def check_content_type_is_json(headers):
"""
Check that a set of HTTP headers have a Content-Type header, and that it
is application/json.
Args:
headers (twisted.web.http_headers.Headers): headers to check
Raises:
RequestSendFailed: if the Content-Type header is missing or isn't JSON
"""
c_type = headers.getRawHeaders(b"Content-Type")
if c_type is None:
raise RequestSendFailed(RuntimeError("No Content-Type header"),
can_retry=False)
c_type = c_type[0].decode("ascii") # only the first header
val, options = cgi.parse_header(c_type)
if val != "application/json":
raise RequestSendFailed(
RuntimeError("Content-Type not application/json: was '%s'" %
c_type),
can_retry=False,
)
def check_content_type_is(headers: Headers,
expected_content_type: str) -> None:
"""
Check that a set of HTTP headers have a Content-Type header, and that it
is the expected value..
Args:
headers: headers to check
Raises:
RequestSendFailed: if the Content-Type header is missing or doesn't match
"""
content_type_headers = headers.getRawHeaders(b"Content-Type")
if content_type_headers is None:
raise RequestSendFailed(
RuntimeError("No Content-Type header received from remote server"),
can_retry=False,
)
c_type = content_type_headers[0].decode("ascii") # only the first header
val, options = cgi.parse_header(c_type)
if val != expected_content_type:
raise RequestSendFailed(
RuntimeError(
f"Remote server sent Content-Type header of '{c_type}', not '{expected_content_type}'",
),
can_retry=False,
)
def check_content_type_is_json(headers: Headers) -> None:
"""
Check that a set of HTTP headers have a Content-Type header, and that it
is application/json.
Args:
headers: headers to check
Raises:
RequestSendFailed: if the Content-Type header is missing or isn't JSON
"""
c_type = headers.getRawHeaders(b"Content-Type")
if c_type is None:
raise RequestSendFailed(
RuntimeError("No Content-Type header received from remote server"),
can_retry=False,
)
c_type = c_type[0].decode("ascii") # only the first header
val, options = cgi.parse_header(c_type)
if val != "application/json":
raise RequestSendFailed(
RuntimeError(
"Remote server sent Content-Type header of '%s', not 'application/json'"
% c_type, ),
can_retry=False,
)
async def _handle_json_response(
reactor: IReactorTime,
timeout_sec: float,
request: MatrixFederationRequest,
response: IResponse,
start_ms: int,
):
"""
Reads the JSON body of a response, with a timeout
Args:
reactor: twisted reactor, for the timeout
timeout_sec: number of seconds to wait for response to complete
request: the request that triggered the response
response: response to the request
start_ms: Timestamp when request was made
Returns:
dict: parsed JSON response
"""
try:
check_content_type_is_json(response.headers)
d = treq.json_content(response)
d = timeout_deferred(d, timeout=timeout_sec, reactor=reactor)
body = await make_deferred_yieldable(d)
except TimeoutError as e:
logger.warning(
"{%s} [%s] Timed out reading response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except Exception as e:
logger.warning(
"{%s} [%s] Error reading response %s %s: %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
e,
)
raise
time_taken_secs = reactor.seconds() - start_ms / 1000
logger.info(
"{%s} [%s] Completed request: %d %s in %.2f secs - %s %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
time_taken_secs,
request.method,
request.uri.decode("ascii"),
)
return body
def _handle_json_response(reactor, timeout_sec, request, response):
"""
Reads the JSON body of a response, with a timeout
Args:
reactor (IReactor): twisted reactor, for the timeout
timeout_sec (float): number of seconds to wait for response to complete
request (MatrixFederationRequest): the request that triggered the response
response (IResponse): response to the request
Returns:
dict: parsed JSON response
"""
try:
check_content_type_is_json(response.headers)
d = treq.json_content(response)
d = timeout_deferred(d, timeout=timeout_sec, reactor=reactor)
body = yield make_deferred_yieldable(d)
except TimeoutError as e:
logger.warning(
"{%s} [%s] Timed out reading response",
request.txn_id,
request.destination,
)
raise RequestSendFailed(e, can_retry=True) from e
except Exception as e:
logger.warning(
"{%s} [%s] Error reading response: %s",
request.txn_id,
request.destination,
e,
)
raise
logger.info(
"{%s} [%s] Completed: %d %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
)
return body
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 _handle_response(
reactor: IReactorTime,
timeout_sec: float,
request: MatrixFederationRequest,
response: IResponse,
start_ms: int,
parser: ByteParser[T],
max_response_size: Optional[int] = None,
) -> T:
"""
Reads the body of a response with a timeout and sends it to a parser
Args:
reactor: twisted reactor, for the timeout
timeout_sec: number of seconds to wait for response to complete
request: the request that triggered the response
response: response to the request
start_ms: Timestamp when request was made
parser: The parser for the response
max_response_size: The maximum size to read from the response, if None
uses the default.
Returns:
The parsed response
"""
if max_response_size is None:
max_response_size = MAX_RESPONSE_SIZE
try:
check_content_type_is(response.headers, parser.CONTENT_TYPE)
d = read_body_with_max_size(response, parser, max_response_size)
d = timeout_deferred(d, timeout=timeout_sec, reactor=reactor)
length = await make_deferred_yieldable(d)
value = parser.finish()
except BodyExceededMaxSize as e:
# The response was too big.
logger.warning(
"{%s} [%s] JSON response exceeded max size %i - %s %s",
request.txn_id,
request.destination,
MAX_RESPONSE_SIZE,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=False) from e
except ValueError as e:
# The content was invalid.
logger.warning(
"{%s} [%s] Failed to parse response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=False) from e
except defer.TimeoutError as e:
logger.warning(
"{%s} [%s] Timed out reading response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except ResponseFailed as e:
logger.warning(
"{%s} [%s] Failed to read response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except Exception as e:
logger.warning(
"{%s} [%s] Error reading response %s %s: %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
e,
)
raise
time_taken_secs = reactor.seconds() - start_ms / 1000
logger.info(
"{%s} [%s] Completed request: %d %s in %.2f secs, got %d bytes - %s %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
time_taken_secs,
length,
request.method,
request.uri.decode("ascii"),
)
return value
async def get_file(
self,
destination: str,
path: str,
output_stream,
args: Optional[QueryArgs] = None,
retry_on_dns_fail: bool = True,
max_size: Optional[int] = None,
ignore_backoff: bool = False,
) -> Tuple[int, Dict[bytes, List[bytes]]]:
"""GETs a file from a given homeserver
Args:
destination: The remote server to send the HTTP request to.
path: The HTTP path to GET.
output_stream: File to write the response body to.
args: Optional dictionary used to create the query string.
ignore_backoff: true to ignore the historical backoff data
and try the request anyway.
Returns:
Resolves with an (int,dict) tuple of
the file length and a dict of the response headers.
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.
"""
request = MatrixFederationRequest(method="GET",
destination=destination,
path=path,
query=args)
response = await self._send_request(
request,
retry_on_dns_fail=retry_on_dns_fail,
ignore_backoff=ignore_backoff)
headers = dict(response.headers.getAllRawHeaders())
try:
d = read_body_with_max_size(response, output_stream, max_size)
d.addTimeout(self.default_timeout, self.reactor)
length = await make_deferred_yieldable(d)
except BodyExceededMaxSize:
msg = "Requested file is too large > %r bytes" % (max_size, )
logger.warning(
"{%s} [%s] %s",
request.txn_id,
request.destination,
msg,
)
raise SynapseError(502, msg, Codes.TOO_LARGE)
except defer.TimeoutError as e:
logger.warning(
"{%s} [%s] Timed out reading response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except ResponseFailed as e:
logger.warning(
"{%s} [%s] Failed to read response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except Exception as e:
logger.warning(
"{%s} [%s] Error reading response: %s",
request.txn_id,
request.destination,
e,
)
raise
logger.info(
"{%s} [%s] Completed: %d %s [%d bytes] %s %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
length,
request.method,
request.uri.decode("ascii"),
)
return length, headers
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). 60s by default.
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: resolves with the http response object on success.
Fails with ``HttpResponseException``: if we get an HTTP response
code >= 300 (except 429).
Fails with ``NotRetryingDestination`` if we are not yet ready
to retry this server.
Fails with ``FederationDeniedError`` if this destination
is not on our federation whitelist
Fails with ``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""
headers_dict = {
b"User-Agent": [self.version_string_bytes],
b"Host": [destination_bytes],
}
with limiter:
# 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"]
self.sign_request(
destination_bytes, method_bytes, url_to_sign_bytes,
headers_dict, json,
)
data = encode_canonical_json(json)
producer = FileBodyProducer(
BytesIO(data),
cooperator=self._cooperator,
)
else:
producer = None
self.sign_request(
destination_bytes, method_bytes, url_to_sign_bytes,
headers_dict,
)
logger.info(
"{%s} [%s] Sending request: %s %s",
request.txn_id, request.destination, request.method,
url_str,
)
# 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.hs.get_reactor(),
)
try:
with Measure(self.clock, "outbound_request"):
response = yield make_deferred_yieldable(
request_deferred,
)
except DNSLookupError as e:
raise_from(RequestSendFailed(e, can_retry=retry_on_dns_fail), e)
except Exception as 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'),
)
if 200 <= response.code < 300:
pass
else:
# :'(
# Update transactions table?
d = treq.content(response)
d = timeout_deferred(
d,
timeout=_sec_timeout,
reactor=self.hs.get_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
defer.returnValue(response)
async def _handle_json_response(
reactor: IReactorTime,
timeout_sec: float,
request: MatrixFederationRequest,
response: IResponse,
start_ms: int,
) -> JsonDict:
"""
Reads the JSON body of a response, with a timeout
Args:
reactor: twisted reactor, for the timeout
timeout_sec: number of seconds to wait for response to complete
request: the request that triggered the response
response: response to the request
start_ms: Timestamp when request was made
Returns:
The parsed JSON response
"""
try:
check_content_type_is_json(response.headers)
buf = StringIO()
d = read_body_with_max_size(response, BinaryIOWrapper(buf),
MAX_RESPONSE_SIZE)
d = timeout_deferred(d, timeout=timeout_sec, reactor=reactor)
def parse(_len: int):
return json_decoder.decode(buf.getvalue())
d.addCallback(parse)
body = await make_deferred_yieldable(d)
except BodyExceededMaxSize as e:
# The response was too big.
logger.warning(
"{%s} [%s] JSON response exceeded max size %i - %s %s",
request.txn_id,
request.destination,
MAX_RESPONSE_SIZE,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=False) from e
except ValueError as e:
# The JSON content was invalid.
logger.warning(
"{%s} [%s] Failed to parse JSON response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=False) from e
except defer.TimeoutError as e:
logger.warning(
"{%s} [%s] Timed out reading response - %s %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
)
raise RequestSendFailed(e, can_retry=True) from e
except Exception as e:
logger.warning(
"{%s} [%s] Error reading response %s %s: %s",
request.txn_id,
request.destination,
request.method,
request.uri.decode("ascii"),
e,
)
raise
time_taken_secs = reactor.seconds() - start_ms / 1000
logger.info(
"{%s} [%s] Completed request: %d %s in %.2f secs - %s %s",
request.txn_id,
request.destination,
response.code,
response.phrase.decode("ascii", errors="replace"),
time_taken_secs,
request.method,
request.uri.decode("ascii"),
)
return body
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