db_replicator.py

# Copyright (c) 2010-2012 OpenStack Foundation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import json
import os
import random
import math
import time
import shutil
import uuid
import errno
import re
from contextlib import contextmanager
from swift import gettext_ as _

from eventlet import GreenPool, sleep, Timeout
from eventlet.green import subprocess

import swift.common.db
from swift.common.constraints import check_drive
from swift.common.utils import get_logger, whataremyips, storage_directory, \
    renamer, mkdirs, lock_parent_directory, config_true_value, \
    unlink_older_than, dump_recon_cache, rsync_module_interpolation, \
    parse_override_options, round_robin_iter, Everything, get_db_files, \
    parse_db_filename, quote, RateLimitedIterator
from swift.common import ring
from swift.common.ring.utils import is_local_device
from swift.common.http import HTTP_NOT_FOUND, HTTP_INSUFFICIENT_STORAGE, \
    is_success
from swift.common.bufferedhttp import BufferedHTTPConnection
from swift.common.exceptions import DriveNotMounted
from swift.common.daemon import Daemon
from swift.common.swob import Response, HTTPNotFound, HTTPNoContent, \
    HTTPAccepted, HTTPBadRequest
from swift.common.recon import DEFAULT_RECON_CACHE_PATH, \
    server_type_to_recon_file


DEBUG_TIMINGS_THRESHOLD = 10


def quarantine_db(object_file, server_type):
    """
    In the case that a corrupt file is found, move it to a quarantined area to
    allow replication to fix it.

    :param object_file: path to corrupt file
    :param server_type: type of file that is corrupt
                        ('container' or 'account')
    """
    object_dir = os.path.dirname(object_file)
    quarantine_dir = os.path.abspath(
        os.path.join(object_dir, '..', '..', '..', '..', 'quarantined',
                     server_type + 's', os.path.basename(object_dir)))
    try:
        renamer(object_dir, quarantine_dir, fsync=False)
    except OSError as e:
        if e.errno not in (errno.EEXIST, errno.ENOTEMPTY):
            raise
        quarantine_dir = "%s-%s" % (quarantine_dir, uuid.uuid4().hex)
        renamer(object_dir, quarantine_dir, fsync=False)


def looks_like_partition(dir_name):
    """
    True if the directory name is a valid partition number, False otherwise.
    """
    try:
        part = int(dir_name)
        return part >= 0
    except ValueError:
        return False


def roundrobin_datadirs(datadirs):
    """
    Generator to walk the data dirs in a round robin manner, evenly
    hitting each device on the system, and yielding any .db files
    found (in their proper places). The partitions within each data
    dir are walked randomly, however.

    :param datadirs: a list of tuples of (path, context, partition_filter) to
                     walk. The context may be any object; the context is not
                     used by this function but is included with each yielded
                     tuple.
    :returns: A generator of (partition, path_to_db_file, context)
    """

    def walk_datadir(datadir, context, part_filter):
        partitions = [pd for pd in os.listdir(datadir)
                      if looks_like_partition(pd) and part_filter(pd)]
        random.shuffle(partitions)
        for partition in partitions:
            part_dir = os.path.join(datadir, partition)
            if not os.path.isdir(part_dir):
                continue
            suffixes = os.listdir(part_dir)
            if not suffixes:
                os.rmdir(part_dir)
                continue
            for suffix in suffixes:
                suff_dir = os.path.join(part_dir, suffix)
                if not os.path.isdir(suff_dir):
                    continue
                hashes = os.listdir(suff_dir)
                if not hashes:
                    os.rmdir(suff_dir)
                    continue
                for hsh in hashes:
                    hash_dir = os.path.join(suff_dir, hsh)
                    if not os.path.isdir(hash_dir):
                        continue
                    object_file = os.path.join(hash_dir, hsh + '.db')
                    # common case
                    if os.path.exists(object_file):
                        yield (partition, object_file, context)
                        continue
                    # look for any alternate db filenames
                    db_files = get_db_files(object_file)
                    if db_files:
                        yield (partition, db_files[-1], context)
                        continue
                    try:
                        os.rmdir(hash_dir)
                    except OSError as e:
                        if e.errno != errno.ENOTEMPTY:
                            raise

    its = [walk_datadir(datadir, context, filt)
           for datadir, context, filt in datadirs]

    rr_its = round_robin_iter(its)
    for datadir in rr_its:
        yield datadir


class ReplConnection(BufferedHTTPConnection):
    """
    Helper to simplify REPLICATEing to a remote server.
    """

    def __init__(self, node, partition, hash_, logger):
        self.logger = logger
        self.node = node
        host = "%s:%s" % (node['replication_ip'], node['replication_port'])
        BufferedHTTPConnection.__init__(self, host)
        self.path = '/%s/%s/%s' % (node['device'], partition, hash_)

    def replicate(self, *args):
        """
        Make an HTTP REPLICATE request

        :param args: list of json-encodable objects

        :returns: bufferedhttp response object
        """
        try:
            body = json.dumps(args)
            self.request('REPLICATE', self.path, body,
                         {'Content-Type': 'application/json'})
            response = self.getresponse()
            response.data = response.read()
            return response
        except (Exception, Timeout):
            self.logger.exception(
                _('ERROR reading HTTP response from %s'), self.node)
            return None


class Replicator(Daemon):
    """
    Implements the logic for directing db replication.
    """

    def __init__(self, conf, logger=None):
        self.conf = conf
        self.logger = logger or get_logger(conf, log_route='replicator')
        self.root = conf.get('devices', '/srv/node')
        self.mount_check = config_true_value(conf.get('mount_check', 'true'))
        self.bind_ip = conf.get('bind_ip', '0.0.0.0')
        self.port = int(conf.get('bind_port', self.default_port))
        concurrency = int(conf.get('concurrency', 8))
        self.cpool = GreenPool(size=concurrency)
        swift_dir = conf.get('swift_dir', '/etc/swift')
        self.ring = ring.Ring(swift_dir, ring_name=self.server_type)
        self._local_device_ids = set()
        self.per_diff = int(conf.get('per_diff', 1000))
        self.max_diffs = int(conf.get('max_diffs') or 100)
        self.interval = float(conf.get('interval') or
                              conf.get('run_pause') or 30)
        if 'run_pause' in conf:
            if 'interval' in conf:
                self.logger.warning(
                    'Option %(type)s-replicator/run_pause is deprecated '
                    'and %(type)s-replicator/interval is already configured. '
                    'You can safely remove run_pause; it is now ignored and '
                    'will be removed in a future version.'
                    % {'type': self.server_type})
            else:
                self.logger.warning(
                    'Option %(type)s-replicator/run_pause is deprecated '
                    'and will be removed in a future version. '
                    'Update your configuration to use option '
                    '%(type)s-replicator/interval.'
                    % {'type': self.server_type})
        self.databases_per_second = float(
            conf.get('databases_per_second', 50))
        self.node_timeout = float(conf.get('node_timeout', 10))
        self.conn_timeout = float(conf.get('conn_timeout', 0.5))
        self.rsync_compress = config_true_value(
            conf.get('rsync_compress', 'no'))
        self.rsync_module = conf.get('rsync_module', '').rstrip('/')
        if not self.rsync_module:
            self.rsync_module = '{replication_ip}::%s' % self.server_type
        self.reclaim_age = float(conf.get('reclaim_age', 86400 * 7))
        swift.common.db.DB_PREALLOCATION = \
            config_true_value(conf.get('db_preallocation', 'f'))
        swift.common.db.QUERY_LOGGING = \
            config_true_value(conf.get('db_query_logging', 'f'))
        self._zero_stats()
        self.recon_cache_path = conf.get('recon_cache_path',
                                         DEFAULT_RECON_CACHE_PATH)
        self.recon_replicator = server_type_to_recon_file(self.server_type)
        self.rcache = os.path.join(self.recon_cache_path,
                                   self.recon_replicator)
        self.extract_device_re = re.compile('%s%s([^%s]+)' % (
            self.root, os.path.sep, os.path.sep))
        self.handoffs_only = config_true_value(conf.get('handoffs_only', 'no'))

    def _zero_stats(self):
        """Zero out the stats."""
        self.stats = {'attempted': 0, 'success': 0, 'failure': 0, 'ts_repl': 0,
                      'no_change': 0, 'hashmatch': 0, 'rsync': 0, 'diff': 0,
                      'remove': 0, 'empty': 0, 'remote_merge': 0,
                      'start': time.time(), 'diff_capped': 0, 'deferred': 0,
                      'failure_nodes': {}}

    def _report_stats(self):
        """Report the current stats to the logs."""
        now = time.time()
        self.logger.info(
            _('Attempted to replicate %(count)d dbs in %(time).5f seconds '
              '(%(rate).5f/s)'),
            {'count': self.stats['attempted'],
             'time': now - self.stats['start'],
             'rate': self.stats['attempted'] /
                (now - self.stats['start'] + 0.0000001)})
        self.logger.info(_('Removed %(remove)d dbs') % self.stats)
        self.logger.info(_('%(success)s successes, %(failure)s failures')
                         % self.stats)
        dump_recon_cache(
            {'replication_stats': self.stats,
             'replication_time': now - self.stats['start'],
             'replication_last': now},
            self.rcache, self.logger)
        self.logger.info(' '.join(['%s:%s' % item for item in
                         sorted(self.stats.items()) if item[0] in
                         ('no_change', 'hashmatch', 'rsync', 'diff', 'ts_repl',
                          'empty', 'diff_capped', 'remote_merge')]))

    def _add_failure_stats(self, failure_devs_info):
        for node, dev in failure_devs_info:
            self.stats['failure'] += 1
            failure_devs = self.stats['failure_nodes'].setdefault(node, {})
            failure_devs.setdefault(dev, 0)
            failure_devs[dev] += 1

    def _rsync_file(self, db_file, remote_file, whole_file=True,
                    different_region=False):
        """
        Sync a single file using rsync. Used by _rsync_db to handle syncing.

        :param db_file: file to be synced
        :param remote_file: remote location to sync the DB file to
        :param whole-file: if True, uses rsync's --whole-file flag
        :param different_region: if True, the destination node is in a
                                 different region

        :returns: True if the sync was successful, False otherwise
        """
        popen_args = ['rsync', '--quiet', '--no-motd',
                      '--timeout=%s' % int(math.ceil(self.node_timeout)),
                      '--contimeout=%s' % int(math.ceil(self.conn_timeout))]
        if whole_file:
            popen_args.append('--whole-file')

        if self.rsync_compress and different_region:
            # Allow for compression, but only if the remote node is in
            # a different region than the local one.
            popen_args.append('--compress')

        popen_args.extend([db_file, remote_file])
        proc = subprocess.Popen(popen_args)
        proc.communicate()
        if proc.returncode != 0:
            self.logger.error(_('ERROR rsync failed with %(code)s: %(args)s'),
                              {'code': proc.returncode, 'args': popen_args})
        return proc.returncode == 0

    def _rsync_db(self, broker, device, http, local_id,
                  replicate_method='complete_rsync', replicate_timeout=None,
                  different_region=False):
        """
        Sync a whole db using rsync.

        :param broker: DB broker object of DB to be synced
        :param device: device to sync to
        :param http: ReplConnection object
        :param local_id: unique ID of the local database replica
        :param replicate_method: remote operation to perform after rsync
        :param replicate_timeout: timeout to wait in seconds
        :param different_region: if True, the destination node is in a
                                 different region
        """
        rsync_module = rsync_module_interpolation(self.rsync_module, device)
        rsync_path = '%s/tmp/%s' % (device['device'], local_id)
        remote_file = '%s/%s' % (rsync_module, rsync_path)
        mtime = os.path.getmtime(broker.db_file)
        if not self._rsync_file(broker.db_file, remote_file,
                                different_region=different_region):
            return False
        # perform block-level sync if the db was modified during the first sync
        if os.path.exists(broker.db_file + '-journal') or \
                os.path.getmtime(broker.db_file) > mtime:
            # grab a lock so nobody else can modify it
            with broker.lock():
                if not self._rsync_file(broker.db_file, remote_file,
                                        whole_file=False,
                                        different_region=different_region):
                    return False
        with Timeout(replicate_timeout or self.node_timeout):
            response = http.replicate(replicate_method, local_id,
                                      os.path.basename(broker.db_file))
        return response and 200 <= response.status < 300

    def _send_replicate_request(self, http, *repl_args):
        with Timeout(self.node_timeout):
            response = http.replicate(*repl_args)
        if not response or not is_success(response.status):
            if response:
                self.logger.error('ERROR Bad response %s from %s',
                                  response.status, http.host)
            return False
        return True

    def _usync_db(self, point, broker, http, remote_id, local_id):
        """
        Sync a db by sending all records since the last sync.

        :param point: synchronization high water mark between the replicas
        :param broker: database broker object
        :param http: ReplConnection object for the remote server
        :param remote_id: database id for the remote replica
        :param local_id: database id for the local replica

        :returns: boolean indicating completion and success
        """
        self.stats['diff'] += 1
        self.logger.increment('diffs')
        self.logger.debug('%s usyncing chunks to %s, starting at row %s',
                          broker.db_file,
                          '%(ip)s:%(port)s/%(device)s' % http.node,
                          point)
        start = time.time()
        sync_table = broker.get_syncs()
        objects = broker.get_items_since(point, self.per_diff)
        diffs = 0
        while len(objects) and diffs < self.max_diffs:
            diffs += 1
            if not self._send_replicate_request(
                    http, 'merge_items', objects, local_id):
                return False
            # replication relies on db order to send the next merge batch in
            # order with no gaps
            point = objects[-1]['ROWID']
            objects = broker.get_items_since(point, self.per_diff)

        self.logger.debug('%s usyncing chunks to %s, finished at row %s (%gs)',
                          broker.db_file,
                          '%(ip)s:%(port)s/%(device)s' % http.node,
                          point, time.time() - start)

        if objects:
            self.logger.debug(
                'Synchronization for %s has fallen more than '
                '%s rows behind; moving on and will try again next pass.',
                broker, self.max_diffs * self.per_diff)
            self.stats['diff_capped'] += 1
            self.logger.increment('diff_caps')
        else:
            with Timeout(self.node_timeout):
                response = http.replicate('merge_syncs', sync_table)
            if response and 200 <= response.status < 300:
                broker.merge_syncs([{'remote_id': remote_id,
                                     'sync_point': point}],
                                   incoming=False)
                return True
        return False

    def _in_sync(self, rinfo, info, broker, local_sync):
        """
        Determine whether or not two replicas of a databases are considered
        to be in sync.

        :param rinfo: remote database info
        :param info: local database info
        :param broker: database broker object
        :param local_sync: cached last sync point between replicas

        :returns: boolean indicating whether or not the replicas are in sync
        """
        if max(rinfo['point'], local_sync) >= info['max_row']:
            self.stats['no_change'] += 1
            self.logger.increment('no_changes')
            return True
        if rinfo['hash'] == info['hash']:
            self.stats['hashmatch'] += 1
            self.logger.increment('hashmatches')
            broker.merge_syncs([{'remote_id': rinfo['id'],
                                 'sync_point': rinfo['point']}],
                               incoming=False)
            return True

    def _http_connect(self, node, partition, db_file):
        """
        Make an http_connection using ReplConnection

        :param node: node dictionary from the ring
        :param partition: partition to send in the url
        :param db_file: DB file

        :returns: ReplConnection object
        """
        hsh, other, ext = parse_db_filename(db_file)
        return ReplConnection(node, partition, hsh, self.logger)

    def _gather_sync_args(self, info):
        """
        Convert local replication_info to sync args tuple.
        """
        sync_args_order = ('max_row', 'hash', 'id', 'created_at',
                           'put_timestamp', 'delete_timestamp', 'metadata')
        return tuple(info[key] for key in sync_args_order)

    def _repl_to_node(self, node, broker, partition, info,
                      different_region=False):
        """
        Replicate a database to a node.

        :param node: node dictionary from the ring to be replicated to
        :param broker: DB broker for the DB to be replication
        :param partition: partition on the node to replicate to
        :param info: DB info as a dictionary of {'max_row', 'hash', 'id',
                     'created_at', 'put_timestamp', 'delete_timestamp',
                     'metadata'}
        :param different_region: if True, the destination node is in a
                                 different region

        :returns: True if successful, False otherwise
        """
        http = self._http_connect(node, partition, broker.db_file)
        sync_args = self._gather_sync_args(info)
        with Timeout(self.node_timeout):
            response = http.replicate('sync', *sync_args)
        if not response:
            return False
        return self._handle_sync_response(node, response, info, broker, http,
                                          different_region=different_region)

    def _handle_sync_response(self, node, response, info, broker, http,
                              different_region=False):
        if response.status == HTTP_NOT_FOUND:  # completely missing, rsync
            self.stats['rsync'] += 1
            self.logger.increment('rsyncs')
            return self._rsync_db(broker, node, http, info['id'],
                                  different_region=different_region)
        elif response.status == HTTP_INSUFFICIENT_STORAGE:
            raise DriveNotMounted()
        elif 200 <= response.status < 300:
            rinfo = json.loads(response.data)
            local_sync = broker.get_sync(rinfo['id'], incoming=False)
            if rinfo.get('metadata', ''):
                broker.update_metadata(json.loads(rinfo['metadata']))
            return self._choose_replication_mode(
                node, rinfo, info, local_sync, broker, http,
                different_region)
        return False

    def _choose_replication_mode(self, node, rinfo, info, local_sync, broker,
                                 http, different_region):
        if self._in_sync(rinfo, info, broker, local_sync):
            self.logger.debug('%s in sync with %s, nothing to do',
                              broker.db_file,
                              '%(ip)s:%(port)s/%(device)s' % node)
            return True

        # if the difference in rowids between the two differs by
        # more than 50% and the difference is greater than per_diff,
        # rsync then do a remote merge.
        # NOTE: difference > per_diff stops us from dropping to rsync
        # on smaller containers, who have only a few rows to sync.
        if (rinfo['max_row'] / float(info['max_row']) < 0.5 and
                info['max_row'] - rinfo['max_row'] > self.per_diff):
            self.stats['remote_merge'] += 1
            self.logger.increment('remote_merges')
            return self._rsync_db(broker, node, http, info['id'],
                                  replicate_method='rsync_then_merge',
                                  replicate_timeout=(info['count'] / 2000),
                                  different_region=different_region)
        # else send diffs over to the remote server
        return self._usync_db(max(rinfo['point'], local_sync),
                              broker, http, rinfo['id'], info['id'])

    def _post_replicate_hook(self, broker, info, responses):
        """
        :param broker: broker instance for the database that just replicated
        :param info: pre-replication full info dict
        :param responses: a list of bools indicating success from nodes
        """
        pass

    def cleanup_post_replicate(self, broker, orig_info, responses):
        """
        Cleanup non primary database from disk if needed.

        :param broker: the broker for the database we're replicating
        :param orig_info: snapshot of the broker replication info dict taken
            before replication
        :param responses: a list of boolean success values for each replication
                          request to other nodes

        :return success: returns False if deletion of the database was
            attempted but unsuccessful, otherwise returns True.
        """
        log_template = 'Not deleting db %s (%%s)' % broker.db_file
        max_row_delta = broker.get_max_row() - orig_info['max_row']
        if max_row_delta < 0:
            reason = 'negative max_row_delta: %s' % max_row_delta
            self.logger.error(log_template, reason)
            return True
        if max_row_delta:
            reason = '%s new rows' % max_row_delta
            self.logger.debug(log_template, reason)
            return True
        if not (responses and all(responses)):
            reason = '%s/%s success' % (responses.count(True), len(responses))
            self.logger.debug(log_template, reason)
            return True
        # If the db has been successfully synced to all of its peers, it can be
        # removed. Callers should have already checked that the db is not on a
        # primary node.
        if not self.delete_db(broker):
            self.logger.debug(
                'Failed to delete db %s', broker.db_file)
            return False
        self.logger.debug('Successfully deleted db %s', broker.db_file)
        return True

    def _reclaim(self, broker, now=None):
        if not now:
            now = time.time()
        return broker.reclaim(now - self.reclaim_age,
                              now - (self.reclaim_age * 2))

    def _replicate_object(self, partition, object_file, node_id):
        """
        Replicate the db, choosing method based on whether or not it
        already exists on peers.

        :param partition: partition to be replicated to
        :param object_file: DB file name to be replicated
        :param node_id: node id of the node to be replicated from
        :returns: a tuple (success, responses). ``success`` is a boolean that
            is True if the method completed successfully, False otherwise.
            ``responses`` is a list of booleans each of which indicates the
            success or not of replicating to a peer node if replication has
            been attempted. ``success`` is False if any of ``responses`` is
            False; when ``responses`` is empty, ``success`` may be either True
            or False.
        """
        start_time = now = time.time()
        self.logger.debug('Replicating db %s', object_file)
        self.stats['attempted'] += 1
        self.logger.increment('attempts')
        shouldbehere = True
        responses = []
        try:
            broker = self.brokerclass(object_file, pending_timeout=30,
                                      logger=self.logger)
            self._reclaim(broker, now)
            info = broker.get_replication_info()
            bpart = self.ring.get_part(
                info['account'], info.get('container'))
            if bpart != int(partition):
                partition = bpart
                # Important to set this false here since the later check only
                # checks if it's on the proper device, not partition.
                shouldbehere = False
                name = '/' + quote(info['account'])
                if 'container' in info:
                    name += '/' + quote(info['container'])
                self.logger.error(
                    'Found %s for %s when it should be on partition %s; will '
                    'replicate out and remove.' % (object_file, name, bpart))
        except (Exception, Timeout) as e:
            if 'no such table' in str(e):
                self.logger.error(_('Quarantining DB %s'), object_file)
                quarantine_db(broker.db_file, broker.db_type)
            else:
                self.logger.exception(_('ERROR reading db %s'), object_file)
            nodes = self.ring.get_part_nodes(int(partition))
            self._add_failure_stats([(failure_dev['replication_ip'],
                                      failure_dev['device'])
                                     for failure_dev in nodes])
            self.logger.increment('failures')
            return False, responses
        if broker.is_reclaimable(now, self.reclaim_age):
            if self.report_up_to_date(info):
                self.delete_db(broker)
            self.logger.timing_since('timing', start_time)
            return True, responses
        failure_devs_info = set()
        nodes = self.ring.get_part_nodes(int(partition))
        local_dev = None
        for node in nodes:
            if node['id'] == node_id:
                local_dev = node
                break
        if shouldbehere:
            shouldbehere = bool([n for n in nodes if n['id'] == node_id])
        # See Footnote [1] for an explanation of the repl_nodes assignment.
        if len(nodes) > 1:
            i = 0
            while i < len(nodes) and nodes[i]['id'] != node_id:
                i += 1
            repl_nodes = nodes[i + 1:] + nodes[:i]
        else:  # Special case if using only a single replica
            repl_nodes = nodes
        more_nodes = self.ring.get_more_nodes(int(partition))
        if not local_dev:
            # Check further if local device is a handoff node
            for node in self.ring.get_more_nodes(int(partition)):
                if node['id'] == node_id:
                    local_dev = node
                    break
        for node in repl_nodes:
            different_region = False
            if local_dev and local_dev['region'] != node['region']:
                # This additional information will help later if we
                # want to handle syncing to a node in different
                # region with some optimizations.
                different_region = True
            success = False
            try:
                success = self._repl_to_node(node, broker, partition, info,
                                             different_region)
            except DriveNotMounted:
                try:
                    repl_nodes.append(next(more_nodes))
                except StopIteration:
                    self.logger.error(
                        _('ERROR There are not enough handoff nodes to reach '
                          'replica count for partition %s'),
                        partition)
                self.logger.error(_('ERROR Remote drive not mounted %s'), node)
            except (Exception, Timeout):
                self.logger.exception(_('ERROR syncing %(file)s with node'
                                        ' %(node)s'),
                                      {'file': object_file, 'node': node})
            if not success:
                failure_devs_info.add((node['replication_ip'], node['device']))
            self.logger.increment('successes' if success else 'failures')
            responses.append(success)
        try:
            self._post_replicate_hook(broker, info, responses)
        except (Exception, Timeout):
            self.logger.exception('UNHANDLED EXCEPTION: in post replicate '
                                  'hook for %s', broker.db_file)
        if not shouldbehere:
            if not self.cleanup_post_replicate(broker, info, responses):
                failure_devs_info.update(
                    [(failure_dev['replication_ip'], failure_dev['device'])
                     for failure_dev in repl_nodes])
        target_devs_info = set([(target_dev['replication_ip'],
                                 target_dev['device'])
                                for target_dev in repl_nodes])
        self.stats['success'] += len(target_devs_info - failure_devs_info)
        self._add_failure_stats(failure_devs_info)

        self.logger.timing_since('timing', start_time)
        if shouldbehere:
            responses.append(True)
        return all(responses), responses

    def delete_db(self, broker):
        object_file = broker.db_file
        hash_dir = os.path.dirname(object_file)
        suf_dir = os.path.dirname(hash_dir)
        with lock_parent_directory(object_file):
            shutil.rmtree(hash_dir, True)
        self.stats['remove'] += 1
        device_name = self.extract_device(object_file)
        self.logger.increment('removes.' + device_name)

        for parent_dir in (suf_dir, os.path.dirname(suf_dir)):
            try:
                os.rmdir(parent_dir)
            except OSError as err:
                if err.errno == errno.ENOTEMPTY:
                    break
                elif err.errno == errno.ENOENT:
                    continue
                else:
                    self.logger.exception(
                        'ERROR while trying to clean up %s', parent_dir)
                    return False
        return True

    def extract_device(self, object_file):
        """
        Extract the device name from an object path.  Returns "UNKNOWN" if the
        path could not be extracted successfully for some reason.

        :param object_file: the path to a database file.
        """
        match = self.extract_device_re.match(object_file)
        if match:
            return match.groups()[0]
        return "UNKNOWN"

    def _partition_dir_filter(self, device_id, partitions_to_replicate):

        def filt(partition_dir):
            partition = int(partition_dir)
            if self.handoffs_only:
                primary_node_ids = [
                    d['id'] for d in self.ring.get_part_nodes(partition)]
                if device_id in primary_node_ids:
                    return False

            if partition not in partitions_to_replicate:
                return False

            return True

        return filt

    def report_up_to_date(self, full_info):
        return True

    def roundrobin_datadirs(self, dirs):
        return RateLimitedIterator(
            roundrobin_datadirs(dirs),
            elements_per_second=self.databases_per_second)

    def run_once(self, *args, **kwargs):
        """Run a replication pass once."""
        override_options = parse_override_options(once=True, **kwargs)

        devices_to_replicate = override_options.devices or Everything()
        partitions_to_replicate = override_options.partitions or Everything()

        self._zero_stats()
        dirs = []
        ips = whataremyips(self.bind_ip)
        if not ips:
            self.logger.error(_('ERROR Failed to get my own IPs?'))
            return

        if self.handoffs_only:
            self.logger.warning(
                'Starting replication pass with handoffs_only enabled. '
                'This mode is not intended for normal '
                'operation; use handoffs_only with care.')

        self._local_device_ids = set()
        found_local = False
        for node in self.ring.devs:
            if node and is_local_device(ips, self.port,
                                        node['replication_ip'],
                                        node['replication_port']):
                found_local = True
                try:
                    dev_path = check_drive(self.root, node['device'],
                                           self.mount_check)
                except ValueError as err:
                    self._add_failure_stats(
                        [(failure_dev['replication_ip'],
                          failure_dev['device'])
                         for failure_dev in self.ring.devs if failure_dev])
                    self.logger.warning('Skipping: %s', err)
                    continue
                if node['device'] not in devices_to_replicate:
                    self.logger.debug(
                        'Skipping device %s due to given arguments',
                        node['device'])
                    continue
                unlink_older_than(
                    os.path.join(dev_path, 'tmp'),
                    time.time() - self.reclaim_age)
                datadir = os.path.join(self.root, node['device'], self.datadir)
                if os.path.isdir(datadir):
                    self._local_device_ids.add(node['id'])
                    part_filt = self._partition_dir_filter(
                        node['id'], partitions_to_replicate)
                    dirs.append((datadir, node['id'], part_filt))
        if not found_local:
            self.logger.error("Can't find itself %s with port %s in ring "
                              "file, not replicating",
                              ", ".join(ips), self.port)
        self.logger.info(_('Beginning replication run'))
        for part, object_file, node_id in self.roundrobin_datadirs(dirs):
            self.cpool.spawn_n(
                self._replicate_object, part, object_file, node_id)
        self.cpool.waitall()
        self.logger.info(_('Replication run OVER'))
        if self.handoffs_only:
            self.logger.warning(
                'Finished replication pass with handoffs_only enabled. '
                'If handoffs_only is no longer required, disable it.')
        self._report_stats()

    def run_forever(self, *args, **kwargs):
        """
        Replicate dbs under the given root in an infinite loop.
        """
        sleep(random.random() * self.interval)
        while True:
            begin = time.time()
            try:
                self.run_once()
            except (Exception, Timeout):
                self.logger.exception(_('ERROR trying to replicate'))
            elapsed = time.time() - begin
            if elapsed < self.interval:
                sleep(self.interval - elapsed)


class ReplicatorRpc(object):
    """Handle Replication RPC calls.  TODO(redbo): document please :)"""

    def __init__(self, root, datadir, broker_class, mount_check=True,
                 logger=None):
        self.root = root
        self.datadir = datadir
        self.broker_class = broker_class
        self.mount_check = mount_check
        self.logger = logger or get_logger({}, log_route='replicator-rpc')

    def _db_file_exists(self, db_path):
        return os.path.exists(db_path)

    def dispatch(self, replicate_args, args):
        if not hasattr(args, 'pop'):
            return HTTPBadRequest(body='Invalid object type')
        op = args.pop(0)
        drive, partition, hsh = replicate_args
        try:
            dev_path = check_drive(self.root, drive, self.mount_check)
        except ValueError:
            return Response(status='507 %s is not mounted' % drive)
        db_file = os.path.join(dev_path,
                               storage_directory(self.datadir, partition, hsh),
                               hsh + '.db')
        if op == 'rsync_then_merge':
            return self.rsync_then_merge(drive, db_file, args)
        if op == 'complete_rsync':
            return self.complete_rsync(drive, db_file, args)
        else:
            # someone might be about to rsync a db to us,
            # make sure there's a tmp dir to receive it.
            mkdirs(os.path.join(self.root, drive, 'tmp'))
            if not self._db_file_exists(db_file):
                return HTTPNotFound()
            return getattr(self, op)(
                self.broker_class(db_file, logger=self.logger), args)

    @contextmanager
    def debug_timing(self, name):
        timemark = time.time()
        yield
        timespan = time.time() - timemark
        if timespan > DEBUG_TIMINGS_THRESHOLD:
            self.logger.debug(
                'replicator-rpc-sync time for %s: %.02fs' % (
                    name, timespan))

    def _parse_sync_args(self, args):
        """
        Convert remote sync args to remote_info dictionary.
        """
        (remote_sync, hash_, id_, created_at, put_timestamp,
         delete_timestamp, metadata) = args[:7]
        remote_metadata = {}
        if metadata:
            try:
                remote_metadata = json.loads(metadata)
            except ValueError:
                self.logger.error("Unable to decode remote metadata %r",
                                  metadata)
        remote_info = {
            'point': remote_sync,
            'hash': hash_,
            'id': id_,
            'created_at': created_at,
            'put_timestamp': put_timestamp,
            'delete_timestamp': delete_timestamp,
            'metadata': remote_metadata,
        }
        return remote_info

    def sync(self, broker, args):
        remote_info = self._parse_sync_args(args)
        return self._handle_sync_request(broker, remote_info)

    def _get_synced_replication_info(self, broker, remote_info):
        """
        Apply any changes to the broker based on remote_info and return the
        current replication info.

        :param broker: the database broker
        :param remote_info: the remote replication info

        :returns: local broker replication info
        """
        return broker.get_replication_info()

    def _handle_sync_request(self, broker, remote_info):
        """
        Update metadata, timestamps, sync points.
        """
        with self.debug_timing('info'):
            try:
                info = self._get_synced_replication_info(broker, remote_info)
            except (Exception, Timeout) as e:
                if 'no such table' in str(e):
                    self.logger.error(_("Quarantining DB %s"), broker)
                    quarantine_db(broker.db_file, broker.db_type)
                    return HTTPNotFound()
                raise
        # TODO(mattoliverau) At this point in the RPC, we have the callers
        # replication info and ours, so it would be cool to be able to make
        # an educated guess here on the size of the incoming replication (maybe
        # average object table row size * difference in ROWIDs or something)
        # and the fallocate_reserve setting so we could return a 507.
        # This would make db fallocate_reserve more or less on par with the
        # object's.
        if remote_info['metadata']:
            with self.debug_timing('update_metadata'):
                broker.update_metadata(remote_info['metadata'])
        sync_timestamps = ('created_at', 'put_timestamp', 'delete_timestamp')
        if any(info[ts] != remote_info[ts] for ts in sync_timestamps):
            with self.debug_timing('merge_timestamps'):
                broker.merge_timestamps(*(remote_info[ts] for ts in
                                          sync_timestamps))
        with self.debug_timing('get_sync'):
            info['point'] = broker.get_sync(remote_info['id'])
        if remote_info['hash'] == info['hash'] and \
                info['point'] < remote_info['point']:
            with self.debug_timing('merge_syncs'):
                translate = {
                    'remote_id': 'id',
                    'sync_point': 'point',
                }
                data = dict((k, remote_info[v]) for k, v in translate.items())
                broker.merge_syncs([data])
                info['point'] = remote_info['point']
        return Response(json.dumps(info))

    def merge_syncs(self, broker, args):
        broker.merge_syncs(args[0])
        return HTTPAccepted()

    def merge_items(self, broker, args):
        broker.merge_items(args[0], args[1])
        return HTTPAccepted()

    def complete_rsync(self, drive, db_file, args):
        old_filename = os.path.join(self.root, drive, 'tmp', args[0])
        if args[1:]:
            db_file = os.path.join(os.path.dirname(db_file), args[1])
        if os.path.exists(db_file):
            return HTTPNotFound()
        if not os.path.exists(old_filename):
            return HTTPNotFound()
        broker = self.broker_class(old_filename, logger=self.logger)
        broker.newid(args[0])
        renamer(old_filename, db_file)
        return HTTPNoContent()

    def _abort_rsync_then_merge(self, db_file, tmp_filename):
        return not (self._db_file_exists(db_file) and
                    os.path.exists(tmp_filename))

    def _post_rsync_then_merge_hook(self, existing_broker, new_broker):
        # subclasses may override to make custom changes to the new broker
        pass

    def rsync_then_merge(self, drive, db_file, args):
        tmp_filename = os.path.join(self.root, drive, 'tmp', args[0])
        if self._abort_rsync_then_merge(db_file, tmp_filename):
            return HTTPNotFound()
        new_broker = self.broker_class(tmp_filename, logger=self.logger)
        existing_broker = self.broker_class(db_file, logger=self.logger)
        db_file = existing_broker.db_file
        point = -1
        objects = existing_broker.get_items_since(point, 1000)
        while len(objects):
            new_broker.merge_items(objects)
            point = objects[-1]['ROWID']
            objects = existing_broker.get_items_since(point, 1000)
            sleep()
        new_broker.merge_syncs(existing_broker.get_syncs())
        self._post_rsync_then_merge_hook(existing_broker, new_broker)
        new_broker.newid(args[0])
        new_broker.update_metadata(existing_broker.metadata)
        if self._abort_rsync_then_merge(db_file, tmp_filename):
            return HTTPNotFound()
        renamer(tmp_filename, db_file)
        return HTTPNoContent()

# Footnote [1]:
#   This orders the nodes so that, given nodes a b c, a will contact b then c,
# b will contact c then a, and c will contact a then b -- in other words, each
# node will always contact the next node in the list first.
#   This helps in the case where databases are all way out of sync, so each
# node is likely to be sending to a different node than it's receiving from,
# rather than two nodes talking to each other, starving out the third.
#   If the third didn't even have a copy and the first two nodes were way out
# of sync, such starvation would mean the third node wouldn't get any copy
# until the first two nodes finally got in sync, which could take a while.
#   This new ordering ensures such starvation doesn't occur, making the data
# more durable.