def request(self, method, uri, data=None, headers=None):
"""
Args:
method (str): HTTP method to use.
uri (str): URI to query.
data (bytes): Data to send in the request body, if applicable.
headers (t.w.http_headers.Headers): Request headers.
Raises:
SynapseError: If the IP is blacklisted.
"""
# A small wrapper around self.agent.request() so we can easily attach
# counters to it
outgoing_requests_counter.labels(method).inc()
# log request but strip `access_token` (AS requests for example include this)
logger.info("Sending request %s %s", method, redact_uri(uri))
try:
body_producer = None
if data is not None:
body_producer = QuieterFileBodyProducer(BytesIO(data))
request_deferred = treq.request(method,
uri,
agent=self.agent,
data=body_producer,
headers=headers,
**self._extra_treq_args)
request_deferred = timeout_deferred(
request_deferred,
60,
self.hs.get_reactor(),
cancelled_to_request_timed_out_error,
)
response = yield make_deferred_yieldable(request_deferred)
incoming_responses_counter.labels(method, response.code).inc()
logger.info("Received response to %s %s: %s", method,
redact_uri(uri), response.code)
defer.returnValue(response)
except Exception as e:
incoming_responses_counter.labels(method, "ERR").inc()
logger.info(
"Error sending request to %s %s: %s %s",
method,
redact_uri(uri),
type(e).__name__,
e.args[0],
)
raise
async def request(
self,
method: str,
uri: str,
data: Optional[bytes] = None,
headers: Optional[Headers] = None,
) -> IResponse:
"""
Args:
method: HTTP method to use.
uri: URI to query.
data: Data to send in the request body, if applicable.
headers: Request headers.
Returns:
Response object, once the headers have been read.
Raises:
RequestTimedOutError if the request times out before the headers are read
"""
outgoing_requests_counter.labels(method).inc()
# log request but strip `access_token` (AS requests for example include this)
logger.debug("Sending request %s %s", method, redact_uri(uri))
with start_active_span(
"outgoing-client-request",
tags={
tags.SPAN_KIND: tags.SPAN_KIND_RPC_CLIENT,
tags.HTTP_METHOD: method,
tags.HTTP_URL: uri,
},
finish_on_close=True,
):
try:
body_producer = None
if data is not None:
body_producer = QuieterFileBodyProducer(
BytesIO(data),
cooperator=self._cooperator,
)
request_deferred = treq.request(
method,
uri,
agent=self.agent,
data=body_producer,
headers=headers,
**self._extra_treq_args,
) # type: defer.Deferred
# we use our own timeout mechanism rather than treq's as a workaround
# for https://twistedmatrix.com/trac/ticket/9534.
request_deferred = timeout_deferred(
request_deferred,
60,
self.hs.get_reactor(),
)
# turn timeouts into RequestTimedOutErrors
request_deferred.addErrback(
_timeout_to_request_timed_out_error)
response = await make_deferred_yieldable(request_deferred)
incoming_responses_counter.labels(method, response.code).inc()
logger.info(
"Received response to %s %s: %s",
method,
redact_uri(uri),
response.code,
)
return response
except Exception as e:
incoming_responses_counter.labels(method, "ERR").inc()
logger.info(
"Error sending request to %s %s: %s %s",
method,
redact_uri(uri),
type(e).__name__,
e.args[0],
)
set_tag(tags.ERROR, True)
set_tag("error_reason", e.args[0])
raise
async def request(self, method, uri, data=None, headers=None):
"""
Args:
method (str): HTTP method to use.
uri (str): URI to query.
data (bytes): Data to send in the request body, if applicable.
headers (t.w.http_headers.Headers): Request headers.
"""
# A small wrapper around self.agent.request() so we can easily attach
# counters to it
outgoing_requests_counter.labels(method).inc()
# log request but strip `access_token` (AS requests for example include this)
logger.debug("Sending request %s %s", method, redact_uri(uri))
with start_active_span(
"outgoing-client-request",
tags={
tags.SPAN_KIND: tags.SPAN_KIND_RPC_CLIENT,
tags.HTTP_METHOD: method,
tags.HTTP_URL: uri,
},
finish_on_close=True,
):
try:
body_producer = None
if data is not None:
body_producer = QuieterFileBodyProducer(
BytesIO(data),
cooperator=self._cooperator,
)
request_deferred = treq.request(method,
uri,
agent=self.agent,
data=body_producer,
headers=headers,
**self._extra_treq_args)
request_deferred = timeout_deferred(
request_deferred,
60,
self.hs.get_reactor(),
cancelled_to_request_timed_out_error,
)
response = await make_deferred_yieldable(request_deferred)
incoming_responses_counter.labels(method, response.code).inc()
logger.info(
"Received response to %s %s: %s",
method,
redact_uri(uri),
response.code,
)
return response
except Exception as e:
incoming_responses_counter.labels(method, "ERR").inc()
logger.info(
"Error sending request to %s %s: %s %s",
method,
redact_uri(uri),
type(e).__name__,
e.args[0],
)
set_tag(tags.ERROR, True)
set_tag("error_reason", e.args[0])
raise
def _send_request(
self,
request,
retry_on_dns_fail=True,
timeout=None,
long_retries=False,
ignore_backoff=False,
backoff_on_404=False,
):
"""
Sends a request to the given server.
Args:
request (MatrixFederationRequest): details of request to be sent
timeout (int|None): number of milliseconds to wait for the response headers
(including connecting to the server), *for each attempt*.
60s by default.
long_retries (bool): whether to use the long retry algorithm.
The regular retry algorithm makes 4 attempts, with intervals
[0.5s, 1s, 2s].
The long retry algorithm makes 11 attempts, with intervals
[4s, 16s, 60s, 60s, ...]
Both algorithms add -20%/+40% jitter to the retry intervals.
Note that the above intervals are *in addition* to the time spent
waiting for the request to complete (up to `timeout` ms).
NB: the long retry algorithm takes over 20 minutes to complete, with
a default timeout of 60s!
ignore_backoff (bool): true to ignore the historical backoff data
and try the request anyway.
backoff_on_404 (bool): Back off if we get a 404
Returns:
Deferred[twisted.web.client.Response]: resolves with the HTTP
response object on success.
Raises:
HttpResponseException: If we get an HTTP response code >= 300
(except 429).
NotRetryingDestination: If we are not yet ready to retry this
server.
FederationDeniedError: If this destination is not on our
federation whitelist
RequestSendFailed: If there were problems connecting to the
remote, due to e.g. DNS failures, connection timeouts etc.
"""
if timeout:
_sec_timeout = timeout / 1000
else:
_sec_timeout = self.default_timeout
if (
self.hs.config.federation_domain_whitelist is not None
and request.destination not in self.hs.config.federation_domain_whitelist
):
raise FederationDeniedError(request.destination)
limiter = yield synapse.util.retryutils.get_retry_limiter(
request.destination,
self.clock,
self._store,
backoff_on_404=backoff_on_404,
ignore_backoff=ignore_backoff,
)
method_bytes = request.method.encode("ascii")
destination_bytes = request.destination.encode("ascii")
path_bytes = request.path.encode("ascii")
if request.query:
query_bytes = encode_query_args(request.query)
else:
query_bytes = b""
# Retreive current span
scope = opentracing.start_active_span(
"outgoing-federation-request",
tags={
opentracing.tags.SPAN_KIND: opentracing.tags.SPAN_KIND_RPC_CLIENT,
opentracing.tags.PEER_ADDRESS: request.destination,
opentracing.tags.HTTP_METHOD: request.method,
opentracing.tags.HTTP_URL: request.path,
},
finish_on_close=True,
)
# Inject the span into the headers
headers_dict = {}
opentracing.inject_active_span_byte_dict(headers_dict, request.destination)
headers_dict[b"User-Agent"] = [self.version_string_bytes]
with limiter, scope:
# XXX: Would be much nicer to retry only at the transaction-layer
# (once we have reliable transactions in place)
if long_retries:
retries_left = MAX_LONG_RETRIES
else:
retries_left = MAX_SHORT_RETRIES
url_bytes = urllib.parse.urlunparse(
(b"matrix", destination_bytes, path_bytes, None, query_bytes, b"")
)
url_str = url_bytes.decode("ascii")
url_to_sign_bytes = urllib.parse.urlunparse(
(b"", b"", path_bytes, None, query_bytes, b"")
)
while True:
try:
json = request.get_json()
if json:
headers_dict[b"Content-Type"] = [b"application/json"]
auth_headers = self.build_auth_headers(
destination_bytes, method_bytes, url_to_sign_bytes, json
)
data = encode_canonical_json(json)
producer = QuieterFileBodyProducer(
BytesIO(data), cooperator=self._cooperator
)
else:
producer = None
auth_headers = self.build_auth_headers(
destination_bytes, method_bytes, url_to_sign_bytes
)
headers_dict[b"Authorization"] = auth_headers
logger.info(
"{%s} [%s] Sending request: %s %s; timeout %fs",
request.txn_id,
request.destination,
request.method,
url_str,
_sec_timeout,
)
try:
with Measure(self.clock, "outbound_request"):
# we don't want all the fancy cookie and redirect handling
# that treq.request gives: just use the raw Agent.
request_deferred = self.agent.request(
method_bytes,
url_bytes,
headers=Headers(headers_dict),
bodyProducer=producer,
)
request_deferred = timeout_deferred(
request_deferred,
timeout=_sec_timeout,
reactor=self.reactor,
)
response = yield request_deferred
except DNSLookupError as e:
raise_from(RequestSendFailed(e, can_retry=retry_on_dns_fail), e)
except Exception as e:
logger.info("Failed to send request: %s", e)
raise_from(RequestSendFailed(e, can_retry=True), e)
logger.info(
"{%s} [%s] Got response headers: %d %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
)
opentracing.set_tag(
opentracing.tags.HTTP_STATUS_CODE, response.code
)
if 200 <= response.code < 300:
pass
else:
# :'(
# Update transactions table?
d = treq.content(response)
d = timeout_deferred(
d, timeout=_sec_timeout, reactor=self.reactor
)
try:
body = yield make_deferred_yieldable(d)
except Exception as e:
# Eh, we're already going to raise an exception so lets
# ignore if this fails.
logger.warn(
"{%s} [%s] Failed to get error response: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e),
)
body = None
e = HttpResponseException(response.code, response.phrase, body)
# Retry if the error is a 429 (Too Many Requests),
# otherwise just raise a standard HttpResponseException
if response.code == 429:
raise_from(RequestSendFailed(e, can_retry=True), e)
else:
raise e
break
except RequestSendFailed as e:
logger.warn(
"{%s} [%s] Request failed: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e.inner_exception),
)
if not e.can_retry:
raise
if retries_left and not timeout:
if long_retries:
delay = 4 ** (MAX_LONG_RETRIES + 1 - retries_left)
delay = min(delay, 60)
delay *= random.uniform(0.8, 1.4)
else:
delay = 0.5 * 2 ** (MAX_SHORT_RETRIES - retries_left)
delay = min(delay, 2)
delay *= random.uniform(0.8, 1.4)
logger.debug(
"{%s} [%s] Waiting %ss before re-sending...",
request.txn_id,
request.destination,
delay,
)
yield self.clock.sleep(delay)
retries_left -= 1
else:
raise
except Exception as e:
logger.warn(
"{%s} [%s] Request failed: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e),
)
raise
return response
async def request(
self,
method: str,
uri: str,
data: Optional[bytes] = None,
headers: Optional[Headers] = None,
) -> IResponse:
"""
Args:
method: HTTP method to use.
uri: URI to query.
data: Data to send in the request body, if applicable.
headers: Request headers.
Returns:
Response object, once the headers have been read.
Raises:
RequestTimedOutError if the request times out before the headers are read
"""
outgoing_requests_counter.labels(method).inc()
# log request but strip `access_token` (AS requests for example include this)
logger.debug("Sending request %s %s", method, redact_uri(uri))
with start_active_span(
"outgoing-client-request",
tags={
tags.SPAN_KIND: tags.SPAN_KIND_RPC_CLIENT,
tags.HTTP_METHOD: method,
tags.HTTP_URL: uri,
},
finish_on_close=True,
):
try:
body_producer = None
if data is not None:
body_producer = QuieterFileBodyProducer(
BytesIO(data),
cooperator=self._cooperator,
)
from skywalking import Layer, Component
from skywalking.trace import tags as sw_tags
from skywalking.trace.context import get_context
from skywalking.trace.tags import Tag
from yarl import URL
context = get_context()
url = URL(uri).with_user(None).with_password(None)
peer = '%s:%d' % (url.host or '', url.port)
span = context.new_exit_span(op=uri or "/", peer=peer)
span.start()
span.layer = Layer.Http
span.component = Component.General
span.tag(Tag(key=sw_tags.HttpMethod, val=method.upper()))
span.tag(Tag(key=sw_tags.HttpUrl, val=uri))
carrier = span.inject()
for item in carrier:
headers.addRawHeader(item.key, item.val)
request_deferred = treq.request(
method,
uri,
agent=self.agent,
data=body_producer,
headers=headers,
# Avoid buffering the body in treq since we do not reuse
# response bodies.
unbuffered=True,
**self._extra_treq_args,
) # type: defer.Deferred
# we use our own timeout mechanism rather than treq's as a workaround
# for https://twistedmatrix.com/trac/ticket/9534.
request_deferred = timeout_deferred(
request_deferred,
60,
self.hs.get_reactor(),
)
# turn timeouts into RequestTimedOutErrors
request_deferred.addErrback(
_timeout_to_request_timed_out_error)
response = await make_deferred_yieldable(request_deferred)
span.tag(
Tag(key=sw_tags.HttpStatus,
val=response.code,
overridable=True))
if response.code >= 400:
span.error_occurred = True
span.stop()
incoming_responses_counter.labels(method, response.code).inc()
logger.info(
"Received response to %s %s: %s",
method,
redact_uri(uri),
response.code,
)
return response
except Exception as e:
incoming_responses_counter.labels(method, "ERR").inc()
logger.info(
"Error sending request to %s %s: %s %s",
method,
redact_uri(uri),
type(e).__name__,
e.args[0],
)
set_tag(tags.ERROR, True)
set_tag("error_reason", e.args[0])
raise
async def _send_request(
self,
request: MatrixFederationRequest,
retry_on_dns_fail: bool = True,
timeout: Optional[int] = None,
long_retries: bool = False,
ignore_backoff: bool = False,
backoff_on_404: bool = False,
) -> IResponse:
"""
Sends a request to the given server.
Args:
request: details of request to be sent
retry_on_dns_fail: true if the request should be retied on DNS failures
timeout: number of milliseconds to wait for the response headers
(including connecting to the server), *for each attempt*.
60s by default.
long_retries: whether to use the long retry algorithm.
The regular retry algorithm makes 4 attempts, with intervals
[0.5s, 1s, 2s].
The long retry algorithm makes 11 attempts, with intervals
[4s, 16s, 60s, 60s, ...]
Both algorithms add -20%/+40% jitter to the retry intervals.
Note that the above intervals are *in addition* to the time spent
waiting for the request to complete (up to `timeout` ms).
NB: the long retry algorithm takes over 20 minutes to complete, with
a default timeout of 60s!
ignore_backoff: true to ignore the historical backoff data
and try the request anyway.
backoff_on_404: Back off if we get a 404
Returns:
Resolves with the HTTP response object on success.
Raises:
HttpResponseException: If we get an HTTP response code >= 300
(except 429).
NotRetryingDestination: If we are not yet ready to retry this
server.
FederationDeniedError: If this destination is not on our
federation whitelist
RequestSendFailed: If there were problems connecting to the
remote, due to e.g. DNS failures, connection timeouts etc.
"""
if timeout:
_sec_timeout = timeout / 1000
else:
_sec_timeout = self.default_timeout
if (self.hs.config.federation.federation_domain_whitelist is not None
and request.destination
not in self.hs.config.federation.federation_domain_whitelist):
raise FederationDeniedError(request.destination)
limiter = await synapse.util.retryutils.get_retry_limiter(
request.destination,
self.clock,
self._store,
backoff_on_404=backoff_on_404,
ignore_backoff=ignore_backoff,
notifier=self.hs.get_notifier(),
replication_client=self.hs.get_replication_command_handler(),
)
method_bytes = request.method.encode("ascii")
destination_bytes = request.destination.encode("ascii")
path_bytes = request.path.encode("ascii")
query_bytes = encode_query_args(request.query)
scope = start_active_span(
"outgoing-federation-request",
tags={
tags.SPAN_KIND: tags.SPAN_KIND_RPC_CLIENT,
tags.PEER_ADDRESS: request.destination,
tags.HTTP_METHOD: request.method,
tags.HTTP_URL: request.path,
},
finish_on_close=True,
)
# Inject the span into the headers
headers_dict: Dict[bytes, List[bytes]] = {}
opentracing.inject_header_dict(headers_dict, request.destination)
headers_dict[b"User-Agent"] = [self.version_string_bytes]
with limiter, scope:
# XXX: Would be much nicer to retry only at the transaction-layer
# (once we have reliable transactions in place)
if long_retries:
retries_left = MAX_LONG_RETRIES
else:
retries_left = MAX_SHORT_RETRIES
url_bytes = request.uri
url_str = url_bytes.decode("ascii")
url_to_sign_bytes = urllib.parse.urlunparse(
(b"", b"", path_bytes, None, query_bytes, b""))
while True:
try:
json = request.get_json()
if json:
headers_dict[b"Content-Type"] = [b"application/json"]
auth_headers = self.build_auth_headers(
destination_bytes, method_bytes, url_to_sign_bytes,
json)
data = encode_canonical_json(json)
producer: Optional[
IBodyProducer] = QuieterFileBodyProducer(
BytesIO(data), cooperator=self._cooperator)
else:
producer = None
auth_headers = self.build_auth_headers(
destination_bytes, method_bytes, url_to_sign_bytes)
headers_dict[b"Authorization"] = auth_headers
logger.debug(
"{%s} [%s] Sending request: %s %s; timeout %fs",
request.txn_id,
request.destination,
request.method,
url_str,
_sec_timeout,
)
outgoing_requests_counter.labels(request.method).inc()
try:
with Measure(self.clock, "outbound_request"):
# we don't want all the fancy cookie and redirect handling
# that treq.request gives: just use the raw Agent.
# To preserve the logging context, the timeout is treated
# in a similar way to `defer.gatherResults`:
# * Each logging context-preserving fork is wrapped in
# `run_in_background`. In this case there is only one,
# since the timeout fork is not logging-context aware.
# * The `Deferred` that joins the forks back together is
# wrapped in `make_deferred_yieldable` to restore the
# logging context regardless of the path taken.
request_deferred = run_in_background(
self.agent.request,
method_bytes,
url_bytes,
headers=Headers(headers_dict),
bodyProducer=producer,
)
request_deferred = timeout_deferred(
request_deferred,
timeout=_sec_timeout,
reactor=self.reactor,
)
response = await make_deferred_yieldable(
request_deferred)
except DNSLookupError as e:
raise RequestSendFailed(
e, can_retry=retry_on_dns_fail) from e
except Exception as e:
raise RequestSendFailed(e, can_retry=True) from e
incoming_responses_counter.labels(request.method,
response.code).inc()
set_tag(tags.HTTP_STATUS_CODE, response.code)
response_phrase = response.phrase.decode("ascii",
errors="replace")
if 200 <= response.code < 300:
logger.debug(
"{%s} [%s] Got response headers: %d %s",
request.txn_id,
request.destination,
response.code,
response_phrase,
)
else:
logger.info(
"{%s} [%s] Got response headers: %d %s",
request.txn_id,
request.destination,
response.code,
response_phrase,
)
# :'(
# Update transactions table?
d = treq.content(response)
d = timeout_deferred(d,
timeout=_sec_timeout,
reactor=self.reactor)
try:
body = await make_deferred_yieldable(d)
except Exception as e:
# Eh, we're already going to raise an exception so lets
# ignore if this fails.
logger.warning(
"{%s} [%s] Failed to get error response: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e),
)
body = None
exc = HttpResponseException(response.code,
response_phrase, body)
# Retry if the error is a 5xx or a 429 (Too Many
# Requests), otherwise just raise a standard
# `HttpResponseException`
if 500 <= response.code < 600 or response.code == 429:
raise RequestSendFailed(exc,
can_retry=True) from exc
else:
raise exc
break
except RequestSendFailed as e:
logger.info(
"{%s} [%s] Request failed: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e.inner_exception),
)
if not e.can_retry:
raise
if retries_left and not timeout:
if long_retries:
delay = 4**(MAX_LONG_RETRIES + 1 - retries_left)
delay = min(delay, 60)
delay *= random.uniform(0.8, 1.4)
else:
delay = 0.5 * 2**(MAX_SHORT_RETRIES - retries_left)
delay = min(delay, 2)
delay *= random.uniform(0.8, 1.4)
logger.debug(
"{%s} [%s] Waiting %ss before re-sending...",
request.txn_id,
request.destination,
delay,
)
# Sleep for the calculated delay, or wake up immediately
# if we get notified that the server is back up.
await self._sleeper.sleep(request.destination,
delay * 1000)
retries_left -= 1
else:
raise
except Exception as e:
logger.warning(
"{%s} [%s] Request failed: %s %s: %s",
request.txn_id,
request.destination,
request.method,
url_str,
_flatten_response_never_received(e),
)
raise
return response