Skip to content

Commit

Permalink
Merge pull request matrix-org#5836 from matrix-org/erikj/lower_bound_…
Browse files Browse the repository at this point in the history
…ttl_well_known

Add a lower bound to well-known TTL.
  • Loading branch information
erikjohnston authored Aug 13, 2019
2 parents 2546f32 + f218705 commit e9906b0
Show file tree
Hide file tree
Showing 4 changed files with 222 additions and 171 deletions.
1 change: 1 addition & 0 deletions changelog.d/5836.misc
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
Add a lower bound to well-known lookup cache time to avoid repeated lookups.
162 changes: 13 additions & 149 deletions synapse/http/federation/matrix_federation_agent.py
Original file line number Diff line number Diff line change
Expand Up @@ -12,10 +12,8 @@
# 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 logging
import random
import time

import attr
from netaddr import IPAddress
Expand All @@ -24,31 +22,16 @@
from twisted.internet import defer
from twisted.internet.endpoints import HostnameEndpoint, wrapClientTLS
from twisted.internet.interfaces import IStreamClientEndpoint
from twisted.web.client import URI, Agent, HTTPConnectionPool, RedirectAgent, readBody
from twisted.web.http import stringToDatetime
from twisted.web.client import URI, Agent, HTTPConnectionPool
from twisted.web.http_headers import Headers
from twisted.web.iweb import IAgent

from synapse.http.federation.srv_resolver import SrvResolver, pick_server_from_list
from synapse.http.federation.well_known_resolver import WellKnownResolver
from synapse.logging.context import make_deferred_yieldable
from synapse.util import Clock
from synapse.util.caches.ttlcache import TTLCache
from synapse.util.metrics import Measure

# period to cache .well-known results for by default
WELL_KNOWN_DEFAULT_CACHE_PERIOD = 24 * 3600

# jitter to add to the .well-known default cache ttl
WELL_KNOWN_DEFAULT_CACHE_PERIOD_JITTER = 10 * 60

# period to cache failure to fetch .well-known for
WELL_KNOWN_INVALID_CACHE_PERIOD = 1 * 3600

# cap for .well-known cache period
WELL_KNOWN_MAX_CACHE_PERIOD = 48 * 3600

logger = logging.getLogger(__name__)
well_known_cache = TTLCache("well-known")


@implementer(IAgent)
Expand Down Expand Up @@ -78,7 +61,7 @@ def __init__(
reactor,
tls_client_options_factory,
_srv_resolver=None,
_well_known_cache=well_known_cache,
_well_known_cache=None,
):
self._reactor = reactor
self._clock = Clock(reactor)
Expand All @@ -93,20 +76,15 @@ def __init__(
self._pool.maxPersistentPerHost = 5
self._pool.cachedConnectionTimeout = 2 * 60

_well_known_agent = RedirectAgent(
Agent(
self._well_known_resolver = WellKnownResolver(
self._reactor,
agent=Agent(
self._reactor,
pool=self._pool,
contextFactory=tls_client_options_factory,
)
),
well_known_cache=_well_known_cache,
)
self._well_known_agent = _well_known_agent

# our cache of .well-known lookup results, mapping from server name
# to delegated name. The values can be:
# `bytes`: a valid server-name
# `None`: there is no (valid) .well-known here
self._well_known_cache = _well_known_cache

@defer.inlineCallbacks
def request(self, method, uri, headers=None, bodyProducer=None):
Expand Down Expand Up @@ -217,7 +195,10 @@ def _route_matrix_uri(self, parsed_uri, lookup_well_known=True):

if lookup_well_known:
# try a .well-known lookup
well_known_server = yield self._get_well_known(parsed_uri.host)
well_known_result = yield self._well_known_resolver.get_well_known(
parsed_uri.host
)
well_known_server = well_known_result.delegated_server

if well_known_server:
# if we found a .well-known, start again, but don't do another
Expand Down Expand Up @@ -280,85 +261,6 @@ def _route_matrix_uri(self, parsed_uri, lookup_well_known=True):
target_port=port,
)

@defer.inlineCallbacks
def _get_well_known(self, server_name):
"""Attempt to fetch and parse a .well-known file for the given server
Args:
server_name (bytes): name of the server, from the requested url
Returns:
Deferred[bytes|None]: either the new server name, from the .well-known, or
None if there was no .well-known file.
"""
try:
result = self._well_known_cache[server_name]
except KeyError:
# TODO: should we linearise so that we don't end up doing two .well-known
# requests for the same server in parallel?
with Measure(self._clock, "get_well_known"):
result, cache_period = yield self._do_get_well_known(server_name)

if cache_period > 0:
self._well_known_cache.set(server_name, result, cache_period)

return result

@defer.inlineCallbacks
def _do_get_well_known(self, server_name):
"""Actually fetch and parse a .well-known, without checking the cache
Args:
server_name (bytes): name of the server, from the requested url
Returns:
Deferred[Tuple[bytes|None|object],int]:
result, cache period, where result is one of:
- the new server name from the .well-known (as a `bytes`)
- None if there was no .well-known file.
- INVALID_WELL_KNOWN if the .well-known was invalid
"""
uri = b"https://%s/.well-known/matrix/server" % (server_name,)
uri_str = uri.decode("ascii")
logger.info("Fetching %s", uri_str)
try:
response = yield make_deferred_yieldable(
self._well_known_agent.request(b"GET", uri)
)
body = yield make_deferred_yieldable(readBody(response))
if response.code != 200:
raise Exception("Non-200 response %s" % (response.code,))

parsed_body = json.loads(body.decode("utf-8"))
logger.info("Response from .well-known: %s", parsed_body)
if not isinstance(parsed_body, dict):
raise Exception("not a dict")
if "m.server" not in parsed_body:
raise Exception("Missing key 'm.server'")
except Exception as e:
logger.info("Error fetching %s: %s", uri_str, e)

# add some randomness to the TTL to avoid a stampeding herd every hour
# after startup
cache_period = WELL_KNOWN_INVALID_CACHE_PERIOD
cache_period += random.uniform(0, WELL_KNOWN_DEFAULT_CACHE_PERIOD_JITTER)
return (None, cache_period)

result = parsed_body["m.server"].encode("ascii")

cache_period = _cache_period_from_headers(
response.headers, time_now=self._reactor.seconds
)
if cache_period is None:
cache_period = WELL_KNOWN_DEFAULT_CACHE_PERIOD
# add some randomness to the TTL to avoid a stampeding herd every 24 hours
# after startup
cache_period += random.uniform(0, WELL_KNOWN_DEFAULT_CACHE_PERIOD_JITTER)
else:
cache_period = min(cache_period, WELL_KNOWN_MAX_CACHE_PERIOD)

return (result, cache_period)


@implementer(IStreamClientEndpoint)
class LoggingHostnameEndpoint(object):
Expand All @@ -374,44 +276,6 @@ def connect(self, protocol_factory):
return self.ep.connect(protocol_factory)


def _cache_period_from_headers(headers, time_now=time.time):
cache_controls = _parse_cache_control(headers)

if b"no-store" in cache_controls:
return 0

if b"max-age" in cache_controls:
try:
max_age = int(cache_controls[b"max-age"])
return max_age
except ValueError:
pass

expires = headers.getRawHeaders(b"expires")
if expires is not None:
try:
expires_date = stringToDatetime(expires[-1])
return expires_date - time_now()
except ValueError:
# RFC7234 says 'A cache recipient MUST interpret invalid date formats,
# especially the value "0", as representing a time in the past (i.e.,
# "already expired").
return 0

return None


def _parse_cache_control(headers):
cache_controls = {}
for hdr in headers.getRawHeaders(b"cache-control", []):
for directive in hdr.split(b","):
splits = [x.strip() for x in directive.split(b"=", 1)]
k = splits[0].lower()
v = splits[1] if len(splits) > 1 else None
cache_controls[k] = v
return cache_controls


@attr.s
class _RoutingResult(object):
"""The result returned by `_route_matrix_uri`.
Expand Down
Loading

0 comments on commit e9906b0

Please sign in to comment.