Source code for calML.restart_servers

#############################################################################
# Author: A. Silenzi, S. Hauf
# Created on September, 2018, 11:04 AM
# Copyright (C) European XFEL GmbH Hamburg. All rights reserved.
#############################################################################

from json import loads

from karabo.middlelayer import (AccessMode, Device, Slot, State, String,
                                VectorString, background, sleep)
from requests import get, put


class RestartCalServers(Device):
    """ A device for reliably restarting calibration servers.

    This device is to be used to initiate a "blank slate" restart
    of the calibration pipeline.

    """
    status = String(accessMode=AccessMode.READONLY)
    host_template = String(displayedName='Server host templ.',
                           accessMode=AccessMode.INITONLY,
                           defaultValue='http://spb-br-sys-cal-{}:8080')

    server_insets = VectorString(displayedName='Server insets',
                                 accessMode=AccessMode.INITONLY,
                                 defaultValue=[str(i) for i in range(8)])

    omit = VectorString(displayedName="Omit from restart",
                        accessMode=AccessMode.RECONFIGURABLE)

    def onInitialization(self):
        self.state = State.ACTIVE

    def get_json(self, url):
        r = get(url)
        if r.status_code != 200:
            self.state = State.ERROR
            rstat = r.status_code
            self.status = f"could not connect to {url} error {rstat}"
            return False
        return r.json()

    def put_json(self, url, json):
        r = put(url, json=json)
        if r.status_code != 200:
            self.state = State.ERROR
            rstat = r.status_code
            self.status = f"could not connect to {url} error {rstat}"
            return False
        return r.json()

    def is_up(self, url):
        d = self.get_json(url)
        if not d:
            return False
        return all(
            [sd.get('status', '').startswith('up') for sd in d['servers']])

    def stop(self, url, name):
        json = {'server': {'name': name, 'command': 'group'}}
        ret = self.put_json(url, json)
        if not ret:
            return False
        json = {'server': {'name': name, 'command': 'down'}}
        return self.put_json(url, json)

    def start(self, url, name):
        json = {'server': {'name': name, 'command': 'up'}}
        return self.put_json(url, json)

    @Slot(displayedName="Restart servers")
    def restartServers(self):
        background(self._restart_servers)

    def _restart_servers(self):
        """ Perform a server restart.
        """
        services = {}
        self.status = "Restarting..."
        self.state = State.STOPPING
        for i in self.server_insets:
            host = self.host_template.format(i)
            dic = self.get_json(f'{host}/api/servers.json')
            if not dic:
                break
            for s_dic in dic['servers']:
                name = s_dic['name']
                this_service = self.serverId.lower().replace("/", "_")
                dont_kill = ['webserver', this_service]
                if self.omit:
                    dont_kill += list({name for s in self.omit if s in name})
                if name not in dont_kill:
                    if name in services:
                        msg = 'warning service name not unique: {}'.\
                            format(name)
                        self.status = msg
                    services[name] = f'{host}/api/servers/{name}.json'

        for name, url in services.items():
            # stop
            if not self.stop(url, name)['success']:
                self.status = f"problem stopping {name}"
            else:
                self.status = f"sucessfully stopped {name}"
                sleep(1)

        for name, url in services.items():
            # wait for service to be down.
            while self.is_up(url):
                nap_time = 0.5
                sleep(nap_time)
            # restart
            if not self.start(url, name)['success']:
                self.status = f"problem starting {name}"
            else:
                # sleep here to prevent users from being too trigger happy
                self.status = f"successfully restarted {name}"
                sleep(1)
        self.state = State.ACTIVE
        self.status = "All servers restarted!"