ProxLB/proxlb

#!/usr/bin/env python3

# ProxLB
# ProxLB (re)balances VM workloads across nodes in Proxmox clusters.
# ProxLB obtains current metrics from all nodes within the cluster for
# further auto balancing by memory, disk or cpu and rebalances the VMs
# over all available nodes in a cluster by having an equal resource usage.
# Copyright (C) 2024 Florian Paul Azim Hoberg @gyptazy <gyptazy@gyptazy.ch>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <https://www.gnu.org/licenses/>.

import argparse
import configparser
import json
import logging
import os
try:
    import proxmoxer
    _imports = True
except ImportError:
    _imports = False
import random
import re
import requests
import sys
import time
import urllib3


# Constants
__appname__ = "ProxLB"
__version__ = "0.9.9"
__author__  = "Florian Paul Azim Hoberg <gyptazy@gyptazy.ch> @gyptazy"
__errors__ = False


# Classes
## Logging class
class SystemdHandler(logging.Handler):
    """ Class to handle logging options. """
    PREFIX = {
        logging.CRITICAL: "<2> " + __appname__ + ": ",
        logging.ERROR: "<3> " + __appname__ + ": ",
        logging.WARNING: "<4> " + __appname__ + ": ",
        logging.INFO: "<6> " + __appname__ + ": ",
        logging.DEBUG: "<7> " + __appname__ + ": ",
        logging.NOTSET: "<7 " + __appname__ + ": ",
    }

    def __init__(self, stream=sys.stdout):
        self.stream = stream
        logging.Handler.__init__(self)

    def emit(self, record):
        try:
            msg = self.PREFIX[record.levelno] + self.format(record) + "\n"
            self.stream.write(msg)
            self.stream.flush()
        except Exception:
            self.handleError(record)


# Functions
def initialize_logger(log_level, update_log_verbosity=False):
    """ Initialize ProxLB logging handler. """
    info_prefix = 'Info: [logger]:'

    root_logger = logging.getLogger()
    root_logger.setLevel(log_level)

    if not update_log_verbosity:
        root_logger.addHandler(SystemdHandler())
        logging.info(f'{info_prefix} Logger got initialized.')
    else:
        logging.info(f'{info_prefix} Logger verbosity got updated to: {log_level}.')


def pre_validations(config_path):
    """ Run pre-validations as sanity checks. """
    info_prefix = 'Info: [pre-validations]:'

    __validate_imports()
    __validate_config_file(config_path)
    logging.info(f'{info_prefix} All pre-validations done.')


def post_validations():
    """ Run post-validations as sanity checks. """
    error_prefix = 'Error: [post-validations]:'
    info_prefix  = 'Info: [post-validations]:'

    if __errors__:
        logging.critical(f'{error_prefix} Not all post-validations succeeded. Please validate!')
    else:
        logging.info(f'{info_prefix} All post-validations succeeded.')


def validate_daemon(daemon, schedule):
    """ Validate if ProxLB runs as a daemon. """
    info_prefix  = 'Info: [daemon]:'

    if bool(int(daemon)):
        logging.info(f'{info_prefix} Running in daemon mode. Next run in {schedule} hours.')
        time.sleep(int(schedule) * 60)
    else:
        logging.info(f'{info_prefix} Not running in daemon mode. Quitting.')
        sys.exit(0)


def __validate_imports():
    """ Validate if all Python imports succeeded. """
    error_prefix = 'Error: [python-imports]:'
    info_prefix  = 'Info: [python-imports]:'

    if not _imports:
        logging.critical(f'{error_prefix} Could not import all dependencies. Please install "proxmoxer".')
        sys.exit(2)
    else:
        logging.info(f'{info_prefix} All required dependencies were imported.')


def __validate_config_file(config_path):
    """ Validate if all Python imports succeeded. """
    error_prefix = 'Error: [config]:'
    info_prefix  = 'Info: [config]:'

    if not os.path.isfile(config_path):
        logging.critical(f'{error_prefix} Could not find config file in: {config_path}.')
        sys.exit(2)
    else:
        logging.info(f'{info_prefix} Configuration file loaded from: {config_path}.')


def initialize_args():
    """ Initialize given arguments for ProxLB. """
    argparser = argparse.ArgumentParser(description='ProxLB')
    argparser.add_argument('-c', '--config',  type=str, help='Path to config file.', required=True)
    argparser.add_argument('-d', '--dry-run', help='Perform a dry-run without doing any actions.', action='store_true', required=False)
    argparser.add_argument('-j', '--json',    help='Return a JSON of the VM movement.', action='store_true', required=False)
    return argparser.parse_args()


def initialize_config_path(app_args):
    """ Initialize path to ProxLB config file. """
    info_prefix  = 'Info: [config]:'

    config_path = app_args.config
    if app_args.config is None:
        config_path = '/etc/proxlb/proxlb.conf'
        logging.info(f'{info_prefix} No config file provided. Falling back to: {config_path}.')
    else:
        logging.info(f'{info_prefix} Using config file: {config_path}.')
    return config_path


def initialize_config_options(config_path):
    """ Read configuration from given config file for ProxLB. """
    error_prefix = 'Error: [config]:'
    info_prefix  = 'Info: [config]:'

    try:
        config = configparser.ConfigParser()
        config.read(config_path)
        # Proxmox config
        proxmox_api_host  = config['proxmox']['api_host']
        proxmox_api_user  = config['proxmox']['api_user']
        proxmox_api_pass  = config['proxmox']['api_pass']
        proxmox_api_ssl_v = config['proxmox']['verify_ssl']
        # Balancing
        balancing_method  = config['balancing'].get('method', 'memory')
        balancing_mode    = config['balancing'].get('mode', 'used')
        balancing_type    = config['balancing'].get('type', 'vm')
        balanciness       = config['balancing'].get('balanciness', 10)
        ignore_nodes      = config['balancing'].get('ignore_nodes', None)
        ignore_vms        = config['balancing'].get('ignore_vms', None)
        # Service
        daemon            = config['service'].get('daemon', 1)
        schedule          = config['service'].get('schedule', 24)
        log_verbosity     = config['service'].get('log_verbosity', 'CRITICAL')
    except configparser.NoSectionError:
        logging.critical(f'{error_prefix} Could not find the required section.')
        sys.exit(2)
    except configparser.ParsingError:
        logging.critical(f'{error_prefix} Unable to parse the config file.')
        sys.exit(2)
    except KeyError:
        logging.critical(f'{error_prefix} Could not find the required options in config file.')
        sys.exit(2)

    logging.info(f'{info_prefix} Configuration file loaded.')
    return proxmox_api_host, proxmox_api_user, proxmox_api_pass, proxmox_api_ssl_v, balancing_method, \
         balancing_mode, balancing_type, balanciness, ignore_nodes, ignore_vms, daemon, schedule, log_verbosity


def api_connect(proxmox_api_host, proxmox_api_user, proxmox_api_pass, proxmox_api_ssl_v):
    """ Connect and authenticate to the Proxmox remote API. """
    error_prefix = 'Error: [api-connection]:'
    warn_prefix  = 'Warning: [api-connection]:'
    info_prefix  = 'Info: [api-connection]:'
    proxmox_api_ssl_v = bool(int(proxmox_api_ssl_v))

    if not proxmox_api_ssl_v:
        requests.packages.urllib3.disable_warnings()
        logging.warning(f'{warn_prefix} API connection does not verify SSL certificate.')

    try:
        api_object = proxmoxer.ProxmoxAPI(proxmox_api_host, user=proxmox_api_user, password=proxmox_api_pass, verify_ssl=proxmox_api_ssl_v)
    except urllib3.exceptions.NameResolutionError:
        logging.critical(f'{error_prefix} Could not resolve the given host: {proxmox_api_host}.')
        sys.exit(2)
    except requests.exceptions.ConnectTimeout:
        logging.critical(f'{error_prefix} Connection time out to host: {proxmox_api_host}.')
        sys.exit(2)
    except requests.exceptions.SSLError:
        logging.critical(f'{error_prefix} SSL certificate verification failed for host: {proxmox_api_host}.')
        sys.exit(2)

    logging.info(f'{info_prefix} API connection succeeded to host: {proxmox_api_host}.')
    return api_object


def get_node_statistics(api_object, ignore_nodes):
    """ Get statistics of cpu, memory and disk for each node in the cluster. """
    info_prefix       = 'Info: [node-statistics]:'
    node_statistics   = {}
    ignore_nodes_list =  ignore_nodes.split(',')

    for node in api_object.nodes.get():
        if node['status'] == 'online' and node['node'] not in ignore_nodes_list:
            node_statistics[node['node']] = {}
            node_statistics[node['node']]['cpu_total']                        = node['maxcpu']
            node_statistics[node['node']]['cpu_assigned']                     = node['cpu']
            node_statistics[node['node']]['cpu_assigned_percent']             = int((node_statistics[node['node']]['cpu_assigned']) / int(node_statistics[node['node']]['cpu_total']) * 100)
            node_statistics[node['node']]['cpu_assigned_percent_last_run']    = 0
            node_statistics[node['node']]['cpu_used']                         = 0
            node_statistics[node['node']]['cpu_free']                         = int(node['maxcpu']) - int(node['cpu'])
            node_statistics[node['node']]['cpu_free_percent']                 = int((node_statistics[node['node']]['cpu_free']) / int(node['maxcpu']) * 100)
            node_statistics[node['node']]['cpu_free_percent_last_run']        = 0
            node_statistics[node['node']]['memory_total']                     = node['maxmem']
            node_statistics[node['node']]['memory_assigned']                  = 0
            node_statistics[node['node']]['memory_assigned_percent']          = int((node_statistics[node['node']]['memory_assigned']) / int(node_statistics[node['node']]['memory_total']) * 100)
            node_statistics[node['node']]['memory_assigned_percent_last_run'] = 0
            node_statistics[node['node']]['memory_used']                      = node['mem']
            node_statistics[node['node']]['memory_free']                      = int(node['maxmem']) - int(node['mem'])
            node_statistics[node['node']]['memory_free_percent']              = int((node_statistics[node['node']]['memory_free']) / int(node['maxmem']) * 100)
            node_statistics[node['node']]['memory_free_percent_last_run']     = 0
            node_statistics[node['node']]['disk_total']                       = node['maxdisk']
            node_statistics[node['node']]['disk_assigned']                    = 0
            node_statistics[node['node']]['disk_assigned_percent']            = int((node_statistics[node['node']]['disk_assigned']) / int(node_statistics[node['node']]['disk_total']) * 100)
            node_statistics[node['node']]['disk_assigned_percent_last_run']   = 0
            node_statistics[node['node']]['disk_used']                        = node['disk']
            node_statistics[node['node']]['disk_free']                        = int(node['maxdisk']) - int(node['disk'])
            node_statistics[node['node']]['disk_free_percent']                = int((node_statistics[node['node']]['disk_free']) / int(node['maxdisk']) * 100)
            node_statistics[node['node']]['disk_free_percent_last_run']       = 0
            logging.info(f'{info_prefix} Added node {node["node"]}.')

    logging.info(f'{info_prefix} Created node statistics.')
    return node_statistics


def get_vm_statistics(api_object, ignore_vms, balancing_type):
    """ Get statistics of cpu, memory and disk for each vm in the cluster. """
    info_prefix        = 'Info: [vm-statistics]:'
    warn_prefix        = 'Warn: [vm-statistics]:'
    vm_statistics      = {}
    ignore_vms_list    =  ignore_vms.split(',')
    group_include      = None
    group_exclude      = None
    vm_ignore          = None
    vm_ignore_wildcard = False

    # Wildcard support: Initially validate if we need to honour
    # any wildcards within the vm_ignore list.
    vm_ignore_wildcard = __validate_ignore_vm_wildcard(ignore_vms)

    for node in api_object.nodes.get():

        # Add all virtual machines if type is vm or all.
        if balancing_type == 'vm' or balancing_type == 'all':
            for vm in api_object.nodes(node['node']).qemu.get():

                # Get the VM tags from API.
                vm_tags       = __get_vm_tags(api_object, node, vm['vmid'], 'vm')
                if vm_tags is not None:
                    group_include, group_exclude, vm_ignore = __get_proxlb_groups(vm_tags)

                # Get wildcard match for VMs to ignore if a wildcard pattern was
                # previously found. Wildcards may slow down the task when using
                # many patterns in the ignore list. Therefore, run this only if
                # a wildcard pattern was found. We also do not need to validate
                # this if the VM is already being ignored by a defined tag.
                if vm_ignore_wildcard and not vm_ignore:
                    vm_ignore = __check_vm_name_wildcard_pattern(vm['name'], ignore_vms_list)

                if vm['status'] == 'running' and vm['name'] not in ignore_vms_list and not vm_ignore:
                    vm_statistics[vm['name']] = {}
                    vm_statistics[vm['name']]['group_include']  = group_include
                    vm_statistics[vm['name']]['group_exclude']  = group_exclude
                    vm_statistics[vm['name']]['cpu_total']      = vm['cpus']
                    vm_statistics[vm['name']]['cpu_used']       = vm['cpu']
                    vm_statistics[vm['name']]['memory_total']   = vm['maxmem']
                    vm_statistics[vm['name']]['memory_used']    = vm['mem']
                    vm_statistics[vm['name']]['disk_total']     = vm['maxdisk']
                    vm_statistics[vm['name']]['disk_used']      = vm['disk']
                    vm_statistics[vm['name']]['vmid']           = vm['vmid']
                    vm_statistics[vm['name']]['node_parent']    = node['node']
                    vm_statistics[vm['name']]['type']           = 'vm'
                    # Rebalancing node will be overwritten after calculations.
                    # If the vm stays on the node, it will be removed at a
                    # later time.
                    vm_statistics[vm['name']]['node_rebalance'] = node['node']
                    logging.info(f'{info_prefix} Added vm {vm["name"]}.')

        # Add all containers if type is ct or all.
        if balancing_type == 'ct' or balancing_type == 'all':
            for vm in api_object.nodes(node['node']).lxc.get():

                logging.warning(f'{warn_prefix} Rebalancing on LXC containers (CT) always requires them to shut down.')
                logging.warning(f'{warn_prefix} {vm["name"]} is from type CT and cannot be live migrated!')
                # Get the VM tags from API.
                vm_tags       = __get_vm_tags(api_object, node, vm['vmid'], 'ct')
                if vm_tags is not None:
                    group_include, group_exclude, vm_ignore = __get_proxlb_groups(vm_tags)

                # Get wildcard match for VMs to ignore if a wildcard pattern was
                # previously found. Wildcards may slow down the task when using
                # many patterns in the ignore list. Therefore, run this only if
                # a wildcard pattern was found. We also do not need to validate
                # this if the VM is already being ignored by a defined tag.
                if vm_ignore_wildcard and not vm_ignore:
                    vm_ignore = __check_vm_name_wildcard_pattern(vm['name'], ignore_vms_list)

                if vm['status'] == 'running' and vm['name'] not in ignore_vms_list and not vm_ignore:
                    vm_statistics[vm['name']] = {}
                    vm_statistics[vm['name']]['group_include']  = group_include
                    vm_statistics[vm['name']]['group_exclude']  = group_exclude
                    vm_statistics[vm['name']]['cpu_total']      = vm['cpus']
                    vm_statistics[vm['name']]['cpu_used']       = vm['cpu']
                    vm_statistics[vm['name']]['memory_total']   = vm['maxmem']
                    vm_statistics[vm['name']]['memory_used']    = vm['mem']
                    vm_statistics[vm['name']]['disk_total']     = vm['maxdisk']
                    vm_statistics[vm['name']]['disk_used']      = vm['disk']
                    vm_statistics[vm['name']]['vmid']           = vm['vmid']
                    vm_statistics[vm['name']]['node_parent']    = node['node']
                    vm_statistics[vm['name']]['type']           = 'ct'
                    # Rebalancing node will be overwritten after calculations.
                    # If the vm stays on the node, it will be removed at a
                    # later time.
                    vm_statistics[vm['name']]['node_rebalance'] = node['node']
                    logging.info(f'{info_prefix} Added vm {vm["name"]}.')

    logging.info(f'{info_prefix} Created VM statistics.')
    return vm_statistics


def update_node_statistics(node_statistics, vm_statistics):
    """ Update node statistics by VMs statistics. """
    info_prefix = 'Info: [node-update-statistics]:'
    warn_prefix = 'Warning: [node-update-statistics]:'

    for vm, vm_value in vm_statistics.items():
        node_statistics[vm_value['node_parent']]['cpu_assigned']            = node_statistics[vm_value['node_parent']]['cpu_assigned'] + int(vm_value['cpu_total'])
        node_statistics[vm_value['node_parent']]['cpu_assigned_percent']    = (node_statistics[vm_value['node_parent']]['cpu_assigned'] / node_statistics[vm_value['node_parent']]['cpu_total']) * 100
        node_statistics[vm_value['node_parent']]['memory_assigned']         = node_statistics[vm_value['node_parent']]['memory_assigned'] + int(vm_value['memory_total'])
        node_statistics[vm_value['node_parent']]['memory_assigned_percent'] = (node_statistics[vm_value['node_parent']]['memory_assigned'] / node_statistics[vm_value['node_parent']]['memory_total']) * 100
        node_statistics[vm_value['node_parent']]['disk_assigned']           = node_statistics[vm_value['node_parent']]['disk_assigned'] + int(vm_value['disk_total'])
        node_statistics[vm_value['node_parent']]['disk_assigned_percent']   = (node_statistics[vm_value['node_parent']]['disk_assigned'] / node_statistics[vm_value['node_parent']]['disk_total']) * 100

        if node_statistics[vm_value['node_parent']]['cpu_assigned_percent'] > 99:
            logging.warning(f'{warn_prefix} Node {vm_value["node_parent"]} is overprovisioned for CPU by {int(node_statistics[vm_value["node_parent"]]["cpu_assigned_percent"])}%.')

        if node_statistics[vm_value['node_parent']]['memory_assigned_percent'] > 99:
            logging.warning(f'{warn_prefix} Node {vm_value["node_parent"]} is overprovisioned for memory by {int(node_statistics[vm_value["node_parent"]]["memory_assigned_percent"])}%.')

        if node_statistics[vm_value['node_parent']]['disk_assigned_percent'] > 99:
            logging.warning(f'{warn_prefix} Node {vm_value["node_parent"]} is overprovisioned for disk by {int(node_statistics[vm_value["node_parent"]]["disk_assigned_percent"])}%.')

    logging.info(f'{info_prefix} Updated node resource assignments by all VMs.')
    logging.debug('node_statistics')
    return node_statistics


def __validate_ignore_vm_wildcard(ignore_vms):
    """ Validate if a wildcard is used for ignored VMs. """
    if '*' in ignore_vms:
        return True


def __check_vm_name_wildcard_pattern(vm_name, ignore_vms_list):
    """ Validate if the VM name is in the ignore list pattern included. """
    for ignore_vm in ignore_vms_list:
        if '*' in ignore_vm:
            if ignore_vm[:-1] in vm_name:
                return True


def __get_vm_tags(api_object, node, vmid, balancing_type):
    """ Get tags for a VM/CT for a given VMID. """
    info_prefix = 'Info: [api-get-vm-tags]:'

    if balancing_type == 'vm':
        vm_config = api_object.nodes(node['node']).qemu(vmid).config.get()

    if balancing_type == 'ct':
        vm_config = api_object.nodes(node['node']).lxc(vmid).config.get()

    logging.info(f'{info_prefix} Got VM/CT tag from API.')
    return vm_config.get('tags', None)


def __get_proxlb_groups(vm_tags):
    """ Get ProxLB related include and exclude groups. """
    info_prefix   = 'Info: [api-get-vm-include-exclude-tags]:'
    group_include = None
    group_exclude = None
    vm_ignore     = None

    group_list = re.split(";", vm_tags)
    for group in group_list:

        if group.startswith('plb_include_'):
            logging.info(f'{info_prefix} Got PLB include group.')
            group_include = group

        if group.startswith('plb_exclude_'):
            logging.info(f'{info_prefix} Got PLB include group.')
            group_exclude = group

        if group.startswith('plb_ignore_vm'):
            logging.info(f'{info_prefix} Got PLB ignore group.')
            vm_ignore = True

    return group_include, group_exclude, vm_ignore


def balancing_calculations(balancing_method, balancing_mode, node_statistics, vm_statistics, balanciness, rebalance, processed_vms):
    """ Calculate re-balancing of VMs on present nodes across the cluster. """
    info_prefix  = 'Info: [rebalancing-calculator]:'

    # Validate for a supported balancing method, mode and if rebalancing is required.
    __validate_balancing_method(balancing_method)
    __validate_balancing_mode(balancing_mode)
    rebalance = __validate_balanciness(balanciness, balancing_method, balancing_mode, node_statistics)

    if rebalance:
        # Get most used/assigned resources of the VM and the most free or less allocated node.
        resources_vm_most_used, processed_vms  = __get_most_used_resources_vm(balancing_method, balancing_mode, vm_statistics, processed_vms)
        resources_node_most_free               = __get_most_free_resources_node(balancing_method, balancing_mode, node_statistics)

        # Update resource statistics for VMs and nodes.
        node_statistics, vm_statistics         = __update_resource_statistics(resources_vm_most_used, resources_node_most_free,
                                                                             vm_statistics, node_statistics, balancing_method, balancing_mode)

        # Start recursion until we do not have any needs to rebalance anymore.
        balancing_calculations(balancing_method, balancing_mode, node_statistics, vm_statistics, balanciness, rebalance, processed_vms)

    # Honour groupings for include and exclude groups for rebalancing VMs.
    node_statistics, vm_statistics = __get_vm_tags_include_groups(vm_statistics, node_statistics, balancing_method, balancing_mode)
    node_statistics, vm_statistics = __get_vm_tags_exclude_groups(vm_statistics, node_statistics, balancing_method, balancing_mode)

    # Remove VMs that are not being relocated.
    vms_to_remove = [vm_name for vm_name, vm_info in vm_statistics.items() if 'node_rebalance' in vm_info and vm_info['node_rebalance'] == vm_info.get('node_parent')]
    for vm_name in vms_to_remove:
        del vm_statistics[vm_name]

    logging.info(f'{info_prefix} Balancing calculations done.')
    return node_statistics, vm_statistics


def __validate_balancing_method(balancing_method):
    """ Validate for valid and supported balancing method. """
    error_prefix = 'Error: [balancing-method-validation]:'
    info_prefix  = 'Info: [balancing-method-validation]]:'

    if balancing_method not in ['memory', 'disk', 'cpu']:
        logging.error(f'{error_prefix} Invalid balancing method: {balancing_method}')
        sys.exit(2)
    else:
        logging.info(f'{info_prefix} Valid balancing method: {balancing_method}')


def __validate_balancing_mode(balancing_mode):
    """ Validate for valid and supported balancing mode. """
    error_prefix = 'Error: [balancing-mode-validation]:'
    info_prefix  = 'Info: [balancing-mode-validation]]:'

    if balancing_mode not in ['used', 'assigned']:
        logging.error(f'{error_prefix} Invalid balancing method: {balancing_mode}')
        sys.exit(2)
    else:
        logging.info(f'{info_prefix} Valid balancing method: {balancing_mode}')


def __validate_balanciness(balanciness, balancing_method, balancing_mode, node_statistics):
    """ Validate for balanciness to ensure further rebalancing is needed. """
    info_prefix                   = 'Info: [balanciness-validation]:'
    node_resource_percent_list    = []
    node_assigned_percent_match = []

    # Remap balancing mode to get the related values from nodes dict.
    if balancing_mode == 'used':
        node_resource_selector = 'free'
    if balancing_mode == 'assigned':
        node_resource_selector = 'assigned'

    for node_name, node_info in node_statistics.items():

        # Save information of nodes from current run to compare them in the next recursion.
        if node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent_last_run'] == node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent']:
            node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent_match'] = True
        else:
            node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent_match'] = False
        # Update value to the current value of the recursion run.
        node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent_last_run'] = node_statistics[node_name][f'{balancing_method}_{node_resource_selector}_percent']

        # If all node resources are unchanged, the recursion can be left.
        for key, value in node_statistics.items():
            node_assigned_percent_match.append(value.get(f'{balancing_method}_{node_resource_selector}_percent_match', False))

        if False not in node_assigned_percent_match:
            return False

        # Add node information to resource list.
        node_resource_percent_list.append(int(node_info[f'{balancing_method}_{node_resource_selector}_percent']))
        logging.debug(f'{info_prefix} Node: {node_name} with values: {node_info}')

    # Create a sorted list of the delta + balanciness between the node resources.
    node_resource_percent_list_sorted    = sorted(node_resource_percent_list)
    node_lowest_percent                  = node_resource_percent_list_sorted[0]
    node_highest_percent                 = node_resource_percent_list_sorted[-1]

    # Validate if the recursion should  be proceeded for further rebalancing.
    if (int(node_lowest_percent) + int(balanciness)) < int(node_highest_percent):
        logging.info(f'{info_prefix} Rebalancing for {balancing_method} is needed. Highest usage: {int(node_highest_percent)}% | Lowest usage: {int(node_lowest_percent)}%.')
        return True
    else:
        logging.info(f'{info_prefix} Rebalancing for {balancing_method} is not needed. Highest usage: {int(node_highest_percent)}% | Lowest usage: {int(node_lowest_percent)}%.')
        return False


def __get_most_used_resources_vm(balancing_method, balancing_mode, vm_statistics, processed_vms):
    """ Get and return the most used resources of a VM by the defined balancing method. """
    info_prefix = 'Info: [get-most-used-resources-vm]:'

    # Remap balancing mode to get the related values from nodes dict.
    if balancing_mode == 'used':
        vm_resource_selector = 'used'
    if balancing_mode == 'assigned':
        vm_resource_selector = 'total'

    vm = max(vm_statistics.items(), key=lambda item: item[1][f'{balancing_method}_{vm_resource_selector}'] if item[0] not in processed_vms else -float('inf'))
    processed_vms.append(vm[0])

    logging.info(f'{info_prefix} {vm}')
    return vm, processed_vms


def __get_most_free_resources_node(balancing_method, balancing_mode, node_statistics):
    """ Get and return the most free resources of a node by the defined balancing method. """
    info_prefix = 'Info: [get-most-free-resources-nodes]:'

    # Return the node information based on the balancing mode.
    if balancing_mode == 'used':
        node = max(node_statistics.items(), key=lambda item: item[1][f'{balancing_method}_free'])
    if balancing_mode == 'assigned':
        node = min(node_statistics.items(), key=lambda item: item[1][f'{balancing_method}_assigned'] if item[1][f'{balancing_method}_assigned_percent'] > 0 or item[1][f'{balancing_method}_assigned_percent'] < 100 else -float('inf'))

    logging.info(f'{info_prefix} {node}')
    return node


def __update_resource_statistics(resource_highest_used_resources_vm, resource_highest_free_resources_node, vm_statistics, node_statistics, balancing_method, balancing_mode):
    """ Update VM and node resource statistics. """
    info_prefix = 'Info: [rebalancing-resource-statistics-update]:'

    if resource_highest_used_resources_vm[1]['node_parent'] != resource_highest_free_resources_node[0]:
        vm_name            = resource_highest_used_resources_vm[0]
        vm_node_parent     = resource_highest_used_resources_vm[1]['node_parent']
        vm_node_rebalance  = resource_highest_free_resources_node[0]
        vm_resource_used   = vm_statistics[resource_highest_used_resources_vm[0]][f'{balancing_method}_used']
        vm_resource_total  = vm_statistics[resource_highest_used_resources_vm[0]][f'{balancing_method}_total']

        # Update dictionaries for new values
        # Assign new rebalance node to vm
        vm_statistics[vm_name]['node_rebalance'] = vm_node_rebalance

        logging.info(f'Moving {vm_name} from {vm_node_parent} to {vm_node_rebalance}')

        # Recalculate values for nodes
        ## Add freed resources to old parent node
        node_statistics[vm_node_parent][f'{balancing_method}_used']                = int(node_statistics[vm_node_parent][f'{balancing_method}_used']) - int(vm_resource_used)
        node_statistics[vm_node_parent][f'{balancing_method}_free']                = int(node_statistics[vm_node_parent][f'{balancing_method}_free']) + int(vm_resource_used)
        node_statistics[vm_node_parent][f'{balancing_method}_free_percent']        = int(int(node_statistics[vm_node_parent][f'{balancing_method}_free']) / int(node_statistics[vm_node_parent][f'{balancing_method}_total']) * 100)
        node_statistics[vm_node_parent][f'{balancing_method}_assigned']            = int(node_statistics[vm_node_parent][f'{balancing_method}_assigned']) - int(vm_resource_total)
        node_statistics[vm_node_parent][f'{balancing_method}_assigned_percent']    = int(int(node_statistics[vm_node_parent][f'{balancing_method}_assigned']) / int(node_statistics[vm_node_parent][f'{balancing_method}_total']) * 100)

        ## Removed newly allocated resources to new rebalanced node
        node_statistics[vm_node_rebalance][f'{balancing_method}_used']             = int(node_statistics[vm_node_rebalance][f'{balancing_method}_used']) + int(vm_resource_used)
        node_statistics[vm_node_rebalance][f'{balancing_method}_free']             = int(node_statistics[vm_node_rebalance][f'{balancing_method}_free']) - int(vm_resource_used)
        node_statistics[vm_node_rebalance][f'{balancing_method}_free_percent']     = int(int(node_statistics[vm_node_rebalance][f'{balancing_method}_free']) / int(node_statistics[vm_node_rebalance][f'{balancing_method}_total']) * 100)
        node_statistics[vm_node_rebalance][f'{balancing_method}_assigned']         = int(node_statistics[vm_node_rebalance][f'{balancing_method}_assigned']) + int(vm_resource_total)
        node_statistics[vm_node_rebalance][f'{balancing_method}_assigned_percent'] = int(int(node_statistics[vm_node_rebalance][f'{balancing_method}_assigned']) / int(node_statistics[vm_node_rebalance][f'{balancing_method}_total']) * 100)

    logging.info(f'{info_prefix} Updated VM and node statistics.')
    return node_statistics, vm_statistics


def __get_vm_tags_include_groups(vm_statistics, node_statistics, balancing_method, balancing_mode):
    """ Get VMs tags for include groups. """
    info_prefix = 'Info: [rebalancing-tags-group-include]:'
    tags_include_vms = {}
    processed_vm = []

    # Create groups of tags with belongings hosts.
    for vm_name, vm_values in vm_statistics.items():
        if vm_values.get('group_include', None):
            if not tags_include_vms.get(vm_values['group_include'], None):
                tags_include_vms[vm_values['group_include']] = [vm_name]
            else:
                tags_include_vms[vm_values['group_include']] = tags_include_vms[vm_values['group_include']] + [vm_name]

    # Update the VMs to the corresponding node to their group assignments.
    for group, vm_names in tags_include_vms.items():
        # Do not take care of tags that have only a single host included.
        if len(vm_names) < 2:
            logging.info(f'{info_prefix} Only one host in group assignment.')
            return node_statistics, vm_statistics

        vm_node_rebalance = False
        logging.info(f'{info_prefix} Create include groups of VM hosts.')
        for vm_name in vm_names:
            if vm_name not in processed_vm:
                if not vm_node_rebalance:
                    vm_node_rebalance = vm_statistics[vm_name]['node_rebalance']
                else:
                    _mocked_vm_object = (vm_name, vm_statistics[vm_name])
                    node_statistics, vm_statistics = __update_resource_statistics(_mocked_vm_object, [vm_node_rebalance], vm_statistics, node_statistics, balancing_method, balancing_mode)
            processed_vm.append(vm_name)

    return node_statistics, vm_statistics


def __get_vm_tags_exclude_groups(vm_statistics, node_statistics, balancing_method, balancing_mode):
    """ Get VMs tags for exclude groups. """
    info_prefix = 'Info: [rebalancing-tags-group-exclude]:'
    tags_exclude_vms = {}
    processed_vm = []

    # Create groups of tags with belongings hosts.
    for vm_name, vm_values in vm_statistics.items():
        if vm_values.get('group_include', None):
            if not tags_exclude_vms.get(vm_values['group_include'], None):
                tags_exclude_vms[vm_values['group_include']] = [vm_name]
            else:
                tags_exclude_vms[vm_values['group_include']] = tags_exclude_vms[vm_values['group_include']] + [vm_name]

    # Update the VMs to the corresponding node to their group assignments.
    for group, vm_names in tags_exclude_vms.items():
        # Do not take care of tags that have only a single host included.
        if len(vm_names) < 2:
            logging.info(f'{info_prefix} Only one host in group assignment.')
            return node_statistics, vm_statistics

        vm_node_rebalance = False
        logging.info(f'{info_prefix} Create exclude groups of VM hosts.')
        for vm_name in vm_names:
            if vm_name not in processed_vm:
                if not vm_node_rebalance:
                    random_node = vm_statistics[vm_name]['node_parent']
                    # Get a random node and make sure that it is not by accident the
                    # currently assigned one.
                    while random_node == vm_statistics[vm_name]['node_parent']:
                        random_node = random.choice(list(node_statistics.keys()))
                else:
                    _mocked_vm_object = (vm_name, vm_statistics[vm_name])
                    node_statistics, vm_statistics = __update_resource_statistics(_mocked_vm_object, [random_node], vm_statistics, node_statistics, balancing_method, balancing_mode)
            processed_vm.append(vm_name)

    return node_statistics, vm_statistics


def __run_vm_rebalancing(api_object, vm_statistics_rebalanced, app_args):
    """ Run & execute the VM rebalancing via API. """
    error_prefix = 'Error: [rebalancing-executor]:'
    info_prefix  = 'Info: [rebalancing-executor]:'

    if len(vm_statistics_rebalanced) > 0 and not app_args.dry_run:
        for vm, value in vm_statistics_rebalanced.items():
            try:
                logging.info(f'{info_prefix} Rebalancing vm {vm} from node {value["node_parent"]} to node {value["node_rebalance"]}.')
                api_object.nodes(value['node_parent']).qemu(value['vmid']).migrate().post(target=value['node_rebalance'],online=1)
            except proxmoxer.core.ResourceException as error_resource:
                logging.critical(f'{error_prefix} {error_resource}')
    else:
        logging.info(f'{info_prefix} No rebalancing needed.')


def __create_json_output(vm_statistics_rebalanced, app_args):
    """ Create a machine parsable json output of VM rebalance statitics. """
    info_prefix  = 'Info: [json-output-generator]:'

    if app_args.json:
        logging.info(f'{info_prefix} Printing json output of VM statistics.')
        print(json.dumps(vm_statistics_rebalanced))


def __create_dry_run_output(vm_statistics_rebalanced, app_args):
    """ Create output for CLI when running in dry-run mode. """
    info_prefix  = 'Info: [dry-run-output-generator]:'
    vm_to_node_list = []

    logging.info(f'{info_prefix} Starting dry-run to rebalance vms to their new nodes.')
    vm_to_node_list.append(['VM', 'Current Node', 'Rebalanced Node', 'VM Type'])
    for vm_name, vm_values in vm_statistics_rebalanced.items():
        vm_to_node_list.append([vm_name, vm_values['node_parent'], vm_values['node_rebalance'], vm_values['type']])

    if len(vm_statistics_rebalanced) > 0:
        logging.info(f'{info_prefix} Printing cli output of VM rebalancing.')
        __print_table_cli(vm_to_node_list)
    else:
        logging.info(f'{info_prefix} No rebalancing needed.')


def __print_table_cli(table):
    """ Pretty print a given table to the cli. """
    longest_cols = [
        (max([len(str(row[i])) for row in table]) + 3)
        for i in range(len(table[0]))
    ]

    row_format = "".join(["{:>" + str(longest_col) + "}" for longest_col in longest_cols])
    for row in table:
        print(row_format.format(*row))


def run_vm_rebalancing(api_object, vm_statistics_rebalanced, app_args):
    """ Run rebalancing of vms to new nodes in cluster. """
    __run_vm_rebalancing(api_object, vm_statistics_rebalanced, app_args)
    __create_json_output(vm_statistics_rebalanced, app_args)
    __create_dry_run_output(vm_statistics_rebalanced, app_args)


def main():
    """ Run ProxLB for balancing VM workloads across a Proxmox cluster. """
    # Initialize PAS.
    initialize_logger('CRITICAL')
    app_args = initialize_args()
    config_path = initialize_config_path(app_args)
    pre_validations(config_path)

    # Parse global config.
    proxmox_api_host, proxmox_api_user, proxmox_api_pass, proxmox_api_ssl_v, balancing_method, balancing_mode, balancing_type, \
        balanciness, ignore_nodes, ignore_vms, daemon, schedule, log_verbosity = initialize_config_options(config_path)

    # Overwrite logging handler with user defined log verbosity.
    initialize_logger(log_verbosity, update_log_verbosity=True)

    while True:
        # API Authentication.
        api_object = api_connect(proxmox_api_host, proxmox_api_user, proxmox_api_pass, proxmox_api_ssl_v)

        # Get metric & statistics for vms and nodes.
        node_statistics = get_node_statistics(api_object, ignore_nodes)
        vm_statistics   = get_vm_statistics(api_object, ignore_vms, balancing_type)
        node_statistics = update_node_statistics(node_statistics, vm_statistics)

        # Calculate rebalancing of vms.
        node_statistics_rebalanced, vm_statistics_rebalanced = balancing_calculations(balancing_method, balancing_mode, node_statistics, vm_statistics, balanciness, rebalance=False, processed_vms=[])

        # Rebalance vms to new nodes within the cluster.
        run_vm_rebalancing(api_object, vm_statistics_rebalanced, app_args)

        # Validate for any errors.
        post_validations()

        # Validate daemon service.
        validate_daemon(daemon, schedule)


if __name__ == '__main__':
    main()