Merge pull request grpc#9776 from apolcyn/add_http2_flow_control_inte…

…rop_tests

add http2 testing interop server uses small data frames and padding

doc/negative-http2-interop-test-descriptions.md → doc/http2-interop-test-descriptions.md

@@ -193,3 +193,79 @@ Server Procedure:
   1. Sets MAX_CONCURRENT_STREAMS to one after the connection is made.
 
 *The assertion that the MAX_CONCURRENT_STREAMS limit is upheld occurs in the http2 library we used.*
+
+### data_frame_padding
+
+This test verifies that the client can correctly receive padded http2 data
+frames. It also stresses the client's flow control (there is a high chance
+that the sender will deadlock if the client's flow control logic doesn't
+correctly account for padding).
+
+Client Procedure:
+(Note this is the same procedure as in the "large_unary" gRPC interop tests.
+Clients should use their "large_unary" gRPC interop test implementations.)
+Procedure:
+ 1. Client calls UnaryCall with:
+
+    ```
+    {
+      response_size: 314159
+      payload:{
+        body: 271828 bytes of zeros
+      }
+    }
+    ```
+
+Client asserts:
+* call was successful
+* response payload body is 314159 bytes in size
+* clients are free to assert that the response payload body contents are zero
+  and comparing the entire response message against a golden response
+
+Server Procedure:
+  1. Reply to the client's request with a `SimpleResponse`, with a payload
+  body length of `SimpleRequest.response_size`. But send it across specific
+  http2 data frames as follows:
+    * Each http2 data frame contains a 5 byte payload and 255 bytes of padding.
+
+  * Note the 5 byte payload and 255 byte padding are partly arbitrary,
+  and other numbers are also ok. With 255 bytes of padding for each 5 bytes of
+  payload containing actual gRPC message, the 300KB response size will
+  multiply into around 15 megabytes of flow control debt, which should stress
+  flow control accounting.
+
+### no_df_padding_sanity_test
+
+This test verifies that the client can correctly receive a series of small
+data frames. Note that this test is intentionally a slight variation of
+"data_frame_padding", with the only difference being that this test doesn't use data
+frame padding when the response is sent. This test is primarily meant to
+prove correctness of the http2 server implementation and highlight failures
+of the "data_frame_padding" test.
+
+Client Procedure:
+(Note this is the same procedure as in the "large_unary" gRPC interop tests.
+Clients should use their "large_unary" gRPC interop test implementations.)
+Procedure:
+ 1. Client calls UnaryCall with:
+
+    ```
+    {
+      response_size: 314159
+      payload:{
+        body: 271828 bytes of zeros
+      }
+    }
+    ```
+
+Client asserts:
+* call was successful
+* response payload body is 314159 bytes in size
+* clients are free to assert that the response payload body contents are zero
+  and comparing the entire response message against a golden response
+
+Server Procedure:
+  1. Reply to the client's request with a `SimpleResponse`, with a payload
+  body length of `SimpleRequest.response_size`. But send it across series of
+  http2 data frames that contain 5 bytes of "payload" and zero bytes of
+  "padding" (the padding flags on the data frames should not be set).

test/http2_test/http2_base_server.py

@@ -39,6 +39,7 @@
 
 _READ_CHUNK_SIZE = 16384
 _GRPC_HEADER_SIZE = 5
+_MIN_SETTINGS_MAX_FRAME_SIZE = 16384
 
 class H2ProtocolBaseServer(twisted.internet.protocol.Protocol):
   def __init__(self):
@@ -121,38 +122,46 @@ def on_request_received_default(self, event):
     )
     self.transport.write(self._conn.data_to_send())
 
-  def on_window_update_default(self, event):
-    # send pending data, if any
-    self.default_send(event.stream_id)
+  def on_window_update_default(self, _, pad_length=None, read_chunk_size=_READ_CHUNK_SIZE):
+    # try to resume sending on all active streams (update might be for connection)
+    for stream_id in self._send_remaining:
+      self.default_send(stream_id, pad_length=pad_length, read_chunk_size=read_chunk_size)
 
   def send_reset_stream(self):
     self._conn.reset_stream(self._stream_id)
     self.transport.write(self._conn.data_to_send())
 
-  def setup_send(self, data_to_send, stream_id):
+  def setup_send(self, data_to_send, stream_id, pad_length=None, read_chunk_size=_READ_CHUNK_SIZE):
     logging.info('Setting up data to send for stream_id: %d' % stream_id)
     self._send_remaining[stream_id] = len(data_to_send)
     self._send_offset = 0
     self._data_to_send = data_to_send
-    self.default_send(stream_id)
+    self.default_send(stream_id, pad_length=pad_length, read_chunk_size=read_chunk_size)
 
-  def default_send(self, stream_id):
+  def default_send(self, stream_id, pad_length=None, read_chunk_size=_READ_CHUNK_SIZE):
     if not self._send_remaining.has_key(stream_id):
       # not setup to send data yet
       return
 
     while self._send_remaining[stream_id] > 0:
       lfcw = self._conn.local_flow_control_window(stream_id)
-      if lfcw == 0:
+      padding_bytes = pad_length + 1 if pad_length is not None else 0
+      if lfcw - padding_bytes <= 0:
+        logging.info('Stream %d. lfcw: %d. padding bytes: %d. not enough quota yet' % (stream_id, lfcw, padding_bytes))
         break
-      chunk_size = min(lfcw, _READ_CHUNK_SIZE)
+      chunk_size = min(lfcw - padding_bytes, read_chunk_size)
       bytes_to_send = min(chunk_size, self._send_remaining[stream_id])
-      logging.info('flow_control_window = %d. sending [%d:%d] stream_id %d' %
-                    (lfcw, self._send_offset, self._send_offset + bytes_to_send,
-                    stream_id))
+      logging.info('flow_control_window = %d. sending [%d:%d] stream_id %d. includes %d total padding bytes' %
+                    (lfcw, self._send_offset, self._send_offset + bytes_to_send + padding_bytes,
+                    stream_id, padding_bytes))
+      # The receiver might allow sending frames larger than the http2 minimum
+      # max frame size (16384), but this test should never send more than 16384
+      # for simplicity (which is always legal).
+      if bytes_to_send + padding_bytes > _MIN_SETTINGS_MAX_FRAME_SIZE:
+        raise ValueError("overload: sending %d" % (bytes_to_send + padding_bytes))
       data = self._data_to_send[self._send_offset : self._send_offset + bytes_to_send]
       try:
-        self._conn.send_data(stream_id, data, False)
+        self._conn.send_data(stream_id, data, end_stream=False, pad_length=pad_length)
       except h2.exceptions.ProtocolError:
         logging.info('Stream %d is closed' % stream_id)
         break
@@ -200,5 +209,5 @@ def parse_received_data(self, stream_id):
     req_proto_str = recv_buffer[5:5+grpc_msg_size]
     sr = messages_pb2.SimpleRequest()
     sr.ParseFromString(req_proto_str)
-    logging.info('Parsed request for stream %d: response_size=%s' % (stream_id, sr.response_size))
+    logging.info('Parsed simple request for stream %d' % stream_id)
     return sr

test/http2_test/http2_test_server.py

@@ -44,6 +44,7 @@
 import test_rst_after_data
 import test_rst_after_header
 import test_rst_during_data
+import test_data_frame_padding
 
 _TEST_CASE_MAPPING = {
   'rst_after_header': test_rst_after_header.TestcaseRstStreamAfterHeader,
@@ -52,6 +53,10 @@
   'goaway': test_goaway.TestcaseGoaway,
   'ping': test_ping.TestcasePing,
   'max_streams': test_max_streams.TestcaseSettingsMaxStreams,
+
+  # Positive tests below:
+  'data_frame_padding': test_data_frame_padding.TestDataFramePadding,
+  'no_df_padding_sanity_test': test_data_frame_padding.TestDataFramePadding,
 }
 
 _exit_code = 0
@@ -73,6 +78,8 @@ def buildProtocol(self, addr):
 
     if self._testcase == 'goaway':
       return t(self._num_streams).get_base_server()
+    elif self._testcase == 'no_df_padding_sanity_test':
+      return t(use_padding=False).get_base_server()
     else:
       return t().get_base_server()
 
@@ -81,7 +88,8 @@ def parse_arguments():
   parser.add_argument('--base_port', type=int, default=8080,
     help='base port to run the servers (default: 8080). One test server is '
     'started on each incrementing port, beginning with base_port, in the '
-    'following order: goaway,max_streams,ping,rst_after_data,rst_after_header,'
+    'following order: data_frame_padding,goaway,max_streams,'
+    'no_df_padding_sanity_test,ping,rst_after_data,rst_after_header,'
     'rst_during_data'
     )
   return parser.parse_args()

test/http2_test/test_data_frame_padding.py

@@ -0,0 +1,94 @@
+# Copyright 2016, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+import http2_base_server
+import logging
+import messages_pb2
+
+# Set the number of padding bytes per data frame to be very large
+# relative to the number of data bytes for each data frame sent.
+_LARGE_PADDING_LENGTH = 255
+_SMALL_READ_CHUNK_SIZE = 5
+
+class TestDataFramePadding(object):
+  """
+    In response to an incoming request, this test sends headers, followed by
+    data, followed by a reset stream frame. Client asserts that the RPC failed.
+    Client needs to deliver the complete message to the application layer.
+  """
+  def __init__(self, use_padding=True):
+    self._base_server = http2_base_server.H2ProtocolBaseServer()
+    self._base_server._handlers['DataReceived'] = self.on_data_received
+    self._base_server._handlers['WindowUpdated'] = self.on_window_update
+    self._base_server._handlers['RequestReceived'] = self.on_request_received
+
+    # _total_updates maps stream ids to total flow control updates received
+    self._total_updates = {}
+    # zero window updates so far for connection window (stream id '0')
+    self._total_updates[0] = 0
+    self._read_chunk_size = _SMALL_READ_CHUNK_SIZE
+
+    if use_padding:
+      self._pad_length = _LARGE_PADDING_LENGTH
+    else:
+      self._pad_length = None
+
+  def get_base_server(self):
+    return self._base_server
+
+  def on_data_received(self, event):
+    logging.info('on data received. Stream id: %d. Data length: %d' % (event.stream_id, len(event.data)))
+    self._base_server.on_data_received_default(event)
+    if len(event.data) == 0:
+      return
+    sr = self._base_server.parse_received_data(event.stream_id)
+    stream_bytes = ''
+    # Check if full grpc msg has been read into the recv buffer yet
+    if sr:
+      response_data = self._base_server.default_response_data(sr.response_size)
+      logging.info('Stream id: %d. total resp size: %d' % (event.stream_id, len(response_data)))
+      # Begin sending the response. Add ``self._pad_length`` padding to each
+      # data frame and split the whole message into data frames each carrying
+      # only self._read_chunk_size of data.
+      # The purpose is to have the majority of the data frame response bytes
+      # be padding bytes, since ``self._pad_length`` >> ``self._read_chunk_size``.
+      self._base_server.setup_send(response_data , event.stream_id, pad_length=self._pad_length, read_chunk_size=self._read_chunk_size)
+
+  def on_request_received(self, event):
+    self._base_server.on_request_received_default(event)
+    logging.info('on request received. Stream id: %s.' % event.stream_id)
+    self._total_updates[event.stream_id] = 0
+
+  # Log debug info and try to resume sending on all currently active streams.
+  def on_window_update(self, event):
+    logging.info('on window update. Stream id: %s. Delta: %s' % (event.stream_id, event.delta))
+    self._total_updates[event.stream_id] += event.delta
+    total = self._total_updates[event.stream_id]
+    logging.info('... - total updates for stream %d : %d' % (event.stream_id, total))
+    self._base_server.on_window_update_default(event, pad_length=self._pad_length, read_chunk_size=self._read_chunk_size)

tools/doxygen/Doxyfile.c++

@@ -780,11 +780,11 @@ doc/fail_fast.md \
 doc/g_stands_for.md \
 doc/health-checking.md \
 doc/http-grpc-status-mapping.md \
+doc/http2-interop-test-descriptions.md \
 doc/internationalization.md \
 doc/interop-test-descriptions.md \
 doc/load-balancing.md \
 doc/naming.md \
-doc/negative-http2-interop-test-descriptions.md \
 doc/server-reflection.md \
 doc/server_reflection_tutorial.md \
 doc/server_side_auth.md \

tools/doxygen/Doxyfile.c++.internal

@@ -780,11 +780,11 @@ doc/fail_fast.md \
 doc/g_stands_for.md \
 doc/health-checking.md \
 doc/http-grpc-status-mapping.md \
+doc/http2-interop-test-descriptions.md \
 doc/internationalization.md \
 doc/interop-test-descriptions.md \
 doc/load-balancing.md \
 doc/naming.md \
-doc/negative-http2-interop-test-descriptions.md \
 doc/server-reflection.md \
 doc/server_reflection_tutorial.md \
 doc/server_side_auth.md \

tools/doxygen/Doxyfile.core

@@ -780,11 +780,11 @@ doc/fail_fast.md \
 doc/g_stands_for.md \
 doc/health-checking.md \
 doc/http-grpc-status-mapping.md \
+doc/http2-interop-test-descriptions.md \
 doc/internationalization.md \
 doc/interop-test-descriptions.md \
 doc/load-balancing.md \
 doc/naming.md \
-doc/negative-http2-interop-test-descriptions.md \
 doc/server-reflection.md \
 doc/server_reflection_tutorial.md \
 doc/server_side_auth.md \

tools/doxygen/Doxyfile.core.internal

@@ -780,11 +780,11 @@ doc/fail_fast.md \
 doc/g_stands_for.md \
 doc/health-checking.md \
 doc/http-grpc-status-mapping.md \
+doc/http2-interop-test-descriptions.md \
 doc/internationalization.md \
 doc/interop-test-descriptions.md \
 doc/load-balancing.md \
 doc/naming.md \
-doc/negative-http2-interop-test-descriptions.md \
 doc/server-reflection.md \
 doc/server_reflection_tutorial.md \
 doc/server_side_auth.md \

tools/internal_ci/linux/grpc_interop_badserver_java.sh

@@ -37,5 +37,5 @@ cd $(dirname $0)/../../..
 
 git submodule update --init
 
-tools/run_tests/run_interop_tests.py -l java --use_docker --http2_badserver_interop $@
+tools/run_tests/run_interop_tests.py -l java --use_docker --http2_server_interop $@
 

tools/internal_ci/linux/grpc_interop_badserver_python.sh

@@ -37,5 +37,5 @@ cd $(dirname $0)/../../..
 
 git submodule update --init
 
-tools/run_tests/run_interop_tests.py -l python --use_docker --http2_badserver_interop $@
+tools/run_tests/run_interop_tests.py -l python --use_docker --http2_server_interop $@
 

tools/jenkins/run_interop.sh

@@ -36,4 +36,4 @@ export LANG=en_US.UTF-8
 # Enter the gRPC repo root
 cd $(dirname $0)/../..
 
-tools/run_tests/run_interop_tests.py -l all -s all --cloud_to_prod --cloud_to_prod_auth --use_docker --http2_interop --http2_badserver_interop -t -j 12 $@ || true
+tools/run_tests/run_interop_tests.py -l all -s all --cloud_to_prod --cloud_to_prod_auth --use_docker --http2_interop --http2_server_interop -t -j 12 $@ || true

tools/run_tests/interop/interop_html_report.template

@@ -106,19 +106,19 @@
   % endfor
 % endif
 
-% if http2_badserver_cases:
-  <h2>HTTP/2 Bad Server Tests</h2>
+% if http2_server_cases:
+  <h2>HTTP/2 Server Tests</h2>
   ## Each column header is the client language.
   <table style="width:100%" border="1">
   <tr bgcolor="#00BFFF">
   <th>Client languages &#9658;<br/>Test Cases &#9660;</th>
-  % for client_lang in client_langs_http2_badserver_cases:
+  % for client_lang in client_langs:
     <th>${client_lang}</th>
   % endfor
   </tr>
-  % for test_case in http2_badserver_cases:
+  % for test_case in http2_server_cases:
     <tr><td><b>${test_case}</b></td>
-    % for client_lang in client_langs_http2_badserver_cases:
+    % for client_lang in client_langs:
       <% 
         shortname = 'cloud_to_cloud:%s:http2_server:%s' % (client_lang, 
                                                            test_case)