site_utils/rpm_control_system/frontend_server.py

#!/usr/bin/python
# Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

import errno
import heapq
import logging
import os
import sys
import socket
import threading
import xmlrpclib

import rpm_logging_config
from config import rpm_config
from MultiThreadedXMLRPCServer import MultiThreadedXMLRPCServer
from rpm_infrastructure_exception import RPMInfrastructureException

import common
from autotest_lib.server import frontend
from autotest_lib.site_utils.rpm_control_system import utils

DEFAULT_RPM_COUNT = 0
TERMINATED = -1

# Indexes for accessing heap entries.
RPM_COUNT = 0
DISPATCHER_URI = 1

LOG_FILENAME_FORMAT = rpm_config.get('GENERAL','frontend_logname_format')
DEFAULT_RPM_ID = rpm_config.get('RPM_INFRASTRUCTURE', 'default_rpm_id')

# Valid state values.
VALID_STATE_VALUES = ['ON', 'OFF', 'CYCLE']

# Servo-interface mapping file
MAPPING_FILE = os.path.join(
        os.path.dirname(__file__),
        rpm_config.get('CiscoPOE', 'servo_interface_mapping_file'))

# Size of the LRU that holds power management unit information related
# to a device, e.g. rpm_hostname, outlet, hydra_hostname, etc.
LRU_SIZE = rpm_config.getint('RPM_INFRASTRUCTURE', 'lru_size')


class RPMFrontendServer(object):
    """
    This class is the frontend server of the RPM Infrastructure. All clients
    will send their power state requests to this central server who will
    forward the requests to an avaliable or already assigned RPM dispatcher
    server.

    Once the dispatcher processes the request it will return the result
    to this frontend server who will send the result back to the client.

    All calls to this server are blocking.

    @var _dispatcher_minheap: Min heap that returns a list of format-
                              [ num_rpm's, dispatcher_uri ]
                              Used to choose the least loaded dispatcher.
    @var _entry_dict: Maps dispatcher URI to an entry (list) inside the min
                     heap. If a dispatcher server shuts down this allows us to
                     invalidate the entry in the minheap.
    @var _lock: Used to protect data from multiple running threads all
                manipulating the same data.
    @var _rpm_dict: Maps rpm hostname's to an already assigned dispatcher
                    server.
    @var _mapping_last_modified: Last-modified time of the servo-interface
                                 mapping file.
    @var _servo_interface: Maps servo hostname to (switch_hostname, interface).
    @var _rpm_info: An LRU cache to hold recently visited rpm information
                    so that we don't hit AFE too often. The elements in
                    the cache are instances of PowerUnitInfo indexed by
                    dut hostnames. POE info is not stored in the cache.
    @var _afe: AFE instance to talk to autotest. Used to retrieve rpm hostname.
    @var _email_handler: Email handler to use to control email notifications.
    """


    def __init__(self, email_handler=None):
        """
        RPMFrontendServer constructor.

        Initializes instance variables.
        """
        self._dispatcher_minheap = []
        self._entry_dict = {}
        self._lock = threading.Lock()
        self._mapping_last_modified = os.path.getmtime(MAPPING_FILE)
        self._servo_interface = utils.load_servo_interface_mapping()
        self._rpm_dict = {}
        self._afe = frontend.AFE()
        self._rpm_info = utils.LRUCache(size=LRU_SIZE)
        self._email_handler = email_handler


    def queue_request(self, device_hostname, new_state):
        """
        Forwards a request to change a device's (a dut or a servo) power state
        to the appropriate dispatcher server.

        This call will block until the forwarded request returns.

        @param device_hostname: Hostname of the device whose power state we want to
                             change.
        @param new_state: [ON, OFF, CYCLE] State to which we want to set the
                          device's outlet to.

        @return: True if the attempt to change power state was successful,
                 False otherwise.

        @raise RPMInfrastructureException: No dispatchers are available or can
                                           be reached.
        """
        # Remove any DNS Zone information and simplify down to just the hostname.
        device_hostname = device_hostname.split('.')[0]
        new_state = new_state.upper()
        # Put new_state in all uppercase letters
        if new_state not in VALID_STATE_VALUES:
            logging.error('Received request to set device %s to invalid '
                          'state %s', device_hostname, new_state)
            return False
        logging.info('Received request to set device: %s to state: %s',
                     device_hostname, new_state)
        powerunit_info = self._get_powerunit_info(device_hostname)
        dispatcher_uri = self._get_dispatcher(powerunit_info)
        if not dispatcher_uri:
            # No dispatchers available.
            raise RPMInfrastructureException('No dispatchers available.')
        client = xmlrpclib.ServerProxy(dispatcher_uri, allow_none=True)
        try:
            # Block on the request and return the result once it arrives.
            return client.queue_request(powerunit_info, new_state)
        except socket.error as er:
            # Dispatcher Server is not reachable. Unregister it and retry.
            logging.error("Can't reach Dispatch Server: %s. Error: %s",
                          dispatcher_uri, errno.errorcode[er.errno])
            if self.is_network_infrastructure_down():
                # No dispatchers can handle this request so raise an Exception
                # to the caller.
                raise RPMInfrastructureException('No dispatchers can be'
                                                 'reached.')
            logging.info('Will attempt forwarding request to other dispatch '
                         'servers.')
            logging.error('Unregistering %s due to error. Recommend resetting '
                          'that dispatch server.', dispatcher_uri)
            self.unregister_dispatcher(dispatcher_uri)
            # Retry forwarding the request.
            return self.queue_request(device_hostname, new_state)


    def is_network_infrastructure_down(self):
        """
        Check to see if we can communicate with any dispatcher servers.

        Only called in the situation that queuing a request to a dispatcher
        server failed.

        @return: False if any dispatcher server is up and the rpm infrastructure
                 can still function. True otherwise.
        """
        for dispatcher_entry in self._dispatcher_minheap:
            dispatcher = xmlrpclib.ServerProxy(
                    dispatcher_entry[DISPATCHER_URI], allow_none=True)
            try:
                if dispatcher.is_up():
                    # Atleast one dispatcher is alive so our network is fine.
                    return False
            except socket.error:
                # Can't talk to this dispatcher so keep looping.
                pass
        logging.error("Can't reach any dispatchers. Check frontend network "
                      'status or all dispatchers are down.')
        return True


    def _get_powerunit_info(self, device_hostname):
        """Get the power management unit information for a device.

        A device could be a chromeos dut or a servo.
        1) ChromeOS dut
        Chromeos dut is managed by RPM. The related information
        we need to know include rpm hostname, rpm outlet, hydra hostname.
        Such information can be retrieved from afe_host_attributes table
        from afe. A local LRU cache is used avoid hitting afe too often.

        2) Servo
        Servo is managed by POE. The related information we need to know
        include poe hostname, poe interface. Such information is
        stored in a local file and read into memory.

        @param device_hostname: A string representing the device's hostname.

        @returns: A PowerUnitInfo object.
        @raises RPMInfrastructureException if failed to get the power
                unit info.

        """
        with self._lock:
            if device_hostname.endswith('servo'):
                # Servos are managed by Cisco POE switches.
                reload_info = utils.reload_servo_interface_mapping_if_necessary(
                        self._mapping_last_modified)
                if reload_info:
                    self._mapping_last_modified, self._servo_interface = reload_info
                switch_if_tuple = self._servo_interface.get(device_hostname)
                if not switch_if_tuple:
                    raise RPMInfrastructureException(
                            'Could not determine POE hostname for %s. '
                            'Please check the servo-interface mapping file.',
                            device_hostname)
                else:
                    return utils.PowerUnitInfo(
                            device_hostname=device_hostname,
                            powerunit_type=utils.PowerUnitInfo.POWERUNIT_TYPES.POE,
                            powerunit_hostname=switch_if_tuple[0],
                            outlet=switch_if_tuple[1],
                            hydra_hostname=None)
            else:
                # Regular DUTs are managed by RPMs.
                if device_hostname in self._rpm_info:
                    return self._rpm_info[device_hostname]
                else:
                    hosts = self._afe.get_hosts(hostname=device_hostname)
                    if not hosts:
                        raise RPMInfrastructureException(
                                'Can not retrieve rpm information '
                                'from AFE for %s, no host found.' % device_hostname)
                    else:
                        info = utils.PowerUnitInfo.get_powerunit_info(hosts[0])
                        self._rpm_info[device_hostname] = info
                        return info


    def _get_dispatcher(self, powerunit_info):
        """
        Private method that looks up or assigns a dispatcher server
        responsible for communicating with the given RPM/POE.

        Will also call _check_dispatcher to make sure it is up before returning
        it.

        @param powerunit_info: A PowerUnitInfo instance.

        @return: URI of dispatcher server responsible for the rpm/poe.
                 None if no dispatcher servers are available.
        """
        powerunit_type = powerunit_info.powerunit_type
        powerunit_hostname = powerunit_info.powerunit_hostname
        with self._lock:
            if self._rpm_dict.get(powerunit_hostname):
                return self._rpm_dict[powerunit_hostname]
            logging.info('No Dispatcher assigned for %s %s.',
                         powerunit_type, powerunit_hostname)
            # Choose the least loaded dispatcher to communicate with the RPM.
            try:
                heap_entry = heapq.heappop(self._dispatcher_minheap)
            except IndexError:
                logging.error('Infrastructure Error: Frontend has no'
                              'registered dispatchers to field out this '
                              'request!')
                return None
            dispatcher_uri = heap_entry[DISPATCHER_URI]
            # Put this entry back in the heap with an RPM Count + 1.
            heap_entry[RPM_COUNT] = heap_entry[RPM_COUNT] + 1
            heapq.heappush(self._dispatcher_minheap, heap_entry)
            logging.info('Assigning %s for %s %s', dispatcher_uri,
                         powerunit_type, powerunit_hostname)
            self._rpm_dict[powerunit_hostname] = dispatcher_uri
            return dispatcher_uri


    def register_dispatcher(self, dispatcher_uri):
        """
        Called by a dispatcher server so that the frontend server knows it is
        available to field out RPM requests.

        Adds an entry to the min heap and entry map for this dispatcher.

        @param dispatcher_uri: Address of dispatcher server we are registering.
        """
        logging.info('Registering uri: %s as a rpm dispatcher.', dispatcher_uri)
        with self._lock:
            heap_entry = [DEFAULT_RPM_COUNT, dispatcher_uri]
            heapq.heappush(self._dispatcher_minheap, heap_entry)
            self._entry_dict[dispatcher_uri] = heap_entry


    def unregister_dispatcher(self, uri_to_unregister):
        """
        Called by a dispatcher server as it exits so that the frontend server
        knows that it is no longer available to field out requests.

        Assigns an rpm count of -1 to this dispatcher so that it will be pushed
        out of the min heap.

        Removes from _rpm_dict all entries with the value of this dispatcher so
        that those RPM's can be reassigned to a new dispatcher.

        @param uri_to_unregister: Address of dispatcher server we are
                                  unregistering.
        """
        logging.info('Unregistering uri: %s as a rpm dispatcher.',
                     uri_to_unregister)
        with self._lock:
            heap_entry = self._entry_dict.get(uri_to_unregister)
            if not heap_entry:
                logging.warning('%s was not registered.', uri_to_unregister)
                return
            # Set this entry's RPM_COUNT to TERMINATED (-1).
            heap_entry[RPM_COUNT] = TERMINATED
            # Remove all RPM mappings.
            for rpm, dispatcher in self._rpm_dict.items():
                if dispatcher == uri_to_unregister:
                    self._rpm_dict[rpm] = None
            self._entry_dict[uri_to_unregister] = None
            # Re-sort the heap and remove any terminated dispatchers.
            heapq.heapify(self._dispatcher_minheap)
            self._remove_terminated_dispatchers()


    def _remove_terminated_dispatchers(self):
        """
        Peek at the head of the heap and keep popping off values until there is
        a non-terminated dispatcher at the top.
        """
        # Heapq guarantees the head of the heap is in the '0' index.
        try:
            # Peek at the next element in the heap.
            top_of_heap = self._dispatcher_minheap[0]
            while top_of_heap[RPM_COUNT] is TERMINATED:
                # Pop off the top element.
                heapq.heappop(self._dispatcher_minheap)
                # Peek at the next element in the heap.
                top_of_heap = self._dispatcher_minheap[0]
        except IndexError:
            # No more values in the heap. Can be thrown by both minheap[0]
            # statements.
            pass


    def suspend_emails(self, hours):
        """Suspend email notifications.

        @param hours: How many hours to suspend email notifications.
        """
        if self._email_handler:
            self._email_handler.suspend_emails(hours)


    def resume_emails(self):
        """Resume email notifications."""
        if self._email_handler:
            self._email_handler.resume_emails()


if __name__ == '__main__':
    """
    Main function used to launch the frontend server. Creates an instance of
    RPMFrontendServer and registers it to a MultiThreadedXMLRPCServer instance.
    """
    if len(sys.argv) != 2:
      print 'Usage: ./%s <log_file_dir>.' % sys.argv[0]
      sys.exit(1)

    email_handler = rpm_logging_config.set_up_logging_to_file(
            sys.argv[1], LOG_FILENAME_FORMAT)
    frontend_server = RPMFrontendServer(email_handler=email_handler)
    # We assume that external clients will always connect to us via the
    # hostname, so listening on the hostname ensures we pick the right network
    # interface.
    address = socket.gethostname()
    port = rpm_config.getint('RPM_INFRASTRUCTURE', 'frontend_port')
    server = MultiThreadedXMLRPCServer((address, port), allow_none=True)
    server.register_instance(frontend_server)
    logging.info('Listening on %s port %d', address, port)
    server.serve_forever()