main.py

# fedavg from https://github.com/WHDY/FedAvg/
# TODO redistribute offline() based on very transaction, not at the beginning of every loop
# TODO when accepting transactions, check comm_round must be in the same, that is to assume by default they only accept the transactions that are in the same round, so the final block contains only updates from the same round
# TODO subnets - potentially resolved by let each miner sends block 
# TODO let's not remove peers because of offline as peers may go back online later, instead just check if they are online or offline. Only remove peers if they are malicious. In real distributed network, remove peers if cannot reach for a long period of time
# assume no resending transaction mechanism if a transaction is lost due to offline or time out. most of the time, unnecessary because workers always send the newer updates, if it's not the last worker's updates
# assume just skip verifying a transaction if offline, in reality it may continue to verify what's left
# PoS also uses resync chain - the chain with highter stake
# only focus on catch malicious worker
# TODO need to make changes in these functions on Sunday
#pow_resync_chain
#update_model_after_chain_resync
# TODO miner sometimes receives worker transactions directly for unknown reason - discard tx if it's not the correct type
# TODO a chain is invalid if a malicious block is identified after this miner is identified as malicious
# TODO Do not associate with blacklisted node. This may be done already.
# TODO KickR continuousness should skip the rounds when nodes are not selected as workers
# TODO update forking log after loading network snapshots
# TODO in reuqest_to_download, forgot to check for maliciousness of the block miner
# future work
# TODO - non-even dataset distribution

import os
import sys
import argparse
import numpy as np
import random
import time
from datetime import datetime
import copy
from sys import getsizeof
import sqlite3
import pickle
from pathlib import Path
import shutil
import torch
import torch.nn.functional as F
from Models import Mnist_2NN, Mnist_CNN, Mnist_CNN_Simplified, Cifar10_CNN, Cifar10_CNN_Simplified, Cifar100_CNN, Cifar100_ResNet
from Device import Device, DevicesInNetwork
from Block import Block
from Blockchain import Blockchain

# set program execution time for logging purpose
date_time = datetime.now().strftime("%m%d%Y_%H%M%S")
log_files_folder_path = f"logs/{date_time}"
NETWORK_SNAPSHOTS_BASE_FOLDER = "snapshots"
# for running on Google Colab
# log_files_folder_path = f"/content/drive/MyDrive/BFA/logs/{date_time}"
# NETWORK_SNAPSHOTS_BASE_FOLDER = "/content/drive/MyDrive/BFA/snapshots"

parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Blockchain_FL_Simulation")

# debug attributes
parser.add_argument('-g', '--gpu', type=str, default='0', help='gpu id to use(e.g. 0,1,2,3)')
parser.add_argument('-v', '--verbose', type=int, default=1, help='print verbose debug log')
parser.add_argument('-sn', '--save_network_snapshots', type=int, default=0, help='only save network_snapshots if this is set to 1; will create a folder with date in the snapshots folder')
parser.add_argument('-dtx', '--destroy_tx_in_block', type=int, default=0, help='currently transactions stored in the blocks are occupying GPU ram and have not figured out a way to move them to CPU ram or harddisk, so turn it on to save GPU ram in order for PoS to run 100+ rounds. NOT GOOD if there needs to perform chain resyncing.')
parser.add_argument('-rp', '--resume_path', type=str, default=None, help='resume from the path of saved network_snapshots; only provide the date')
parser.add_argument('-sf', '--save_freq', type=int, default=10, help='save frequency of the network_snapshot')
parser.add_argument('-sm', '--save_most_recent', type=int, default=1, help='in case of saving space, keep only the recent specified number of snapshops; 0 means keep all')

# FL attributes
parser.add_argument('-data', '--dataset', type=str, default='mnist')
parser.add_argument('-B', '--batchsize', type=int, default=10, help='local train batch size')
parser.add_argument('-mn', '--model_name', type=str, default='mnist_cnn', help='the model to train')
parser.add_argument('-lr', "--learning_rate", type=float, default=0.01, help="learning rate, use value from origin paper as default")
parser.add_argument('-op', '--optimizer', type=str, default="SGD", help='optimizer to be used, by default implementing stochastic gradient descent')
parser.add_argument('-iid', '--IID', type=int, default=0, help='the way to allocate data to devices')
parser.add_argument('-max_ncomm', '--max_num_comm', type=int, default=100, help='maximum number of communication rounds, may terminate early if converges')
parser.add_argument('-nd', '--num_devices', type=int, default=20, help='numer of the devices in the simulation network')
parser.add_argument('-st', '--shard_test_data', type=int, default=0, help='it is easy to see the global models are consistent across devices when the test dataset is NOT sharded')
# parser.add_argument('-nm', '--num_malicious', type=int, default=0, help="number of malicious nodes in the network. malicious node's data sets will be introduced Gaussian noise")
parser.add_argument('-nm', '--num_malicious', type=str, default='0,0', help="number of malicious nodes in the network. order by worker and miner. e.g. 4,1 assign 4 malicious workers and 1 malicious miners.")
parser.add_argument('-nv', '--noise_variance', type=int, default=1, help="noise variance level of the injected Gaussian Noise")
parser.add_argument('-le', '--default_local_epochs', type=int, default=2, help='local train epoch. Train local model by this same num of epochs for each worker, if -mt is not specified')

# blockchain system consensus attributes
parser.add_argument('-ur', '--unit_reward', type=int, default=1, help='unit reward for providing data, verification of signature, validation and so forth')
parser.add_argument('-ko', '--knock_out_rounds', type=int, default=6, help="a worker or miner device is kicked out of the device's peer list(put in black list) if it's identified as malicious for this number of rounds")
parser.add_argument('-lo', '--lazy_worker_knock_out_rounds', type=int, default=10, help="a worker device is kicked out of the device's peer list(put in black list) if it does not provide updates for this number of rounds, due to too slow or just lazy to do updates and only accept the model udpates.(do not care lazy miner as they will just not receive rewards)")
parser.add_argument('-pow', '--pow_difficulty', type=int, default=0, help="if set to 0, meaning miners are not using PoW")
parser.add_argument('-cons', '--consensus', type=str, default='PoW', help="including PoW, PoS, PBFT")


# blockchain FL miner restriction tuning parameters
parser.add_argument('-mt', '--miner_acception_wait_time', type=float, default=float("inf"), help="default time window for miners to accept transactions, in seconds.")
parser.add_argument('-wt', '--worker_acception_wait_time', type=float, default=float("inf"), help="default time window for workers to accept transactions, in seconds. 0 means no time limit, and each device will just perform same amount(-le) of epochs per round like in FedAvg paper")
parser.add_argument('-ml', '--miner_accepted_transactions_size_limit', type=float, default=0.0, help="no further transactions will be accepted by miner after this limit. 0 means no size limit. either this or -mt has to be specified, or both. This param determines the final block_size")
parser.add_argument('-mp', '--miner_poe_propagated_block_wait_time', type=float, default=float("inf"), help="this wait time is counted from the beginning of the comm round, used to simulate forking events in PoE")
parser.add_argument('-vh', '--validate_threshold', type=float, default=0.05, help="a threshold value of accuracy difference to determine malicious worker") #TODO
parser.add_argument('-md', '--malicious_updates_discount', type=float, default=0.0, help="do not entirely drop the voted negative worker transaction because that risks the same worker dropping the entire transactions and repeat its accuracy again and again and will be kicked out. Apply a discount factor instead to the false negative worker's updates are by some rate applied so it won't repeat")
parser.add_argument('-mmo', '--malicious_miner_on', type=int, default=0, help="let malicious miner flip voting result")
parser.add_argument('-mwo', '--malicious_worker_on', type=int, default=0, help="let malicious worker flip voting result")



# distributed system attributes
parser.add_argument('-ns', '--network_stability', type=float, default=1.0, help='the odds a device is online')
parser.add_argument('-els', '--even_link_speed_strength', type=int, default=1, help="This variable is used to simulate transmission delay. Default value 1 means every device is assigned to the same link speed strength -dts bytes/sec. If set to 0, link speed strength is randomly initiated between 0 and 1, meaning a device will transmit  -els*-dts bytes/sec - during experiment, one transaction is around 35k bytes.")
parser.add_argument('-dts', '--base_data_transmission_speed', type=float, default=70000.0, help="volume of data can be transmitted per second when -els == 1. set this variable to determine transmission speed (bandwidth), which further determines the transmission delay - during experiment, one transaction is around 35k bytes.")
parser.add_argument('-ecp', '--even_computation_power', type=int, default=1, help="This variable is used to simulate strength of hardware equipment. The calculation time will be shrunk down by this value. Default value 1 means evenly assign computation power to 1. If set to 0, power is randomly initiated as an int between 0 and 4, both included.")

# simulation attributes
parser.add_argument('-ha', '--hard_assign', type=str, default='*,*', help="hard assign number of roles in the network, order by worker and miner. e.g. 12,8 assign 12 workers and 8 miners. \"*,*\" means completely random role-assigning in each communication round")
parser.add_argument('-aio', '--all_in_one', type=int, default=1, help='let all nodes be aware of each other in the network while registering')
parser.add_argument('-cs', '--check_signature', type=int, default=1, help='if set to 0, all signatures are assumed to be verified to save execution time')

# parser.add_argument('-la', '--least_assign', type=str, default='*,*,*', help='the assigned number of roles are at least guaranteed in the network')

if __name__=="__main__":

	# create logs/ if not exists
	if not os.path.exists('logs'):
		os.makedirs('logs')

	# get arguments
	args = parser.parse_args()
	args = args.__dict__ #convert the parsed arguments into a dictionary format
	
	# detect CUDA
	dev = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

	# pre-define system variables
	latest_round_num = 0

	''' If network_snapshot is specified, continue from left '''
	if args['resume_path']:
		if not args['save_network_snapshots']:
			print("NOTE: save_network_snapshots is set to 0. New network_snapshots won't be saved by conituing.")
		network_snapshot_save_path = f"{NETWORK_SNAPSHOTS_BASE_FOLDER}/{args['resume_path']}"
		latest_network_snapshot_file_name = sorted([f for f in os.listdir(network_snapshot_save_path) if not f.startswith('.')], key = lambda fn: int(fn.split('_')[-1]) , reverse=True)[0]
		print(f"Loading network snapshot from {args['resume_path']}/{latest_network_snapshot_file_name}")
		print("BE CAREFUL - loaded dev env must be the same as the current dev env, namely, cpu, gpu or gpu parallel")
		latest_round_num = int(latest_network_snapshot_file_name.split('_')[-1])
		devices_in_network = pickle.load(open(f"{network_snapshot_save_path}/{latest_network_snapshot_file_name}", "rb"))
		devices_list = list(devices_in_network.devices_set.values())
		log_files_folder_path = f"logs/{args['resume_path']}"
		# for colab
		# log_files_folder_path = f"/content/drive/MyDrive/BFA/logs/{args['resume_path']}"
		# original arguments file
		args_used_file = f"{log_files_folder_path}/args_used.txt"
		file = open(args_used_file,"r") 
		log_whole_text = file.read() 
		lines_list = log_whole_text.split("\n")
		for line in lines_list:
			# abide by the original specified rewards
			if line.startswith('--unit_reward'):
				rewards = int(line.split(" ")[-1])
			# get number of roles
			if line.startswith('--hard_assign'):
				roles_requirement = line.split(" ")[-1].split(',')
			# get mining consensus
			if line.startswith('--pow_difficulty'):
				mining_consensus = 'PoW' if int(line.split(" ")[-1]) else 'PoS'
			if line.startswith('--num_malicious'):
				num_malicious = line.split(" ")[-1].split(',')
		# determine roles to assign
		try:
			workers_needed = int(roles_requirement[0])
		except:
			workers_needed = 1
		try:
			miners_needed = int(roles_requirement[-1])
		except:
			miners_needed = 1
	else:
		''' SETTING UP FROM SCRATCH'''
		
		# 0. create log_files_folder_path if not resume
		os.mkdir(log_files_folder_path)

		# 1. save arguments used
		with open(f'{log_files_folder_path}/args_used.txt', 'w') as f:
			f.write("Command line arguments used -\n")
			f.write(' '.join(sys.argv[1:]))
			f.write("\n\nAll arguments used -\n")
			for arg_name, arg in args.items():
				f.write(f'\n--{arg_name} {arg}')
				
		# 2. create network_snapshot folder
		if args['save_network_snapshots']:
			network_snapshot_save_path = f"{NETWORK_SNAPSHOTS_BASE_FOLDER}/{date_time}"
			os.mkdir(network_snapshot_save_path)

		# 3. assign system variables
		# for demonstration purposes, this reward is for every rewarded action
		rewards = args["unit_reward"]
		
		# 4. get number of roles needed in the network
		roles_requirement = list(map(int, args['hard_assign'].split(',')))
		# determine roles to assign
		try:
			workers_needed = roles_requirement[0]
		except:
			workers_needed = 1
		try:
			miners_needed = roles_requirement[-1]
		except:
			miners_needed = 1

		# 5. check arguments eligibility

		num_devices = args['num_devices']
		num_malicious = list(map(int, args['num_malicious'].split(',')))
		# num_malicious = args['num_malicious']
		
		if num_devices < workers_needed + miners_needed:
			sys.exit("ERROR: Roles assigned to the devices exceed the maximum number of allowed devices in the network.")

		if num_devices < 2:
			sys.exit("ERROR: There are not enough devices in the network.\n The system needs at least one miner and one worker to start the operation.\nSystem aborted.")

		
		if num_malicious:
			if (num_malicious[0]+num_malicious[-1]) > num_devices:
				sys.exit("ERROR: The number of malicious nodes cannot exceed the total number of devices set in this network")
			else:
				print(f"Malicious workers rate: {(num_malicious[0]/workers_needed)*100:.2f}%. Malicious miners rate: {(num_malicious[-1]/miners_needed)*100:.2f}%")

		# 6. create neural net based on the input model name
		net = None
		if args['model_name'] == 'mnist_2nn':
			net = Mnist_2NN()
		elif args['model_name'] == 'mnist_cnn':
			net = Mnist_CNN()#Mnist_CNN_Simplified()
		elif args['model_name'] == 'cifar10_cnn':
			net = Cifar10_CNN_Simplified()#Cifar10_CNN()
		elif args['model_name'] == 'cifar100_cnn':
			net = Cifar100_CNN()
		elif args['model_name'] == 'cifar100_resnet':
			net = Cifar100_ResNet()
		else:
			sys.exit("ERROR: Invalid model name. Please choose from mnist_2nn, mnist_cnn, cifar10_cnn, cifar100_resnet.")

		# 7. assign GPU(s) if available to the net, otherwise CPU
		# os.environ['CUDA_VISIBLE_DEVICES'] = args['gpu']
		if torch.cuda.device_count() > 1:
			net = torch.nn.DataParallel(net)
		print(f"{torch.cuda.device_count()} GPUs are available to use!")
		net = net.to(dev)

		# 8. set loss_function
		loss_func = F.cross_entropy

		# 9. create devices in the network
		devices_in_network = DevicesInNetwork(data_set_name=args['dataset'], is_iid=args['IID'], batch_size = args['batchsize'], learning_rate =  args['learning_rate'], loss_func = loss_func, opti = args['optimizer'], num_devices=num_devices, roles_requirement=roles_requirement, network_stability=args['network_stability'], net=net, dev=dev, knock_out_rounds=args['knock_out_rounds'], lazy_worker_knock_out_rounds=args['lazy_worker_knock_out_rounds'], shard_test_data=args['shard_test_data'], miner_acception_wait_time=args['miner_acception_wait_time'], worker_acception_wait_time=args['worker_acception_wait_time'], miner_accepted_transactions_size_limit=args['miner_accepted_transactions_size_limit'], validate_threshold=args['validate_threshold'], pow_difficulty=args['pow_difficulty'], even_link_speed_strength=args['even_link_speed_strength'], base_data_transmission_speed=args['base_data_transmission_speed'], even_computation_power=args['even_computation_power'], malicious_updates_discount=args['malicious_updates_discount'], num_malicious=num_malicious, noise_variance=args['noise_variance'], check_signature=args['check_signature'], not_resync_chain=args['destroy_tx_in_block'])
		del net
		devices_list = list(devices_in_network.devices_set.values())

		# 10. register devices and initialize global parameterms
		for device in devices_list:
			# set initial global weights
			device.init_global_parameters()
			device.init_candidate_parameters()
			# helper function for registration simulation - set devices_list and aio
			device.set_devices_dict_and_aio(devices_in_network.devices_set, args["all_in_one"])
			# simulate peer registration, with respect to device idx order
			device.register_in_the_network()
		# remove its own from peer list if there is
		for device in devices_list:
			device.remove_peers(device)

		# 11. build logging files/database path
		# create log files
		open(f"{log_files_folder_path}/correctly_kicked_workers.txt", 'w').close()
		open(f"{log_files_folder_path}/mistakenly_kicked_workers.txt", 'w').close()
		open(f"{log_files_folder_path}/false_positive_malious_nodes_inside_slipped.txt", 'w').close()
		open(f"{log_files_folder_path}/false_negative_good_nodes_inside_victims.txt", 'w').close()
		open(f"{log_files_folder_path}/kicked_lazy_workers.txt", 'w').close()

		# 12. setup the mining consensus #consensus
		mining_consensus = args['consensus']

	# create malicious worker identification database 创建恶意节点识别数据库？
	conn = sqlite3.connect(f'{log_files_folder_path}/malicious_wokrer_identifying_log.db') #This line establishes a connection to a SQLite database file
	conn_cursor = conn.cursor() #This line creates a cursor object associated with the database connection. The cursor is used to execute SQL commands and navigate the database.
	conn_cursor.execute("""CREATE TABLE if not exists  malicious_workers_log ( 
	device_seq text, 
	if_malicious integer,
	correctly_identified_by text,
	incorrectly_identified_by text,
	in_round integer,
	when_resyncing text
	)""") #This line executes an SQL command to create a table named malicious_workers_log if it doesn't already exist. 

	# VBFL starts here
	for comm_round in range(latest_round_num + 1, args['max_num_comm']+1):
		# create round specific log folder
		log_files_folder_path_comm_round = f"{log_files_folder_path}/comm_{comm_round}"
		if os.path.exists(log_files_folder_path_comm_round):
			print(f"Deleting {log_files_folder_path_comm_round} and create a new one.")
			shutil.rmtree(log_files_folder_path_comm_round) #shutil.rmtree() is used torecursively remove a directory and all its contents
		os.mkdir(log_files_folder_path_comm_round)
		# free cuda memory
		if dev == torch.device("cuda"):
			with torch.cuda.device('cuda'):
				torch.cuda.empty_cache()
		print(f"\nCommunication round {comm_round}")
		comm_round_start_time = time.time()
		# (RE)ASSIGN ROLES 分配角色
		workers_to_assign = workers_needed
		miners_to_assign = miners_needed #TODO 优化目标：worker和miner的比例
		workers_this_round = []
		miners_this_round = []
		random.shuffle(devices_list)
		for device in devices_list:
			if workers_to_assign:
				device.assign_worker_role()
				workers_to_assign -= 1
			elif miners_to_assign:
				device.assign_miner_role()
				miners_to_assign -= 1
			else:
				device.assign_role()
			if device.return_role() == 'worker':
				workers_this_round.append(device)
			else:
				miners_this_round.append(device)
			# determine if online at the beginning (essential for step 1 when worker needs to associate with an online device)
			device.online_switcher()

		''' DEBUGGING CODE '''
		if args['verbose']:

			# show devices initial chain length and if online
			for device in devices_list:
				if device.is_online():
					print(f'{device.return_idx()} {device.return_role()} online - ', end='')
				else:
					print(f'{device.return_idx()} {device.return_role()} offline - ', end='')
				# debug chain length
				print(f"chain length {device.return_blockchain_object().return_chain_length()}")
		
			# show device roles
			print(f"\nThere are {len(workers_this_round)} workers and {len(miners_this_round)} miners in this round.")
			print("\nworkers this round are")
			for worker in workers_this_round:
				print(f"d_{worker.return_idx().split('_')[-1]} online - {worker.is_online()} with chain len {worker.return_blockchain_object().return_chain_length()}")
			print("\nminers this round are")
			for miner in miners_this_round:
				print(f"d_{miner.return_idx().split('_')[-1]} online - {miner.is_online()} with chain len {miner.return_blockchain_object().return_chain_length()}")
			print()

			# show peers with round number
			print(f"+++++++++ Round {comm_round} Beginning Peer Lists +++++++++")
			for device_seq, device in devices_in_network.devices_set.items():
				peers = device.return_peers()
				print(f"d_{device_seq.split('_')[-1]} - {device.return_role()[0]} has peer list ", end='')
				for peer in peers:
					print(f"d_{peer.return_idx().split('_')[-1]} - {peer.return_role()[0]}", end=', ')
				print()
			print(f"+++++++++ Round {comm_round} Beginning Peer Lists +++++++++")

		''' DEBUGGING CODE ENDS '''

		# re-init round vars - 变量重置
		#in real distributed system, they could still fall behind in comm round, but here we assume they will all go into the next round together, 
		#thought device may go offline somewhere in the previous round and their variables were not therefore reset
		for miner in miners_this_round:
			miner.miner_reset_vars_for_new_round()
		for worker in workers_this_round:
			worker.worker_reset_vars_for_new_round()

		# DOESN'T MATTER ANY MORE AFTER TRACKING TIME, but let's keep it - orginal purpose: shuffle the list(for worker, this will affect the order of dataset portions to be trained)
		random.shuffle(workers_this_round)
		random.shuffle(miners_this_round)
		
		''' workers and miners take turns to perform jobs '''

		print(''' Step 0 - workers assign associated miners.\n''')
		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			# resync chain(block could be dropped due to fork from last round)
			if worker.resync_chain(mining_consensus):
				worker.update_model_after_chain_resync(log_files_folder_path_comm_round, conn, conn_cursor)
			# worker (should) perform local update and associate
			print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} will associate with miners, if online...")
			# worker associates with a miner to accept finally mined block
			if worker.online_switcher():
				associated_miners = worker.associate_with_miner()
				if associated_miners:
					for associated_miner in associated_miners:
						associated_miner.add_device_to_association(worker)
				else:
					print(f"Cannot find a qualified miner in {worker.return_idx()} peer list.")

		print(''' Step 1 - workers do local updates.\n''')
		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			if worker.online_switcher():
				print(f'worker {worker_iter+1}/{len(workers_this_round)} is online and will do local updates...')
				worker.worker_local_update(rewards, log_files_folder_path_comm_round, comm_round, local_epochs=args['default_local_epochs'])#get worker.local_update_time, worker.local_train_parameters, worker.local_updates_rewards_per_transaction
				worker.return_local_updates_and_signature(comm_round) #get worker.local_updates_dict, i.e., local update transaction 
			else:
				print(f"worker {worker.return_idx()} offline and unable do local updates")

		print(''' Step 2 - miners accept local updates and broadcast to other miners in their respective peer lists.\n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			# resync chain
			if miner.resync_chain(mining_consensus):
				miner.update_model_after_chain_resync(log_files_folder_path, conn, conn_cursor)
			# miner accepts local updates from its associated workers 
			associated_workers = list(miner.return_associated_workers())
			if not associated_workers:
				print(f"No workers are associated with miner {miner.return_idx()} {miner_iter+1}/{len(miners_this_round)} for this communication round.")
				continue
			miner_link_speed = miner.return_link_speed()
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} is accepting workers' updates with link speed {miner_link_speed} bytes/s, and miner wait time is specified as {args['miner_acception_wait_time']} seconds.")			
			transaction_arrival_queue = {}# used for arrival time easy sorting for later miner broadcasting (and miners' acception order)
			for worker_iter in range(len(associated_workers)):
				worker = associated_workers[worker_iter]
				if not worker.return_idx() in miner.return_black_list():
					if worker.online_switcher():
						worker_link_speed = worker.return_link_speed()
						lower_link_speed = miner_link_speed if miner_link_speed < worker_link_speed else worker_link_speed
						transmission_delay = getsizeof(str(worker.local_updates_dict)) / lower_link_speed
						local_update_transaction_arrival_time = worker.local_update_time + transmission_delay
						if miner.online_switcher():
							if args['miner_acception_wait_time'] >= local_update_transaction_arrival_time:
								transaction_arrival_queue[local_update_transaction_arrival_time] = worker.local_updates_dict
								print(f"miner {miner.return_idx()} has accepted local update transaction from worker {worker.return_idx()}.")
						else:
							print(f"miner {miner.return_idx()} offline and unable to accept this transaction")
					else:
						print(f"worker {worker.return_idx()} offline and unable do local updates")
				else:
					print(f"worker {worker.return_idx()} in miner {miner.return_idx()}'s black list. This worker's transactions won't be accpeted.")
			miner.set_unordered_arrival_time_accepted_worker_transactions(transaction_arrival_queue)
			# in case miner off line for accepting broadcasted transactions but can later back online to validate the transactions itself receives
			miner.set_transaction_for_final_validating_queue(sorted(transaction_arrival_queue.items())) #MAYBE HAVE PROBLEM HERE
			
			# broadcast to other miners
			if transaction_arrival_queue:
				miner.miner_broadcast_worker_transactions()
			else:
				print("No transactions have been received by this miner, probably due to workers and/or miners offline or timeout while doing local updates or transmitting updates, or all workers are in miner's black list.")


		print(''' Step 2.5 - with the broadcasted workers transactions, miners decide the final transaction arrival order \n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			accepted_broadcasted_miner_transactions = miner.return_accepted_broadcasted_worker_transactions() #[ {'source_miner_link_speed': ,'broadcasted_transactions': self.unordered_arrival_time_accepted_worker_transactions}]
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} is calculating the final transactions arrival order by combining the direct received worker transactions and received broadcasted transactions...")			
			#Calculate the arrival time of broadcasted transactions by considering the transmission delay based on the link speed between the broadcasting miner and the current miner.
			accepted_broadcasted_transactions_arrival_queue = {}
			if accepted_broadcasted_miner_transactions:
				self_miner_link_speed = miner.return_link_speed()
				for broadcasting_miner_record in accepted_broadcasted_miner_transactions:
					broadcasting_miner_link_speed = broadcasting_miner_record['source_miner_link_speed']
					lower_link_speed = self_miner_link_speed if self_miner_link_speed < broadcasting_miner_link_speed else broadcasting_miner_link_speed
					for arrival_time_at_broadcasting_miner, broadcasted_transaction in broadcasting_miner_record['broadcasted_transactions'].items():
						transmission_delay = getsizeof(str(broadcasted_transaction))/lower_link_speed
						accepted_broadcasted_transactions_arrival_queue[transmission_delay + arrival_time_at_broadcasting_miner] = broadcasted_transaction
			else:
				print(f"miner {miner.return_idx()} {miner_iter+1}/{len(miners_this_round)} did not receive any broadcasted worker transaction this round.")
			
			# mix the boardcasted transactions with the direct accepted transactions
			final_transactions_arrival_queue = sorted({**miner.return_unordered_arrival_time_accepted_worker_transactions(), **accepted_broadcasted_transactions_arrival_queue}.items()) 
			# Set the final transaction queue for the miner
			miner.set_transaction_for_final_validating_queue(final_transactions_arrival_queue)
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} done calculating the ordered final transactions arrival order. Total {len(final_transactions_arrival_queue)} accepted transactions need to be validated.")

		print(''' Step 3 - miners do self and cross-validation(validate local updates from workers) by the order of transaction arrival time.\n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			final_transactions_arrival_queue = miner.return_final_transactions_validating_queue()
			if final_transactions_arrival_queue:
				# miner asynchronously does one epoch of update and validate on its own test set
				local_validation_time = miner.miner_update_model_by_one_epoch_and_validate_local_accuracy(args['optimizer']) #after this line, "self.miner_local_accuracy" can be get
				print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} is validating received worker transactions...")
				for (arrival_time, unconfirmmed_transaction) in final_transactions_arrival_queue:
					if miner.online_switcher():
						# validation won't begin until miner locally done one epoch of update and validation(worker transactions will be queued)
						if arrival_time < local_validation_time:
							arrival_time = local_validation_time
						#validate
						validation_time, post_validation_unconfirmmed_transaction = miner.validate_worker_transaction(unconfirmmed_transaction, rewards, log_files_folder_path, comm_round, args['malicious_miner_on'])
						if validation_time:
							miner.add_post_validation_transaction_to_queue((arrival_time + validation_time, miner.return_link_speed(), post_validation_unconfirmmed_transaction)) #three-metric tuple
							print(f"A validation process has been done for the transaction from worker {post_validation_unconfirmmed_transaction['worker_device_idx']} by miner {miner.return_idx()}")
					else:
						print(f"A validation process is skipped for the transaction from worker {post_validation_unconfirmmed_transaction['worker_device_idx']} by miner {miner.return_idx()} due to miner offline.")
				print(f"The validation process has been done for total {len(final_transactions_arrival_queue)} transactions by miner {miner.return_idx()}.")			
			else:
				print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} did not receive any transaction from worker or miner in this round.")

		print(''' Step 4 - miners aggregate their candidate models using the validated local updates from workers.\n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			if miner.online_switcher():
					post_validation_transactions_by_miner = miner.return_post_validation_transactions_queue()
					local_params_used_by_miner = miner.return_local_params_used_by_miner(post_validation_transactions_by_miner)
					miner.set_local_updates_used_info_by_miner(post_validation_transactions_by_miner)
					miner.aggregate_candidate_model(local_params_used_by_miner, rewards, log_files_folder_path_comm_round, comm_round)#get miner.aggregate_time, miner.candidate_parameters, miner.aggregate_rewards
					miner.return_candidate_model_and_signature(comm_round) #get miner.candidate_model_dict
			else:
				print(f"miner {miner.return_idx()} {miner_iter+1}/{len(miners_this_round)} is offline and unable to aggregate the candidate model.")

		print(''' Step 5 - workers accept candidate models and broadcast to other workers in their respective peer lists.\n''')
		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			associated_miners = list(worker.return_associated_miners())
			if not associated_miners:
				print(f"No miners are associated with worker {worker.return_idx()} for this communication round.")
				continue
			worker_link_speed = worker.return_link_speed()
			print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} is accepting miners' candidate models with link speed {worker_link_speed} bytes/s, if online...")
			# used for arrival time easy sorting for later worker broadcasting (and acception order)
			candadite_arrival_queue = {}
			print(f"worker wait time is specified as {args['worker_acception_wait_time']} seconds. let each miner aggregate_candidate_model till time limit")
			for miner_iter in range(len(associated_miners)):
				miner = associated_miners[miner_iter]
				if not miner.return_idx() in worker.return_black_list():						
					miner_link_speed = miner.return_link_speed() 
					lower_link_speed = worker_link_speed if worker_link_speed < miner_link_speed else miner_link_speed
					if miner.online_switcher():
						candidate_transmission_delay = getsizeof(str(miner.candidate_model_dict))/lower_link_speed
						candidate_arrival_time = candidate_transmission_delay + miner.aggregate_time
						if candidate_arrival_time > args['worker_acception_wait_time']:
							print(f'{miner.return_idx()}-miner candidate model arrival time exceeds the worker waiting time')
							break
						if worker.online_switcher():
							# accept this transaction only if the worker is online
							print(f"Worker {worker.return_idx()} has accepted this candidate.")
							candadite_arrival_queue[candidate_arrival_time] = miner.candidate_model_dict
						else:
							print(f"Worker {worker.return_idx()} offline and unable to accept this candidate")
					else:
						print(f"miner {miner.return_idx()} offline and unable aggregate candidate model")
				else:
					print(f"miner {miner.return_idx()} in worker {worker.return_idx()}'s black list. This miner's transactions won't be accpeted.")

			worker.set_unordered_arrival_time_accepted_miner_candidate(candadite_arrival_queue) #dict of arrival_time and candidate_model_dict
			# in case worker off line for accepting broadcasted transactions but can later back online to validate the transactions itself receives
			worker.set_candidate_for_final_validating_queue(sorted(candadite_arrival_queue.items())) #MAYBE HAVE PROBLEM HERE

			# broadcast to other workers
			if candadite_arrival_queue:
				worker.worker_broadcast_miner_candidate() #TODO 
			else:
				print("No transactions have been received by this worker, probably due to workers and/or miners offline or timeout while doing local updates or transmitting updates, or all miners are in worker's black list.")


		print(''' Step 5.5 - with the broadcasted miners candidate models, workers decide the final arrival order.\n ''')
		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			accepted_broadcasted_miner_candidate = worker.return_accepted_broadcasted_miner_candidate()
			print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} is calculating the final candidate arrival order by combining the direct miner candidate received and received broadcasted candidate...")
			#Calculate the arrival time of broadcasted candidate by considering the transmission delay based on the link speed between the broadcasting worker and the current worker.
			accepted_broadcasted_candidate_arrival_queue = {}
			if accepted_broadcasted_miner_candidate:
				self_worker_link_speed = worker.return_link_speed()
				for broadcasting_worker_record in accepted_broadcasted_miner_candidate:
					broadcasting_worker_link_speed = broadcasting_worker_record['source_worker_link_speed']
					lower_link_speed = self_worker_link_speed if self_worker_link_speed < broadcasting_worker_link_speed else broadcasting_worker_link_speed
					for arrival_time_at_broadcasting_worker, broadcasted_transaction in broadcasting_worker_record['broadcasted_candidate'].items():
						transmission_delay = getsizeof(str(broadcasted_transaction))/lower_link_speed
						accepted_broadcasted_candidate_arrival_queue[transmission_delay + arrival_time_at_broadcasting_worker] = broadcasted_transaction
			else:
				print(f"worker {worker.return_idx()} {worker_iter+1}/{len(workers_this_round)} did not receive any broadcasted miner transaction this round.")			
			# mix the boardcasted candidate with the direct accepted candidate
			final_candidate_arrival_queue = sorted({**worker.return_unordered_arrival_time_accepted_miner_candidate(), **accepted_broadcasted_candidate_arrival_queue}.items()) 
			# Set the final transaction queue for the worker
			worker.set_candidate_for_final_validating_queue(final_candidate_arrival_queue)
			print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} done calculating the ordered final candidate arrival order. Total {len(final_candidate_arrival_queue)} accepted candidate need to be validated.")

		
		print(''' Step 6 - workers verify miners' signature and miners' candidate models by the order of transaction arrival time.\n''')
		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			final_candidate_arrival_queue = worker.return_final_candidate_validating_queue()
			if final_candidate_arrival_queue:
				print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} is validating received miner candidate models...")
				for (arrival_time, unconfirmmed_candidate) in final_candidate_arrival_queue:
					if worker.online_switcher():
						candidate_validation_time, post_validation_candidate = worker.validate_miner_candidate(unconfirmmed_candidate, rewards, log_files_folder_path, comm_round, args['validate_threshold'], args['malicious_worker_on'])
						if candidate_validation_time:
							worker.add_post_validation_candidate_to_queue((arrival_time + candidate_validation_time, worker.return_link_speed(), post_validation_candidate))
							print(f"A validation process has been done for the transaction from miner {post_validation_candidate['miner_idx']} by worker {worker.return_idx()}")			
					else:
						print(f"A candidate validation process is skipped from miner {post_validation_candidate['miner_idx']} by worker {worker.return_idx()} due to worker offline.")
			else:
				print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} did not receive any candidate from worker or miner in this round.")
        

		print(''' Step 7 - workers send post validation candidate transactions to associated miner and miner broadcasts these to other miners in their respecitve peer lists.\n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			# resync chain
			if miner.resync_chain(mining_consensus):
				miner.update_model_after_chain_resync(log_files_folder_path, conn, conn_cursor)
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} accepting workers' post-validation candidate transactions...")
			associated_workers = list(miner.return_associated_workers())
			if not associated_workers:
				print(f"No workers are associated with miner {miner.return_idx()} for this communication round.")
				continue
			self_miner_link_speed = miner.return_link_speed()
			worker_validated_candidate_transactions_arrival_queue = {}
			for worker_iter in range(len(associated_workers)):
				worker = associated_workers[worker_iter]
				print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(associated_workers)} of miner {miner.return_idx()} is sending signature verified transaction...")
				post_validation_candidates_queue_by_worker = worker.return_post_validation_candidate_queue()
				post_validation_unconfirmmed_candidate_transaction_iter = 1
				for (worker_sending_time, source_worker_link_spped, post_validation_unconfirmmed_candidate_transaction) in post_validation_candidates_queue_by_worker:
					if worker.online_switcher() and miner.online_switcher():
						lower_link_speed = self_miner_link_speed if self_miner_link_speed < source_worker_link_spped else source_worker_link_spped
						transmission_delay = getsizeof(str(post_validation_unconfirmmed_candidate_transaction))/lower_link_speed
						worker_validated_candidate_transactions_arrival_queue[worker_sending_time + transmission_delay] = post_validation_unconfirmmed_candidate_transaction
						print(f"miner {miner.return_idx()} has accepted {post_validation_unconfirmmed_candidate_transaction_iter}/{len(post_validation_candidates_queue_by_worker)} post-validation candidate transaction from worker {worker.return_idx()}")
					else:
						print(f"miner {miner.return_idx()} has not accepted {post_validation_unconfirmmed_candidate_transaction_iter}/{len(post_validation_candidates_queue_by_worker)} post-validation candidate transaction from worker {worker.return_idx()} due to one of devices or both offline.")
					post_validation_unconfirmmed_candidate_transaction_iter += 1
			miner.set_unordered_arrival_time_accepted_worker_validated_candidate_transactions(worker_validated_candidate_transactions_arrival_queue)
			# in case worker off line for accepting broadcasted transactions but can later back online to validate the transactions itself receives
			worker.set_candidate_transactions_for_final_mining_queue(sorted(worker_validated_candidate_transactions_arrival_queue.items()))
			if worker_validated_candidate_transactions_arrival_queue:
				miner.miner_broadcast_worker_validated_candidate_transactions()
			else:
				print("No post-validation candidate transactions have been received by this miner.")

			
			
		print(''' Step 7.5 - with the broadcasted validated candidate transactions, miners decide the final candidate transaction arrival order.\n ''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			accepted_broadcasted_worker_validated_candidate_transactions = miner.return_accepted_miner_broadcasted_worker_validated_candidate_transactions()
			self_miner_link_speed = miner.return_link_speed()
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} calculating the final transactions arrival order by combining the direct accepted worker candidate transactions received and received broadcasted candidate transactions...")
			accepted_broadcasted_candidate_transactions_arrival_queue = {}
			if accepted_broadcasted_worker_validated_candidate_transactions:
				# calculate broadcasted transactions arrival time
				for broadcasting_miner_record in accepted_broadcasted_worker_validated_candidate_transactions:
					broadcasting_miner_link_speed = broadcasting_miner_record['source_miner_link_speed']
					lower_link_speed = self_miner_link_speed if self_miner_link_speed < broadcasting_miner_link_speed else broadcasting_miner_link_speed
					for arrival_time_at_broadcasting_miner, broadcasted_candidate_transaction in broadcasting_miner_record['broadcasted_validated_candidate_transactions'].items():
						transmission_delay = getsizeof(str(broadcasted_candidate_transaction))/lower_link_speed
						accepted_broadcasted_candidate_transactions_arrival_queue[transmission_delay + arrival_time_at_broadcasting_miner] = broadcasted_candidate_transaction
			else:
				print(f"miner {miner.return_idx()} {miner_iter+1}/{len(miners_this_round)} did not receive any broadcasted worker candidate transaction this round.")
			# mix the boardcasted transactions with the direct accepted transactions
			final_candidate_transactions_arrival_queue = sorted({**miner.return_unordered_arrival_time_accepted_worker_validated_candidate_transactions(), **accepted_broadcasted_candidate_transactions_arrival_queue}.items())
			miner.set_candidate_transactions_for_final_mining_queue(final_candidate_transactions_arrival_queue)
			print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} done calculating the ordered final candidate transactions arrival order. Total {len(final_candidate_transactions_arrival_queue)} accepted candidate transactions (There are duplicates).")


		print(''' Step 8 - miners do self and cross-verification (verify signature) by the order of transaction arrival time and record the transactions in the candidate block according to the limit size. Also mine and propagate the block.\n''')
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			final_candidate_transactions_arrival_queue = miner.return_final_candidate_transactions_mining_queue()
			valid_worker_sig_candidate_transacitons = []
			invalid_worker_sig_candidate_transacitons = []
			begin_mining_time = 0
			new_begin_mining_time = begin_mining_time
			if final_candidate_transactions_arrival_queue:
				print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} is verifying received worker candidate transactions...")
				time_limit = miner.return_miner_acception_wait_time()
				size_limit = miner.return_miner_accepted_transactions_size_limit()
				for (arrival_time, unconfirmmed_candidate_transaction) in final_candidate_transactions_arrival_queue:
					if miner.online_switcher():
						if time_limit:
							if arrival_time > time_limit:
								break
						if size_limit:
							if getsizeof(str(valid_worker_sig_candidate_transacitons+invalid_worker_sig_candidate_transacitons)) > size_limit:
								break
						# verify worker signature of this transaction
						verification_time, is_worker_sig_valid = miner.verify_worker_transaction(unconfirmmed_candidate_transaction)
						if verification_time:
							if is_worker_sig_valid:
								this_candidate_tx_info = {
								'validation_done_by_worker': unconfirmmed_candidate_transaction['validation_done_by'],
								'validation_reward_for_worker': unconfirmmed_candidate_transaction['validation_rewards'],
								'validation_time_by_worker': unconfirmmed_candidate_transaction['validation_time'],
								'worker_rsa_pub_key': unconfirmmed_candidate_transaction['worker_rsa_pub_key'],
								'worker_signature': unconfirmmed_candidate_transaction['worker_signature'],
								'candidate_direction': unconfirmmed_candidate_transaction['candidate_direction'],
								'candidate_model_accuracy': unconfirmmed_candidate_transaction['candidate_validation_accuracy'],
								'miner_device_idx': miner.return_idx(),
								'miner_verification_time': verification_time,
								'miner_verification_rewards_for_this_tx': rewards
								}
								# worker's transaction signature valid
								found_same_candidate_transaction = False
								for valid_worker_sig_candidate_transaciton in valid_worker_sig_candidate_transacitons:
									if valid_worker_sig_candidate_transaciton['miner_idx'] == unconfirmmed_candidate_transaction['miner_idx']:
										found_same_candidate_transaction = True
										break
								if not found_same_candidate_transaction:
									valid_worker_sig_candidate_transaciton = copy.deepcopy(unconfirmmed_candidate_transaction)
									del valid_worker_sig_candidate_transaciton['validation_done_by']
									del valid_worker_sig_candidate_transaciton['validation_rewards']
									del valid_worker_sig_candidate_transaciton['validation_time']
									del valid_worker_sig_candidate_transaciton['worker_rsa_pub_key']
									del valid_worker_sig_candidate_transaciton['worker_signature']
									del valid_worker_sig_candidate_transaciton['candidate_direction']
									del valid_worker_sig_candidate_transaciton['candidate_validation_accuracy']
									valid_worker_sig_candidate_transaciton['supported_workers'] = []
									valid_worker_sig_candidate_transaciton['opposed_workers'] = []
									valid_worker_sig_candidate_transacitons.append(valid_worker_sig_candidate_transaciton)
								if unconfirmmed_candidate_transaction['candidate_direction']:
									valid_worker_sig_candidate_transaciton['supported_workers'].append(this_candidate_tx_info)
								else:
									valid_worker_sig_candidate_transaciton['opposed_workers'].append(this_candidate_tx_info)
								transaction_to_sign = valid_worker_sig_candidate_transaciton
							else:
								# worker's transaction signature invalid
								invalid_worker_sig_candidate_transaciton = copy.deepcopy(unconfirmmed_candidate_transaction)
								invalid_worker_sig_candidate_transaciton['miner_verification_time'] = verification_time
								invalid_worker_sig_candidate_transaciton['miner_verification_rewards_for_this_tx'] = rewards
								invalid_worker_sig_candidate_transacitons.append(invalid_worker_sig_candidate_transaciton)
								transaction_to_sign = invalid_worker_sig_candidate_transaciton
							signing_time = miner.sign_candidate_transaction(transaction_to_sign)
							new_begin_mining_time = arrival_time + verification_time + signing_time							
					else:
						print(f"A verification process is skipped for the candidate transaction from worker {unconfirmmed_candidate_transaction['validation_done_by']} by miner {miner.return_idx()} due to miner offline.")
						new_begin_mining_time = arrival_time
					begin_mining_time = new_begin_mining_time if new_begin_mining_time > begin_mining_time else begin_mining_time


				# confirm the Leader in this round
				leader_idx, max_candidate_model_accuracy, sorted_valid_sig_transacitons = miner.find_leader_and_max_accuracy_among_valid_candidate_transacitons(valid_worker_sig_candidate_transacitons)
				if miner.return_idx() == leader_idx:
					if miner.online_switcher():
						print(f"miner {miner.return_idx()} is the leader of this round with max candidate model accuracy {max_candidate_model_accuracy}.")
						transactions_to_record_in_block = {}
						transactions_to_record_in_block['valid_worker_sig_transacitons'] = sorted_valid_sig_transacitons
						transactions_to_record_in_block['invalid_worker_sig_transacitons'] = invalid_worker_sig_candidate_transacitons

						# put transactions into candidate block and begin mining
						# block index starts from 1
						start_time_point = time.time()
						candidate_block = Block(idx=miner.return_blockchain_object().return_chain_length()+1, transactions=transactions_to_record_in_block, miner_rsa_pub_key=miner.return_rsa_pub_key())

						# prepare the candidate block 
						miner_computation_power = miner.return_computation_power() 
						if not miner_computation_power:
							block_generation_time_spent = float('inf')
							miner.set_block_generation_time_point(float('inf'))
							print(f"{miner.return_idx()} - miner mines a block in INFINITE time...")
							continue
						recorded_transactions = candidate_block.return_transactions()
						if recorded_transactions['valid_worker_sig_transacitons'] or recorded_transactions['invalid_worker_sig_transacitons']:
							print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} mining the block...")
							# return the last block and add previous hash
							last_block = miner.return_blockchain_object().return_last_block()
							if last_block is None:
								# will mine the genesis block
								candidate_block.set_previous_block_hash(None)
							else:
								candidate_block.set_previous_block_hash(last_block.compute_hash(hash_entire_block=True))
							candidate_block.set_mined_by(miner.return_idx())
							miner.sign_block(candidate_block)
							current_hash = candidate_block.compute_hash()
							candidate_block.set_hash(current_hash)
							# record mining time
							block_generation_time_spent = (time.time() - start_time_point)/miner_computation_power
							miner.set_block_generation_time_point(block_generation_time_spent)
							print(f"{miner.return_idx()} - miner mines a block in {block_generation_time_spent} seconds.")
							# immediately propagate the block
							miner.propagated_the_block(miner.return_block_generation_time_point(), candidate_block)
						else:
							print("No transaction to mine for this block.")
							continue
					else:#TODO: handle Leader goes offline while mining a block
						print(f"Unfortunately, {miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} goes offline while mining a block. This if-successful-mined block is not propagated.")
				else:
					#miner {miner.return_idx()} is not the leader of this round
					miner.set_mined_rewards(0)
			else:
				print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} did not receive any transaction from worker or miner in this round.")

		
		print(''' Step 9 - miners decide if adding a propagated block or its own mined block as the legitimate block, and request its associated devices to download this block. \n''')
		forking_happened = False
		# comm_round_block_gen_time regarded as the time point when the winning miner mines its block, 
		#calculated from the beginning of the round. If there is forking in PoW or rewards info out of sync in PoS, 
    	#this time is the avg time point of all the appended time by any device
		comm_round_block_gen_time = []
		for miner_iter in range(len(miners_this_round)):
			miner = miners_this_round[miner_iter]
			unconfirmed_candidate_block = miner.return_unconfirmed_candidate_block()
			block_arrival_time = miner.return_unconfirmed_candidate_block_arrival_time()
			if unconfirmed_candidate_block:
				verified_block, verification_time = miner.verify_block(unconfirmed_candidate_block, unconfirmed_candidate_block.return_mined_by())
				if verified_block:
					if miner.online_switcher():
						added = miner.add_block(verified_block) #return True if added, False if not added
					else:
						print(f"Unfortunately, miner {miner.return_idx()} goes offline while adding this block to its chain.")
					if added: #or if miner.return_the_added_block():
						# requesting devices in its associations to download this block
						miner.request_to_download(verified_block, block_arrival_time + verification_time)
						break	
				miner.add_to_round_end_time(block_arrival_time + verification_time)
			else:
				print(f"{miner.return_idx()} - miner {miner_iter+1}/{len(miners_this_round)} does not receive a propagated block and has not mined its own block yet.")

		for worker_iter in range(len(workers_this_round)):
			worker = workers_this_round[worker_iter]
			ordered_downloaded_block_processing_queue = sorted(worker.return_unordered_downloaded_block_processing_queue().items())
			if ordered_downloaded_block_processing_queue:
				print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} processing the downloaded block...")
				for (arrival_time, downloaded_block) in ordered_downloaded_block_processing_queue:
					if worker.online_switcher():
						verified_block, verification_time = worker.verify_block(downloaded_block, downloaded_block.return_mined_by())
						if verified_block:
							worker.add_block(verified_block)
							worker.set_block_download_time(arrival_time + verification_time)
							break
					else:
						print(f"Unfortunately, worker {worker.return_idx()} goes offline while processing this downloaded block.")
						worker.set_block_download_time(arrival_time)
						break
			else:
				print(f"{worker.return_idx()} - worker {worker_iter+1}/{len(workers_this_round)} does not receive any block.")
  
		# CHECK FOR FORKING
		added_blocks_miner_set = set()
		for device in devices_list:
			the_added_block = device.return_the_added_block()
			if the_added_block:
				print(f"{device.return_role()} {device.return_idx()} has added a block mined by {the_added_block.return_mined_by()}")
				added_blocks_miner_set.add(the_added_block.return_mined_by())
				block_generation_time_point = devices_in_network.devices_set[the_added_block.return_mined_by()].return_block_generation_time_point()
				# commented, as we just want to plot the legitimate block gen time, and the wait time is to avoid forking. 
				#Also the logic is wrong. Should track the time to the slowest worker after its global model update
				# if mining_consensus == 'PoS':
				# 	if args['miner_pos_propagated_block_wait_time'] != float("inf"):
				# 		block_generation_time_point += args['miner_pos_propagated_block_wait_time']
				comm_round_block_gen_time.append(block_generation_time_point)
		if len(added_blocks_miner_set) > 1:
			print("WARNING: a forking event just happened!")
			forking_happened = True
			with open(f"{log_files_folder_path}/forking_and_no_valid_block_log.txt", 'a') as file:
				file.write(f"Forking in round {comm_round}\n")
		else:
			print("No forking event happened.")
		
		print(''' Step 9.5 last step - process the added block - 1.collect usable candidate models\n 2.malicious nodes identification\n 3.get rewards\n This code block is skipped if no valid block was generated in this round''')
		all_devices_round_ends_time = []
		for device in devices_list:
			if device.return_the_added_block() and device.online_switcher():
				# collect usable updated params, malicious nodes identification, get rewards
				processing_time = device.process_block(device.return_the_added_block(), log_files_folder_path, conn, conn_cursor) ###
				device.other_tasks_at_the_end_of_comm_round(comm_round, log_files_folder_path)
				device.add_to_round_end_time(processing_time)
				all_devices_round_ends_time.append(device.return_round_end_time())

		print(''' Logging Accuracies by Devices ''')
		for device in devices_list:
			device.accuracy_this_round = device.validate_model_weights()
			with open(f"{log_files_folder_path_comm_round}/accuracy_comm_{comm_round}.txt", "a") as file:
				is_malicious_node = "M" if device.return_is_malicious() else "B"
				file.write(f"{device.return_idx()} {device.return_role()} {is_malicious_node}: {device.accuracy_this_round}\n")

		# logging time, mining_consensus and forking
		# get the slowest device end time
		comm_round_spent_time = time.time() - comm_round_start_time
		with open(f"{log_files_folder_path_comm_round}/accuracy_comm_{comm_round}.txt", "a") as file:
			# corner case when all miners in this round are malicious devices so their blocks are rejected
			try:
				comm_round_block_gen_time = max(comm_round_block_gen_time)
				file.write(f"comm_round_block_gen_time: {comm_round_block_gen_time}\n")
			except:
				no_block_msg = "No valid block has been generated this round."
				print(no_block_msg)
				file.write(f"comm_round_block_gen_time: {no_block_msg}\n")
				with open(f"{log_files_folder_path}/forking_and_no_valid_block_log.txt", 'a') as file2:
					# TODO this may be caused by "no transaction to mine" for the miner. Forgot to check for block miner's maliciousness in request_to_downlaod()
					file2.write(f"No valid block in round {comm_round}\n")
			try:
				slowest_round_ends_time = max(all_devices_round_ends_time)
				file.write(f"slowest_device_round_ends_time: {slowest_round_ends_time}\n")
			except:
				# corner case when all transactions are rejected by miners
				file.write("slowest_device_round_ends_time: No valid block has been generated this round.\n")
				with open(f"{log_files_folder_path}/forking_and_no_valid_block_log.txt", 'r+') as file2:
					no_valid_block_msg = f"No valid block in round {comm_round}\n"
					if file2.readlines()[-1] != no_valid_block_msg:
						file2.write(no_valid_block_msg)
			file.write(f"mining_consensus: {mining_consensus} {args['pow_difficulty']}\n")
			file.write(f"forking_happened: {forking_happened}\n")
			file.write(f"comm_round_spent_time_on_this_machine: {comm_round_spent_time}\n")
		conn.commit()

		# if no forking, log the block miner
		if not forking_happened:
			legitimate_block = None
			for device in devices_list:
				legitimate_block = device.return_the_added_block()
				if legitimate_block is not None:
					# skip the device who's been identified malicious and cannot get a block from miners
					break
			with open(f"{log_files_folder_path_comm_round}/accuracy_comm_{comm_round}.txt", "a") as file:
				if legitimate_block is None:
					file.write("block_mined_by: no valid block generated this round\n")
				else:
					block_mined_by = legitimate_block.return_mined_by()
					is_malicious_node = "M" if devices_in_network.devices_set[block_mined_by].return_is_malicious() else "B"
					file.write(f"block_mined_by: {block_mined_by} {is_malicious_node}\n")
		else:
			with open(f"{log_files_folder_path_comm_round}/accuracy_comm_{comm_round}.txt", "a") as file:
				file.write(f"block_mined_by: Forking happened\n")

		print(''' Logging Stake by Devices ''')
		for device in devices_list:
			device.accuracy_this_round = device.validate_model_weights()
			with open(f"{log_files_folder_path_comm_round}/stake_comm_{comm_round}.txt", "a") as file:
				is_malicious_node = "M" if device.return_is_malicious() else "B"
				file.write(f"{device.return_idx()} {device.return_role()} {is_malicious_node}: {device.return_stake()}\n")

		# a temporary workaround to free GPU mem by delete txs stored in the blocks. Not good when need to resync chain
		if args['destroy_tx_in_block']:
			for device in devices_list:
				last_block = device.return_blockchain_object().return_last_block()
				if last_block:
					last_block.free_tx()

		# save network_snapshot if reaches save frequency
		if args['save_network_snapshots'] and (comm_round == 1 or comm_round % args['save_freq'] == 0):
			if args['save_most_recent']:
				paths = sorted(Path(network_snapshot_save_path).iterdir(), key=os.path.getmtime)
				if len(paths) > args['save_most_recent']:
					for _ in range(len(paths) - args['save_most_recent']):
						# make it 0 byte as os.remove() moves file to the bin but may still take space
						# https://stackoverflow.com/questions/53028607/how-to-remove-the-file-from-trash-in-drive-in-colab
						open(paths[_], 'w').close() 
						os.remove(paths[_])
			snapshot_file_path = f"{network_snapshot_save_path}/snapshot_r_{comm_round}"
			print(f"Saving network snapshot to {snapshot_file_path}")
			pickle.dump(devices_in_network, open(snapshot_file_path, "wb"))