sunetdrive/templates/script/restart-nextcloud-farm.erb

#!/usr/bin/env python3.9
# vim: set filetype=python:

import subprocess
import sys
import time
import urllib.parse

import requests
from drive_utils import (build_fqdn, run_remote_command,
                         smoketest_nextcloud_node)


def add_downtime(fqdn: str,
                 apikey: str,
                 monitor_host: str = 'monitor.drive.test.sunet.se') -> None:
    if not apikey:
        return
    print("Adding downtime for: {}".format(fqdn))
    action = 'schedule_host_svc_downtime'

    start_time = int(time.time())
    end_time = start_time + (10 * 60)  # 10 minutes
    data = {
        'comment_data': 'Reboot from script',
        'start_time': start_time,
        'end_time': end_time
    }

    post_url = 'https://{}/thruk/r/hosts/{}/cmd/{}'.format(
        monitor_host, fqdn, action)
    headers = {'X-Thruk-Auth-Key': apikey}
    requests.post(post_url, data=data, headers=headers)


def remove_downtime(fqdn: str,
                    apikey: str,
                    monitor_host: str = 'monitor.drive.test.sunet.se') -> None:
    if not apikey:
        return
    print("Removing downtime for: {}".format(fqdn))
    get_url = 'https://{}/thruk/r/hosts?name={}&columns=services'.format(
        monitor_host, fqdn)
    headers = {'X-Thruk-Auth-Key': apikey}
    req = requests.get(get_url, headers=headers)
    action = 'del_active_service_downtimes'
    for service in req.json()[0]['services']:
        post_url = 'https://{}/thruk/r/services/{}/{}/cmd/{}'.format(
            monitor_host, fqdn, urllib.parse.quote(service), action)
        requests.post(post_url, headers=headers)


def run_command(command: list[str]) -> tuple:
    with subprocess.Popen(command,
                          stdout=subprocess.PIPE,
                          stderr=subprocess.PIPE) as proc:
        outs, errs = proc.communicate()
    try:
        reply = outs.decode().strip('\n')
    except AttributeError:
        reply = str()
    return (reply, errs)


def main() -> int:
    customer = "<%= @customer %>"
    environment = "<%= @environment %>"
    apikey_test = "<%= @apikey_test %>"
    apikey_prod = "<%= @apikey_prod %>"

    cosmos_command = ['sudo run-cosmos']
    nc_upgrade_command = 'sudo /usr/local/bin/occ upgrade'
    repair_command = 'sudo /usr/local/bin/occ maintenance:repair'
    reboot_command = ['sudo /usr/local/bin/safer_reboot']
    server_type = "node"
    if customer == "common":
        customer = "gss"
        server_type = "gss"

    for number in reversed(range(1, 4)):
        fqdn = build_fqdn(customer, environment, number, server_type)
        add_downtime(fqdn, apikey_test)
        add_downtime(fqdn, apikey_prod, monitor_host="monitor.drive.sunet.se")
        print("Upgrading: {}".format(fqdn))

        print("\tRunning cosmos command at {}".format(fqdn))
        run_remote_command(fqdn,
                           cosmos_command,
                           user="script",
                           output=subprocess.DEVNULL)
        print("\tRunning reboot command at {}".format(fqdn))
        run_remote_command(fqdn, reboot_command, user="script")
        success = False
        for testnumber in reversed(range(1, 32, 2)):
            print("\tSleeping for {} seconds before smoketest on {}".format(
                testnumber, fqdn))
            time.sleep(testnumber)

            if smoketest_nextcloud_node(fqdn):
                success = True

                break

        remove_downtime(fqdn, apikey_test)
        remove_downtime(fqdn,
                        apikey_prod,
                        monitor_host="monitor.drive.sunet.se")
        if success:
            print("Upgrade cycle succeeded on {} ".format(fqdn))
        else:
            print("Smoketest failed on {} after server reboot command".format(
                fqdn))

            return 6
    fqdn = build_fqdn(customer, environment, 3, server_type)
    print("Running nextcloud upgrade command at {}".format(fqdn))
    run_remote_command(fqdn, [nc_upgrade_command], user="script")
    print("Running repair command on {}".format(fqdn))
    run_remote_command(fqdn, [repair_command], user="script")
    print("All {}-servers successfully upgraded for {}".format(
        environment, customer))

    return 0


if __name__ == "__main__":
    sys.exit(main())