int(df_input.shape[0])
            # print('Part number of records to process:: ', count_row)

            if count_row > 0:

                # Deriving rows
                df_input['Encrypt_Acct_Nbr'] = df_input.apply(lambda row: self.encrypt_acctNbr(row), axis=1)
                df_input['Encrypt_Name'] = df_input.apply(lambda row: self.encrypt_Name(row), axis=1)
                df_input['Encrypt_Phone'] = df_input.apply(lambda row: self.encrypt_Phone(row), axis=1)
                df_input['Encrypt_Email'] = df_input.apply(lambda row: self.encrypt_Email(row), axis=1)

                # Dropping original columns
                df_input.drop(['Acct_Nbr', 'Name', 'Phone', 'Email'], axis=1, inplace=True)

                # Renaming new columns with the old column names
                df_input.rename(columns={'Encrypt_Acct_Nbr':'Acct_Nbr'}, inplace=True)
                df_input.rename(columns={'Encrypt_Name': 'Name'}, inplace=True)
                df_input.rename(columns={'Encrypt_Phone': 'Phone'}, inplace=True)
                df_input.rename(columns={'Encrypt_Email': 'Email'}, inplace=True)

                # New Column List Orders
                column_order = ['Acct_Nbr', 'Name', 'Acct_Addr_1', 'Acct_Addr_2', 'Phone', 'Email', 'Serial_No']
                df_fin = df_input.reindex(column_order, axis=1)

                fin_dict[Thread_Name] = df_fin

            return 0
        except Exception as e:
            df_error = p.DataFrame({'Acct_Nbr':str(e), 'Name':'', 'Acct_Addr_1':'', 'Acct_Addr_2':'', 'Phone':'', 'Email':'', 'Serial_No':''})
            fin_dict[Thread_Name] = df_error

            return 1

    def getEncryptWQ(self):
        item_dict = {}
        item = ''

        while True:
            try:
                #item_dict = q.get()
                item_dict = q.get_nowait()

                for k, v in item_dict.items():
                    # Assigning Target File Basic Name
                    item = str(k)

                if ((item == 'TEND') | (item == '')):
                    break

                if ((item != 'TEND') | (item != '')):
                    self.getEncrypt(item_dict)

                q.task_done()
            except Exception:
                break

    def getEncryptParallel(self, df_payload):
        start_pos = 0
        end_pos = 0
        l_dict = {}
        c_dict = {}
        min_val_list = {}
        cnt = 0
        num_worker_threads = self.num_worker_threads
        split_df = p.DataFrame()
        df_ret = p.DataFrame()

        # Assigning Target File Basic Name
        df_input = df_payload

        # Checking total count of rows
        count_row = df_input.shape[0]
        print('Total number of records to process:: ', count_row)

        interval = int(count_row / num_worker_threads) + 1
        actual_worker_task = int(count_row / interval) + 1

        for i in range(actual_worker_task):
            t = threading.Thread(target=self.getEncryptWQ)
            t.start()
            threads.append(t)
            name = str(t.getName())

            if ((start_pos + interval) < count_row):
                end_pos = start_pos + interval
            else:
                end_pos = start_pos + (count_row - start_pos)

            split_df = df_input.iloc[start_pos:end_pos]
            l_dict[name] = split_df

            if ((start_pos > count_row) | (start_pos == count_row)):
                break
            else:
                start_pos = start_pos + interval

            q.put(l_dict)
            cnt += 1

        # block until all tasks are done
        q.join()

        # stop workers
        for i in range(actual_worker_task):
            c_dict['TEND'] = p.DataFrame()
            q.put(c_dict)

        for t in threads:
            t.join()

        for k, v in fin_dict.items():
            min_val_list[int(k.replace('Thread-',''))] = v

        min_val = min(min_val_list, key=int)

        for k, v in sorted(fin_dict.items(), key=lambda k:int(k[0].replace('Thread-',''))):
            if int(k.replace('Thread-','')) == min_val:
                df_ret = fin_dict[k]
            else:
                d_frames = [df_ret, fin_dict[k]]
                df_ret = p.concat(d_frames)

        # Releasing Memory
        del[[split_df]]
        gc.collect()

        return df_ret

    def getDecrypt(self, df_encrypted_dict):
        try:
            df_input = p.DataFrame()
            df_fin = p.DataFrame()

            # Assigning Target File Basic Name
            for k, v in df_encrypted_dict.items():
                Thread_Name = k
                df_input = v

            # Checking total count of rows
            count_row = int(df_input.shape[0])

            if count_row > 0:

                # Deriving rows
                df_input['Decrypt_Acct_Nbr'] = df_input.apply(lambda row: self.decrypt_acctNbr(row), axis=1)
                df_input['Decrypt_Name'] = df_input.apply(lambda row: self.decrypt_Name(row), axis=1)
                df_input['Decrypt_Phone'] = df_input.apply(lambda row: self.decrypt_Phone(row), axis=1)
                df_input['Decrypt_Email'] = df_input.apply(lambda row: self.decrypt_Email(row), axis=1)

                # Dropping original columns
                df_input.drop(['Acct_Nbr', 'Name', 'Phone', 'Email'], axis=1, inplace=True)

                # Renaming new columns with the old column names
                df_input.rename(columns={'Decrypt_Acct_Nbr':'Acct_Nbr'}, inplace=True)
                df_input.rename(columns={'Decrypt_Name': 'Name'}, inplace=True)
                df_input.rename(columns={'Decrypt_Phone': 'Phone'}, inplace=True)
                df_input.rename(columns={'Decrypt_Email': 'Email'}, inplace=True)

                # New Column List Orders
                column_order = ['Acct_Nbr', 'Name', 'Acct_Addr_1', 'Acct_Addr_2', 'Phone', 'Email']
                df_fin = df_input.reindex(column_order, axis=1)

                fin_dict_1[Thread_Name] = df_fin

            return 0

        except Exception as e:
            df_error = p.DataFrame({'Acct_Nbr': str(e), 'Name': '', 'Acct_Addr_1': '', 'Acct_Addr_2': '', 'Phone': '', 'Email': ''})
            fin_dict_1[Thread_Name] = df_error

            return 1

    def getDecryptWQ(self):
        item_dict = {}
        item = ''

        while True:
            try:
                #item_dict = q.get()
                item_dict = m.get_nowait()

                for k, v in item_dict.items():
                    # Assigning Target File Basic Name
                    item = str(k)

                if ((item == 'TEND') | (item == '')):
                    return True
                    #break

                if ((item != 'TEND') | (item != '')):
                    self.getDecrypt(item_dict)

                m.task_done()
            except Exception:
                break


    def getDecryptParallel(self, df_payload):
        start_pos = 0
        end_pos = 0
        l_dict_1 = {}
        c_dict_1 = {}
        cnt = 0
        num_worker_threads = self.num_worker_threads
        split_df = p.DataFrame()
        df_ret_1 = p.DataFrame()

        min_val_list = {}

        # Assigning Target File Basic Name
        df_input_1 = df_payload

        # Checking total count of rows
        count_row = df_input_1.shape[0]
        print('Total number of records to process:: ', count_row)

        interval = int(count_row / num_worker_threads) + 1
        actual_worker_task = int(count_row / interval) + 1

        for i in range(actual_worker_task):
            t_1 = threading.Thread(target=self.getDecryptWQ)
            t_1.start()
            threads.append(t_1)
            name = str(t_1.getName())

            if ((start_pos + interval) < count_row):
                end_pos = start_pos + interval
            else:
                end_pos = start_pos + (count_row - start_pos)

            split_df = df_input_1.iloc[start_pos:end_pos]
            l_dict_1[name] = split_df

            if ((start_pos > count_row) | (start_pos == count_row)):
                break
            else:
                start_pos = start_pos + interval

            m.put(l_dict_1)
            cnt += 1

        # block until all tasks are done
        m.join()

        # stop workers
        for i in range(actual_worker_task):
            c_dict_1['TEND'] = p.DataFrame()
            m.put(c_dict_1)

        for t_1 in threads:
            t_1.join()

        for k, v in fin_dict_1.items():
            min_val_list[int(k.replace('Thread-',''))] = v

        min_val = min(min_val_list, key=int)

        for k, v in sorted(fin_dict_1.items(), key=lambda k:int(k[0].replace('Thread-',''))):
            if int(k.replace('Thread-','')) == min_val:
                df_ret_1 = fin_dict_1[k]
            else:
                d_frames = [df_ret_1, fin_dict_1[k]]
                df_ret_1 = p.concat(d_frames)

        # Releasing Memory
        del[[split_df]]
        gc.collect()

        return df_ret_1

Let’s explain the key snippet from the code. For your information, we’re not going to describe all the encryption methods such as –

# Encryption Method
encrypt_acctNbr
encrypt_Name
encrypt_Phone
encrypt_Email

# Decryption Method
decrypt_acctNbr
decrypt_Name
decrypt_Phone
decrypt_Email

As we’ve already described the logic of these methods in our previous post.

# Checking total count of rows
count_row = df_input.shape[0]
print('Total number of records to process:: ', count_row)

interval = int(count_row / num_worker_threads) + 1
actual_worker_task = int(count_row / interval) + 1

Fetching the total number of rows from the dataframe. Based on the row count, the application will derive the actual number of threads that will be used for parallelism.

for i in range(actual_worker_task):
    t = threading.Thread(target=self.getEncryptWQ)
    t.start()
    threads.append(t)
    name = str(t.getName())

    if ((start_pos + interval) < count_row):
        end_pos = start_pos + interval
    else:
        end_pos = start_pos + (count_row - start_pos)

    split_df = df_input.iloc[start_pos:end_pos]
    l_dict[name] = split_df

    if ((start_pos > count_row) | (start_pos == count_row)):
        break
    else:
        start_pos = start_pos + interval

    q.put(l_dict)
    cnt += 1

Here, the application is splitting the data into multiple groups of smaller data packs & then combining them into (key, value) dictionary & finally placed them into the individual queue.

# block until all tasks are done
q.join()

This will join the queue process. This will ensure that queues are free after consuming the data.

# stop workers
for i in range(actual_worker_task):
    c_dict['TEND'] = p.DataFrame()
    q.put(c_dict)

for t in threads:
    t.join()

The above lines are essential. As this will help the process to identify that no more data are left to send at the queue. And, the main thread will wait until all the threads are done.

for k, v in fin_dict.items():
    min_val_list[int(k.replace('Thread-',''))] = v

min_val = min(min_val_list, key=int)

Once, all the jobs are done. The application will find the minimum thread value & based on that we can sequence all the data chunks as explained in our previous image & finally clubbed them together to form the complete csv.

for k, v in sorted(fin_dict.items(), key=lambda k:int(k[0].replace('Thread-',''))):
    if int(k.replace('Thread-','')) == min_val:
        df_ret = fin_dict[k]
    else:
        d_frames = [df_ret, fin_dict[k]]
        df_ret = p.concat(d_frames)

As already explained, using the starting point of our data dictionary element, the application is clubbing the data back to the main csv.

Next method, which we’ll be explaining is –

getEncryptWQ

Please find the key lines –

while True:
    try:
        #item_dict = q.get()
        item_dict = q.get_nowait()

        for k, v in item_dict.items():
            # Assigning Target File Basic Name
            item = str(k)

        if ((item == 'TEND') | (item == '')):
            break

        if ((item != 'TEND') | (item != '')):
            self.getEncrypt(item_dict)

        q.task_done()
    except Exception:
        break

This method will consume the data & processing it for encryption or decryption. This will continue to do the work until or unless it receives the key value as TEND or the queue is empty.

Let’s compare the statistics between Windows & MAC.

Let’s see the file structure first –

Windows (16 GB – Core 2) Vs Mac (10 GB – Core 2):

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Find the complete directory from both the machine.

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Here is the final output –

So, we’ve achieved our target goal.

Let me know – how do you like this post. Please share your suggestion & comments.

I’ll be back with another installment from the Python verse.

Till then – Happy Avenging!

	Building a real-time… on Projecting real-time KPIs by i…
	RAG implementation o… on RAG implementation of LLMs by…
	Realtime reading fro… on Live visual reading using Conv…
	Neural prophet… on Predicting real-time Covid-19…
	Projecting real-time… on Python-based dash framework vi…

Improvement of Pandas data processing performance using Multi-threading with the Queue (Another crossover of Space Stone, Reality Stone & Power Stone)

Windows (16 GB – Core 2) Vs Mac (10 GB – Core 2):

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Find the complete directory from both the machine.

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Like this:

Related

Published by SatyakiDe

Leave a ReplyCancel reply

Windows (16 GB – Core 2) Vs Mac (10 GB – Core 2):

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Find the complete directory from both the machine.

Windows (16 GB – Core 2):

Mac (10 GB – Core 2):

Share this:

Like this:

Related

Published by SatyakiDe

Leave a ReplyCancel reply

Discover more from Satyaki De's Blog