github-octodns/octodns/provider/dyn.py

#
#
#

from __future__ import absolute_import, division, print_function, \
    unicode_literals

from collections import defaultdict
from dyn.tm.errors import DynectGetError
from dyn.tm.services.dsf import DSFARecord, DSFAAAARecord, DSFFailoverChain, \
    DSFMonitor, DSFNode, DSFRecordSet, DSFResponsePool, DSFRuleset, \
    TrafficDirector, get_all_dsf_monitors, get_all_dsf_services, \
    get_response_pool
from dyn.tm.session import DynectSession
from dyn.tm.zones import Zone as DynZone
from logging import getLogger
from threading import Lock
from uuid import uuid4

from ..record import Record
from .base import BaseProvider


class _CachingDynZone(DynZone):
    log = getLogger('_CachingDynZone')

    _cache = {}

    @classmethod
    def get(cls, zone_name, create=False):
        cls.log.debug('get: zone_name=%s, create=%s', zone_name, create)
        # This works in dyn zone names, without the trailing .
        try:
            dyn_zone = cls._cache[zone_name]
            cls.log.debug('get:   cache hit')
        except KeyError:
            cls.log.debug('get:   cache miss')
            try:
                dyn_zone = _CachingDynZone(zone_name)
                cls.log.debug('get:   fetched')
            except DynectGetError:
                if not create:
                    cls.log.debug("get:   does't exist")
                    return None
                # this value shouldn't really matter, it's not tied to
                # whois or anything
                hostname = 'hostmaster@{}'.format(zone_name[:-1])
                # Try again with the params necessary to create
                dyn_zone = _CachingDynZone(zone_name, ttl=3600,
                                           contact=hostname,
                                           serial_style='increment')
                cls.log.debug('get:   created')
            cls._cache[zone_name] = dyn_zone

        return dyn_zone

    @classmethod
    def flush_zone(cls, zone_name):
        '''Flushes the zone cache, if there is one'''
        cls.log.debug('flush_zone: zone_name=%s', zone_name)
        try:
            del cls._cache[zone_name]
        except KeyError:
            pass

    def __init__(self, zone_name, *args, **kwargs):
        super(_CachingDynZone, self).__init__(zone_name, *args, **kwargs)
        self.flush_cache()

    def flush_cache(self):
        self._cached_records = None

    def get_all_records(self):
        if self._cached_records is None:
            self._cached_records = \
                super(_CachingDynZone, self).get_all_records()
        return self._cached_records

    def publish(self):
        super(_CachingDynZone, self).publish()
        self.flush_cache()


class DynProvider(BaseProvider):
    '''
    Dynect Managed DNS provider

    dyn:
        class: octodns.provider.dyn.DynProvider
        # Your dynect customer name (required)
        customer: cust
        # Your dynect username (required)
        username: user
        # Your dynect password (required)
        password: pass
        # Whether or not to support TrafficDirectors and enable GeoDNS
        # (optional, default is false)
        traffic_directors_enabled: true

    Note: due to the way dyn.tm.session.DynectSession is managing things we can
    only really have a single DynProvider configured. When you create a
    DynectSession it's stored in a thread-local singleton. You don't invoke
    methods on this session or a client that holds on to it. The client
    libraries grab their per-thread session by accessing the singleton through
    DynectSession.get_session(). That fundamentally doesn't support having more
    than one account active at a time. See DynProvider._check_dyn_sess for some
    related bits.
    '''

    RECORDS_TO_TYPE = {
        'a_records': 'A',
        'aaaa_records': 'AAAA',
        'alias_records': 'ALIAS',
        'caa_records': 'CAA',
        'cname_records': 'CNAME',
        'mx_records': 'MX',
        'naptr_records': 'NAPTR',
        'ns_records': 'NS',
        'ptr_records': 'PTR',
        'sshfp_records': 'SSHFP',
        'spf_records': 'SPF',
        'srv_records': 'SRV',
        'txt_records': 'TXT',
    }
    TYPE_TO_RECORDS = {v: k for k, v in RECORDS_TO_TYPE.items()}
    SUPPORTS = set(TYPE_TO_RECORDS.keys())

    # https://help.dyn.com/predefined-geotm-regions-groups/
    REGION_CODES = {
        'NA': 11,  # Continental North America
        'SA': 12,  # Continental South America
        'EU': 13,  # Contentinal Europe
        'AF': 14,  # Continental Africa
        'AS': 15,  # Contentinal Asia
        'OC': 16,  # Contentinal Austrailia/Oceania
        'AN': 17,  # Continental Antartica
    }

    _sess_create_lock = Lock()

    def __init__(self, id, customer, username, password,
                 traffic_directors_enabled=False, *args, **kwargs):
        self.log = getLogger('DynProvider[{}]'.format(id))
        self.log.debug('__init__: id=%s, customer=%s, username=%s, '
                       'password=***, traffic_directors_enabled=%s', id,
                       customer, username, traffic_directors_enabled)
        # we have to set this before calling super b/c SUPPORTS_GEO requires it
        self.traffic_directors_enabled = traffic_directors_enabled
        super(DynProvider, self).__init__(id, *args, **kwargs)
        self.customer = customer
        self.username = username
        self.password = password

        self._cache = {}
        self._traffic_directors = None
        self._traffic_director_monitors = None

    @property
    def SUPPORTS_GEO(self):
        return self.traffic_directors_enabled

    def _check_dyn_sess(self):
        # We don't have to worry about locking for the check since the
        # underlying pieces are pre-thread. We can check to see if this thread
        # has a session and if so we're good to go.
        if DynectSession.get_session() is None:
            # We need to create a new session for this thread and DynectSession
            # creation is not thread-safe so we have to do the locking. If we
            # don't and multiple sessions start creattion before the the first
            # has finished (long time b/c it makes http calls) the subsequent
            # creates will blow away DynectSession._instances, potentially
            # multiple times if there are multiple creates in flight. Only the
            # last of these initial concurrent creates will exist in
            # DynectSession._instances dict and the others will be lost. When
            # this thread later tries to make api calls there won't be an
            # accessible session available for it to use.
            with self._sess_create_lock:
                DynectSession(self.customer, self.username, self.password)

    def _data_for_A(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [r.address for r in records]
        }

    _data_for_AAAA = _data_for_A

    def _data_for_ALIAS(self, _type, records):
        # See note on ttl in _kwargs_for_ALIAS
        record = records[0]
        return {
            'type': _type,
            'ttl': record.ttl,
            'value': record.alias
        }

    def _data_for_CAA(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [{'flags': r.flags, 'tag': r.tag, 'value': r.value}
                       for r in records],
        }

    def _data_for_CNAME(self, _type, records):
        record = records[0]
        return {
            'type': _type,
            'ttl': record.ttl,
            'value': record.cname,
        }

    def _data_for_MX(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [{'preference': r.preference, 'exchange': r.exchange}
                       for r in records],
        }

    def _data_for_NAPTR(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [{
                'order': r.order,
                'preference': r.preference,
                'flags': r.flags,
                'service': r.services,
                'regexp': r.regexp,
                'replacement': r.replacement,
            } for r in records]
        }

    def _data_for_NS(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [r.nsdname for r in records]
        }

    def _data_for_PTR(self, _type, records):
        record = records[0]
        return {
            'type': _type,
            'ttl': record.ttl,
            'value': record.ptrdname,
        }

    def _data_for_SPF(self, _type, records):
        record = records[0]
        return {
            'type': _type,
            'ttl': record.ttl,
            'values': [r.txtdata for r in records]
        }

    _data_for_TXT = _data_for_SPF

    def _data_for_SSHFP(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [{
                'algorithm': r.algorithm,
                'fingerprint_type': r.fptype,
                'fingerprint': r.fingerprint,
            } for r in records],
        }

    def _data_for_SRV(self, _type, records):
        return {
            'type': _type,
            'ttl': records[0].ttl,
            'values': [{
                'priority': r.priority,
                'weight': r.weight,
                'port': r.port,
                'target': r.target,
            } for r in records],
        }

    @property
    def traffic_directors(self):
        if self._traffic_directors is None:
            self._check_dyn_sess()

            tds = defaultdict(dict)
            for td in get_all_dsf_services():
                try:
                    fqdn, _type = td.label.split(':', 1)
                except ValueError as e:
                    self.log.warn("Failed to load TraficDirector '%s': %s",
                                  td.label, e.message)
                    continue
                tds[fqdn][_type] = td
            self._traffic_directors = dict(tds)

        return self._traffic_directors

    def _populate_traffic_directors(self, zone):
        self.log.debug('_populate_traffic_directors: zone=%s', zone.name)
        td_records = set()
        for fqdn, types in self.traffic_directors.items():
            # TODO: skip subzones
            if not fqdn.endswith(zone.name):
                continue

            for _type, td in types.items():
                # critical to call rulesets once, each call loads them :-(
                rulesets = td.rulesets

                # We start out with something that will always change show
                # change in case this is a busted TD. This will prevent us from
                # creating a duplicate td. We'll overwrite this with real data
                # provide we have it
                geo = {}
                data = {
                    'geo': geo,
                    'type': _type,
                    'ttl': td.ttl,
                    'values': ['0.0.0.0']
                }
                for ruleset in rulesets:
                    try:
                        record_set = ruleset.response_pools[0].rs_chains[0] \
                            .record_sets[0]
                    except IndexError:
                        # problems indicate a malformed ruleset, ignore it
                        continue
                    _type = record_set.rdata_class
                    if ruleset.label.startswith('default:'):
                        data_for = getattr(self, '_data_for_{}'.format(_type))
                        data.update(data_for(_type, record_set.records))
                    else:
                        # We've stored the geo in label
                        try:
                            code, _ = ruleset.label.split(':', 1)
                        except ValueError:
                            continue
                        values = [r.address for r in record_set.records]
                        geo[code] = values

                name = zone.hostname_from_fqdn(fqdn)
                record = Record.new(zone, name, data, source=self)
                zone.add_record(record)
                td_records.add(record)

        return td_records

    def populate(self, zone, target=False, lenient=False):
        self.log.debug('populate: name=%s, target=%s, lenient=%s', zone.name,
                       target, lenient)

        before = len(zone.records)

        self._check_dyn_sess()

        td_records = set()
        if self.traffic_directors_enabled:
            td_records = self._populate_traffic_directors(zone)

        dyn_zone = _CachingDynZone.get(zone.name[:-1])

        if dyn_zone:
            values = defaultdict(lambda: defaultdict(list))
            for _type, records in dyn_zone.get_all_records().items():
                if _type == 'soa_records':
                    continue
                _type = self.RECORDS_TO_TYPE[_type]
                for record in records:
                    record_name = zone.hostname_from_fqdn(record.fqdn)
                    values[record_name][_type].append(record)

            for name, types in values.items():
                for _type, records in types.items():
                    data_for = getattr(self, '_data_for_{}'.format(_type))
                    data = data_for(_type, records)
                    record = Record.new(zone, name, data, source=self,
                                        lenient=lenient)
                    if record not in td_records:
                        zone.add_record(record)

        self.log.info('populate:   found %s records',
                      len(zone.records) - before)

    def _kwargs_for_A(self, record):
        return [{
            'address': v,
            'ttl': record.ttl,
        } for v in record.values]

    _kwargs_for_AAAA = _kwargs_for_A

    def _kwargs_for_CAA(self, record):
        return [{
            'flags': v.flags,
            'tag': v.tag,
            'value': v.value,
        } for v in record.values]

    def _kwargs_for_CNAME(self, record):
        return [{
            'cname': record.value,
            'ttl': record.ttl,
        }]

    def _kwargs_for_ALIAS(self, record):
        # NOTE: Dyn's UI doesn't allow editing of ALIAS ttl, but the API seems
        # to accept and store the values we send it just fine. No clue if they
        # do anything with them. I'd assume they just obey the TTL of the
        # record that we're pointed at which makes sense.
        return [{
            'alias': record.value,
            'ttl': record.ttl,
        }]

    def _kwargs_for_MX(self, record):
        return [{
            'preference': v.preference,
            'exchange': v.exchange,
            'ttl': record.ttl,
        } for v in record.values]

    def _kwargs_for_NAPTR(self, record):
        return [{
            'flags': v.flags,
            'order': v.order,
            'preference': v.preference,
            'regexp': v.regexp,
            'replacement': v.replacement,
            'services': v.service,
            'ttl': record.ttl,
        } for v in record.values]

    def _kwargs_for_NS(self, record):
        return [{
            'nsdname': v,
            'ttl': record.ttl,
        } for v in record.values]

    def _kwargs_for_PTR(self, record):
        return [{
            'ptrdname': record.value,
            'ttl': record.ttl,
        }]

    def _kwargs_for_SSHFP(self, record):
        return [{
            'algorithm': v.algorithm,
            'fptype': v.fingerprint_type,
            'fingerprint': v.fingerprint,
        } for v in record.values]

    def _kwargs_for_SPF(self, record):
        return [{
            'txtdata': v,
            'ttl': record.ttl,
        } for v in record.chunked_values]

    def _kwargs_for_SRV(self, record):
        return [{
            'port': v.port,
            'priority': v.priority,
            'target': v.target,
            'weight': v.weight,
            'ttl': record.ttl,
        } for v in record.values]

    _kwargs_for_TXT = _kwargs_for_SPF

    def _traffic_director_monitor(self, fqdn):
        if self._traffic_director_monitors is None:
            self._traffic_director_monitors = \
                {m.label: m for m in get_all_dsf_monitors()}

        try:
            return self._traffic_director_monitors[fqdn]
        except KeyError:
            monitor = DSFMonitor(fqdn, protocol='HTTPS', response_count=2,
                                 probe_interval=60, retries=2, port=443,
                                 active='Y', host=fqdn[:-1], timeout=10,
                                 header='User-Agent: Dyn Monitor',
                                 path='/_dns')
            self._traffic_director_monitors[fqdn] = monitor
            return monitor

    def _find_or_create_pool(self, td, pools, label, _type, values,
                             monitor_id=None):
        for pool in pools:
            if pool.label != label:
                continue
            records = pool.rs_chains[0].record_sets[0].records
            record_values = sorted([r.address for r in records])
            if record_values == values:
                # it's a match
                return pool
        # we need to create the pool
        _class = {
            'A': DSFARecord,
            'AAAA': DSFAAAARecord
        }[_type]
        records = [_class(v) for v in values]
        record_set = DSFRecordSet(_type, label, serve_count=len(records),
                                  records=records, dsf_monitor_id=monitor_id)
        chain = DSFFailoverChain(label, record_sets=[record_set])
        pool = DSFResponsePool(label, rs_chains=[chain])
        pool.create(td)
        return pool

    def _mod_rulesets(self, td, change):
        new = change.new

        # Response Pools
        pools = {}

        # Get existing pools. This should be simple, but it's not b/c the dyn
        # api is a POS. We need all response pools so we can GC and check to
        # make sure that what we're after doesn't already exist.
        # td.all_response_pools just returns thin objects that don't include
        # their rs_chains (and children down to actual records.) We could just
        # foreach over those turning them into full DSFResponsePool objects
        # with get_response_pool, but that'd be N round-trips. We can avoid
        # those round trips in cases where the pools are in use in rules where
        # they're already full objects.

        # First up populate all the full pools we have under rules, the _
        # prevents a td.refresh we don't need :-( seriously?
        existing_rulesets = td._rulesets
        for ruleset in existing_rulesets:
            for pool in ruleset.response_pools:
                pools[pool.response_pool_id] = pool
        # Now we need to find any pools that aren't referenced by rules
        for pool in td.all_response_pools:
            rpid = pool.response_pool_id
            if rpid not in pools:
                # we want this one, but it's thin, inflate it
                pools[rpid] = get_response_pool(rpid, td)
        # now that we have full objects for the complete set of existing pools,
        # a list will be more useful
        pools = pools.values()

        # Rulesets

        # add the default
        label = 'default:{}'.format(uuid4().hex)
        ruleset = DSFRuleset(label, 'always', [])
        ruleset.create(td, index=0)
        pool = self._find_or_create_pool(td, pools, 'default', new._type,
                                         new.values)
        # There's no way in the client lib to create a ruleset with an existing
        # pool (ref'd by id) so we have to do this round-a-bout.
        active_pools = {
            'default': pool.response_pool_id
        }
        ruleset.add_response_pool(pool.response_pool_id)

        monitor_id = self._traffic_director_monitor(new.fqdn).dsf_monitor_id
        # Geos ordered least to most specific so that parents will always be
        # created before their children (and thus can be referenced
        geos = sorted(new.geo.items(), key=lambda d: d[0])
        for _, geo in geos:
            if geo.subdivision_code:
                criteria = {
                    'province': geo.subdivision_code.lower()
                }
            elif geo.country_code:
                criteria = {
                    'country': geo.country_code
                }
            else:
                criteria = {
                    'region': self.REGION_CODES[geo.continent_code]
                }

            label = '{}:{}'.format(geo.code, uuid4().hex)
            ruleset = DSFRuleset(label, 'geoip', [], {
                'geoip': criteria
            })
            # Something you have to call create others the constructor does it
            ruleset.create(td, index=0)

            first = geo.values[0]
            pool = self._find_or_create_pool(td, pools, first, new._type,
                                             geo.values, monitor_id)
            active_pools[geo.code] = pool.response_pool_id
            ruleset.add_response_pool(pool.response_pool_id)

            # look for parent rulesets we can add in the chain
            for code in geo.parents:
                try:
                    pool_id = active_pools[code]
                    # looking at client lib code, index > exists appends
                    ruleset.add_response_pool(pool_id, index=999)
                except KeyError:
                    pass
            # and always add default as the last
            pool_id = active_pools['default']
            ruleset.add_response_pool(pool_id, index=999)

        # we're done with active_pools as a lookup, convert it in to a set of
        # the ids in use
        active_pools = set(active_pools.values())
        # Clean up unused response_pools
        for pool in pools:
            if pool.response_pool_id in active_pools:
                continue
            pool.delete()

        # Clean out the old rulesets
        for ruleset in existing_rulesets:
            ruleset.delete()

    def _mod_geo_Create(self, dyn_zone, change):
        new = change.new
        fqdn = new.fqdn
        _type = new._type
        label = '{}:{}'.format(fqdn, _type)
        node = DSFNode(new.zone.name, fqdn)
        td = TrafficDirector(label, ttl=new.ttl, nodes=[node], publish='Y')
        self.log.debug('_mod_geo_Create: td=%s', td.service_id)
        self._mod_rulesets(td, change)
        self.traffic_directors[fqdn] = {
            _type: td
        }

    def _mod_geo_Update(self, dyn_zone, change):
        new = change.new
        if not new.geo:
            # New record doesn't have geo we're going from a TD to a regular
            # record
            self._mod_Create(dyn_zone, change)
            self._mod_geo_Delete(dyn_zone, change)
            return
        try:
            td = self.traffic_directors[new.fqdn][new._type]
        except KeyError:
            # There's no td, this is actually a create, we must be going from a
            # non-geo to geo record so delete the regular record as well
            self._mod_geo_Create(dyn_zone, change)
            self._mod_Delete(dyn_zone, change)
            return
        self._mod_rulesets(td, change)

    def _mod_geo_Delete(self, dyn_zone, change):
        existing = change.existing
        fqdn_tds = self.traffic_directors[existing.fqdn]
        _type = existing._type
        fqdn_tds[_type].delete()
        del fqdn_tds[_type]

    def _mod_Create(self, dyn_zone, change):
        new = change.new
        kwargs_for = getattr(self, '_kwargs_for_{}'.format(new._type))
        for kwargs in kwargs_for(new):
            dyn_zone.add_record(new.name, new._type, **kwargs)

    def _mod_Delete(self, dyn_zone, change):
        existing = change.existing
        if existing.name:
            target = '{}.{}'.format(existing.name, existing.zone.name[:-1])
        else:
            target = existing.zone.name[:-1]
        _type = self.TYPE_TO_RECORDS[existing._type]
        for rec in dyn_zone.get_all_records()[_type]:
            if rec.fqdn == target:
                rec.delete()

    def _mod_Update(self, dyn_zone, change):
        self._mod_Delete(dyn_zone, change)
        self._mod_Create(dyn_zone, change)

    def _apply_traffic_directors(self, desired, changes, dyn_zone):
        self.log.debug('_apply_traffic_directors: zone=%s', desired.name)
        unhandled_changes = []
        for c in changes:
            # we only mess with changes that have geo info somewhere
            if getattr(c.new, 'geo', False) or getattr(c.existing, 'geo',
                                                       False):
                mod = getattr(self, '_mod_geo_{}'.format(c.__class__.__name__))
                mod(dyn_zone, c)
            else:
                unhandled_changes.append(c)

        return unhandled_changes

    def _apply_regular(self, desired, changes, dyn_zone):
        self.log.debug('_apply_regular: zone=%s', desired.name)
        for c in changes:
            mod = getattr(self, '_mod_{}'.format(c.__class__.__name__))
            mod(dyn_zone, c)

    # TODO: detect "extra" changes when monitors are out of date or failover
    # chains are wrong etc.

    def _apply(self, plan):
        desired = plan.desired
        changes = plan.changes
        self.log.debug('_apply: zone=%s, len(changes)=%d', desired.name,
                       len(changes))

        self._check_dyn_sess()

        dyn_zone = _CachingDynZone.get(desired.name[:-1], create=True)

        if self.traffic_directors_enabled:
            # any changes left over don't involve geo
            changes = self._apply_traffic_directors(desired, changes, dyn_zone)

        self._apply_regular(desired, changes, dyn_zone)

        dyn_zone.publish()