powerapi/power_client.py

import math
import time

import requests
import socket
from killswitch import KillSwitch
from timber import LumberJack, axe
from config import _CLIENT_CONFIG as _c

lj = LumberJack(log_file='power_client.log',
                log_level='d')

class PowerClient:
    def __init__(self):
        self.DRYRUN = _c['dryrun']
        self.RETRY_ATTEMPTS = _c['retry']['attempts']
        self.RETRY_COUNT = 0
        self.config_check_count = 0
        self.HEALTHY_RETRY_SECONDS = _c['retry']['seconds']['healthy']
        self.UNHEALTHY_RETRY_SECONDS = _c['retry']['seconds']['unhealthy']
        self.RETRY_SECONDS = self.HEALTHY_RETRY_SECONDS
        self.config_check_cycles = _c['config_check']['cycles']
        self.power_server = _c['server']['address']
        self.server_port = _c['server']['port']
        self.api_base_endpoint = _c['api']['base_endpoint']
        self.hostname = self.get_hostname()
        self.api_config_endpoint = self.get_api_config_endpoint()

        self.config_url = self.get_config_url()
        self.url = self.get_url()

        # check for new config at init
        self.update_config()

    def get_config(self):
        c = requests.get(self.config_url)
        if c.status_code == 200:
            return c.json()
            # return json.loads(c.json())
        else:
            return None

    def update_config(self):
        axe.info('Checking for new configuration')
        c = self.get_config()
        if c:
            if self.DRYRUN != c['dryrun']:
                axe.info(f'New DRYRUN configuration: {c["dryrun"]}')
                self.DRYRUN = c['dryrun']

            if self.RETRY_ATTEMPTS != c['retry']['attempts']:
                axe.info(f"New RETRY_ATTEMPTS configuration: {c['retry']['attempts']}")
                self.RETRY_ATTEMPTS = c['retry']['attempts']

            if self.HEALTHY_RETRY_SECONDS != c['retry']['seconds']['healthy']:
                axe.info(f"New HEALTHY_RETRY_SECONDS configuration: {c['retry']['seconds']['healthy']}")
                self.HEALTHY_RETRY_SECONDS = c['retry']['seconds']['healthy']
                self.RETRY_SECONDS = self.HEALTHY_RETRY_SECONDS

            if self.UNHEALTHY_RETRY_SECONDS != c['retry']['seconds']['unhealthy']:
                axe.info(f"New UNHEALTHY_RETRY_SECONDS configuration: {c['retry']['seconds']['unhealthy']}")
                self.UNHEALTHY_RETRY_SECONDS = c['retry']['seconds']['unhealthy']

            if self.config_check_cycles != c['config_check']['cycles']:
                axe.info(f"New config_check_cycles configuration: {c['config_check']['cycles']}")
                self.config_check_cycles = c['config_check']['cycles']

            if self.power_server != c['server']['address']:
                axe.info(f"New power_server configuration: {c['server']['address']}")
                self.power_server = c['server']['address']
                self.set_config_url()
                self.set_url()

            if self.server_port != c['server']['port']:
                axe.info(f"New server_port configuration: {c['server']['port']}")
                self.server_port = c['server']['port']
                self.set_config_url()
                self.set_url()

            if self.api_base_endpoint != c['api']['base_endpoint']:
                axe.info(f'New base api endpoint: {c["api"]["base_endpoint"]}')
                self.api_base_endpoint = c['api']['base_endpoint']
                self.set_api_config_endpoint()
                self.set_config_url()
                self.set_url()
        else:
            axe.error(f'Unable to get to config endpoint: {self.config_url}')

    def get_api_config_endpoint(self):
        return f'{self.api_base_endpoint}/_CONFIG/client/{self.hostname}'

    def get_config_url(self):
        return f'http://{self.power_server}:{self.server_port}/{self.api_config_endpoint}'

    def get_url(self):
        return f'http://{self.power_server}:{self.server_port}/{self.api_base_endpoint}/{self.hostname}'

    def set_api_config_endpoint(self):
        self.api_config_endpoint = self.get_api_config_endpoint()
        self.set_config_url()

    def set_config_url(self):
        self.config_url = self.get_config_url()

    def set_url(self):
        self.url = self.get_url()

    @staticmethod
    def get_hostname():
        try:
            hn = socket.gethostname()
        except Exception as e:
            hn = None
            axe.error(f"Could not get hostname because an error occurred:\n{e}")
        return hn

    def get_status_code(self):
        return requests.get(self.url)

    def check_power(self):
        axe.debug(f'Checking {self.url} for power with {self.RETRY_ATTEMPTS} retry attempts, DRYRUN={self.DRYRUN}')

        while True:
            self.config_check_count += 1
            self.RETRY_COUNT += 1
            response = self.get_status_code()
            remaining_tries = self.RETRY_ATTEMPTS - self.RETRY_COUNT

            if self.config_check_cycles == self.config_check_count:
                self.update_config()
                self.config_check_count = 0

            if response.ok and response.status_code == 200:
                if self.RETRY_COUNT == 1:
                    axe.info(f'Power is up')
                else:
                    axe.info(f'Power is back up after {self.RETRY_COUNT} checks')
                self.RETRY_COUNT = 0
                self.RETRY_SECONDS = self.HEALTHY_RETRY_SECONDS
            else:
                self.RETRY_SECONDS = self.UNHEALTHY_RETRY_SECONDS
                if self.RETRY_COUNT < int(math.floor(self.RETRY_ATTEMPTS / 2)):
                    axe.warning(f'Attempt {self.RETRY_COUNT} request failed, power may be down. \
                             Will try {remaining_tries} more times before initiating shutdown.')
                elif self.RETRY_COUNT < self.RETRY_ATTEMPTS:
                    axe.error(f'Attempt {self.RETRY_COUNT} requests failed, power appears down.\
                             Will try {remaining_tries} more times before initiating shutdown.')
                else:
                    axe.critical(f'Attempt {self.RETRY_COUNT} requests failed, power appears down.\
                                 Engaging Killswitch!')
                    ks = KillSwitch(DRYRUN=self.DRYRUN)
                    ks.power_down()
                    break

            time.sleep(self.RETRY_SECONDS)

if __name__ == "__main__":
    pc = PowerClient()
    pc.check_power()