diff options
Diffstat (limited to 'test/cpp/microbenchmarks')
-rw-r--r-- | test/cpp/microbenchmarks/BUILD | 93 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_call_create.cc | 28 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_chttp2_hpack.cc | 163 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_chttp2_transport.cc | 81 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_closure.cc | 3 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_cq.cc | 14 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_error.cc | 2 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_fullstack_streaming_ping_pong.cc | 194 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_fullstack_streaming_pump.cc | 3 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_fullstack_trickle.cc | 2 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_fullstack_unary_ping_pong.cc | 3 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/bm_metadata.cc | 17 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/helpers.h | 2 | ||||
-rw-r--r-- | test/cpp/microbenchmarks/noop-benchmark.cc | 2 |
14 files changed, 540 insertions, 67 deletions
diff --git a/test/cpp/microbenchmarks/BUILD b/test/cpp/microbenchmarks/BUILD new file mode 100644 index 0000000000..38619666dc --- /dev/null +++ b/test/cpp/microbenchmarks/BUILD @@ -0,0 +1,93 @@ +# Copyright 2017, 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. + +licenses(["notice"]) # 3-clause BSD + +cc_test( + name = "noop-benchmark", + srcs = ["noop-benchmark.cc"], + deps = ["//external:benchmark"], + linkopts = ["-pthread"], +) + +cc_library( + name = "helpers", + srcs = ["helpers.cc"], + hdrs = ["helpers.h", "fullstack_fixtures.h", "fullstack_context_mutators.h"], + deps = ["//:grpc++", "//external:benchmark", "//test/core/util:grpc_test_util", "//src/proto/grpc/testing:echo_proto"], + linkopts = ["-pthread"], +) + +cc_test( + name = "bm_closure", + srcs = ["bm_closure.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_cq", + srcs = ["bm_cq.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_error", + srcs = ["bm_error.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_fullstack_streaming_ping_pong", + srcs = ["bm_fullstack_streaming_ping_pong.cc"], + deps = [":helpers"], + + ) +cc_test( + name = "bm_fullstack_streaming_pump", + srcs = ["bm_fullstack_streaming_pump.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_fullstack_trickle", + srcs = ["bm_fullstack_trickle.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_fullstack_unary_ping_pong", + srcs = ["bm_fullstack_unary_ping_pong.cc"], + deps = [":helpers"], +) + +cc_test( + name = "bm_metadata", + srcs = ["bm_metadata.cc"], + deps = [":helpers"], +) diff --git a/test/cpp/microbenchmarks/bm_call_create.cc b/test/cpp/microbenchmarks/bm_call_create.cc index 1ef8caa690..cc37f0c9e9 100644 --- a/test/cpp/microbenchmarks/bm_call_create.cc +++ b/test/cpp/microbenchmarks/bm_call_create.cc @@ -34,6 +34,7 @@ /* This benchmark exists to ensure that the benchmark integration is * working */ +#include <benchmark/benchmark.h> #include <string.h> #include <sstream> @@ -44,8 +45,8 @@ #include <grpc/support/string_util.h> extern "C" { -#include "src/core/ext/client_channel/client_channel.h" -#include "src/core/ext/load_reporting/load_reporting_filter.h" +#include "src/core/ext/filters/client_channel/client_channel.h" +#include "src/core/ext/filters/load_reporting/load_reporting_filter.h" #include "src/core/lib/channel/channel_stack.h" #include "src/core/lib/channel/compress_filter.h" #include "src/core/lib/channel/connected_channel.h" @@ -60,7 +61,6 @@ extern "C" { #include "src/cpp/client/create_channel_internal.h" #include "src/proto/grpc/testing/echo.grpc.pb.h" #include "test/cpp/microbenchmarks/helpers.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" auto &force_library_initialization = Library::get(); @@ -221,7 +221,7 @@ namespace dummy_filter { static void StartTransportStreamOp(grpc_exec_ctx *exec_ctx, grpc_call_element *elem, - grpc_transport_stream_op *op) {} + grpc_transport_stream_op_batch *op) {} static void StartTransportOp(grpc_exec_ctx *exec_ctx, grpc_channel_element *elem, @@ -296,7 +296,7 @@ void SetPollsetSet(grpc_exec_ctx *exec_ctx, grpc_transport *self, /* implementation of grpc_transport_perform_stream_op */ void PerformStreamOp(grpc_exec_ctx *exec_ctx, grpc_transport *self, - grpc_stream *stream, grpc_transport_stream_op *op) { + grpc_stream *stream, grpc_transport_stream_op_batch *op) { grpc_closure_sched(exec_ctx, op->on_complete, GRPC_ERROR_NONE); } @@ -346,13 +346,15 @@ class SendEmptyMetadata { memset(&op_, 0, sizeof(op_)); op_.on_complete = grpc_closure_init(&closure_, DoNothing, nullptr, grpc_schedule_on_exec_ctx); + op_.send_initial_metadata = true; + op_.payload = &op_payload_; } class Op { public: Op(grpc_exec_ctx *exec_ctx, SendEmptyMetadata *p, grpc_call_stack *s) { grpc_metadata_batch_init(&batch_); - p->op_.send_initial_metadata = &batch_; + p->op_payload_.send_initial_metadata.send_initial_metadata = &batch_; } void Finish(grpc_exec_ctx *exec_ctx) { grpc_metadata_batch_destroy(exec_ctx, &batch_); @@ -366,7 +368,8 @@ class SendEmptyMetadata { const gpr_timespec deadline_ = gpr_inf_future(GPR_CLOCK_MONOTONIC); const gpr_timespec start_time_ = gpr_now(GPR_CLOCK_MONOTONIC); const grpc_slice method_ = grpc_slice_from_static_string("/foo/bar"); - grpc_transport_stream_op op_; + grpc_transport_stream_op_batch op_; + grpc_transport_stream_op_batch_payload op_payload_; grpc_closure closure_; }; @@ -488,13 +491,16 @@ namespace isolated_call_filter { static void StartTransportStreamOp(grpc_exec_ctx *exec_ctx, grpc_call_element *elem, - grpc_transport_stream_op *op) { + grpc_transport_stream_op_batch *op) { if (op->recv_initial_metadata) { - grpc_closure_sched(exec_ctx, op->recv_initial_metadata_ready, - GRPC_ERROR_NONE); + grpc_closure_sched( + exec_ctx, + op->payload->recv_initial_metadata.recv_initial_metadata_ready, + GRPC_ERROR_NONE); } if (op->recv_message) { - grpc_closure_sched(exec_ctx, op->recv_message_ready, GRPC_ERROR_NONE); + grpc_closure_sched(exec_ctx, op->payload->recv_message.recv_message_ready, + GRPC_ERROR_NONE); } grpc_closure_sched(exec_ctx, op->on_complete, GRPC_ERROR_NONE); } diff --git a/test/cpp/microbenchmarks/bm_chttp2_hpack.cc b/test/cpp/microbenchmarks/bm_chttp2_hpack.cc index 563db758f7..55d2d2f58d 100644 --- a/test/cpp/microbenchmarks/bm_chttp2_hpack.cc +++ b/test/cpp/microbenchmarks/bm_chttp2_hpack.cc @@ -33,6 +33,7 @@ /* Microbenchmarks around CHTTP2 HPACK operations */ +#include <grpc/support/alloc.h> #include <grpc/support/log.h> #include <string.h> #include <sstream> @@ -40,6 +41,7 @@ extern "C" { #include "src/core/ext/transport/chttp2/transport/hpack_encoder.h" #include "src/core/ext/transport/chttp2/transport/hpack_parser.h" #include "src/core/lib/slice/slice_internal.h" +#include "src/core/lib/slice/slice_string_helpers.h" #include "src/core/lib/transport/static_metadata.h" } #include "test/cpp/microbenchmarks/helpers.h" @@ -69,6 +71,7 @@ template <class Fixture> static void BM_HpackEncoderEncodeHeader(benchmark::State &state) { TrackCounters track_counters; grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT; + static bool logged_representative_output = false; grpc_metadata_batch b; grpc_metadata_batch_init(&b); @@ -87,8 +90,17 @@ static void BM_HpackEncoderEncodeHeader(benchmark::State &state) { grpc_slice_buffer outbuf; grpc_slice_buffer_init(&outbuf); while (state.KeepRunning()) { - grpc_chttp2_encode_header(&exec_ctx, &c, (uint32_t)state.iterations(), &b, - state.range(0), state.range(1), &stats, &outbuf); + uint32_t stream_id = static_cast<uint32_t>(state.iterations()); + grpc_chttp2_encode_header(&exec_ctx, &c, stream_id, &b, state.range(0), + state.range(1), &stats, &outbuf); + if (!logged_representative_output) { + logged_representative_output = true; + for (size_t i = 0; i < outbuf.count; i++) { + char *s = grpc_dump_slice(outbuf.slices[i], GPR_DUMP_HEX); + gpr_log(GPR_DEBUG, "%" PRIdPTR ": %s", i, s); + gpr_free(s); + } + } grpc_slice_buffer_reset_and_unref_internal(&exec_ctx, &outbuf); grpc_exec_ctx_flush(&exec_ctx); } @@ -131,6 +143,28 @@ class SingleInternedElem { } }; +template <int kLength> +class SingleInternedBinaryElem { + public: + static std::vector<grpc_mdelem> GetElems(grpc_exec_ctx *exec_ctx) { + grpc_slice bytes = MakeBytes(); + std::vector<grpc_mdelem> out = {grpc_mdelem_from_slices( + exec_ctx, grpc_slice_intern(grpc_slice_from_static_string("abc-bin")), + grpc_slice_intern(bytes))}; + grpc_slice_unref(bytes); + return out; + } + + private: + static grpc_slice MakeBytes() { + std::vector<char> v; + for (int i = 0; i < kLength; i++) { + v.push_back(static_cast<char>(rand())); + } + return grpc_slice_from_copied_buffer(v.data(), v.size()); + } +}; + class SingleInternedKeyElem { public: static std::vector<grpc_mdelem> GetElems(grpc_exec_ctx *exec_ctx) { @@ -149,6 +183,24 @@ class SingleNonInternedElem { } }; +template <int kLength> +class SingleNonInternedBinaryElem { + public: + static std::vector<grpc_mdelem> GetElems(grpc_exec_ctx *exec_ctx) { + return {grpc_mdelem_from_slices( + exec_ctx, grpc_slice_from_static_string("abc-bin"), MakeBytes())}; + } + + private: + static grpc_slice MakeBytes() { + std::vector<char> v; + for (int i = 0; i < kLength; i++) { + v.push_back(static_cast<char>(rand())); + } + return grpc_slice_from_copied_buffer(v.data(), v.size()); + } +}; + class RepresentativeClientInitialMetadata { public: static std::vector<grpc_mdelem> GetElems(grpc_exec_ctx *exec_ctx) { @@ -195,8 +247,29 @@ BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedKeyElem) ->Args({0, 16384}); BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedElem) ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedBinaryElem<1>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedBinaryElem<3>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedBinaryElem<10>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedBinaryElem<31>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleInternedBinaryElem<100>) + ->Args({0, 16384}); BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedElem) ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedBinaryElem<1>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedBinaryElem<3>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedBinaryElem<10>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedBinaryElem<31>) + ->Args({0, 16384}); +BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, + SingleNonInternedBinaryElem<100>) + ->Args({0, 16384}); // test with a tiny frame size, to highlight continuation costs BENCHMARK_TEMPLATE(BM_HpackEncoderEncodeHeader, SingleNonInternedElem) ->Args({0, 1}); @@ -255,6 +328,8 @@ static void BM_HpackParserParseHeader(benchmark::State &state) { } grpc_exec_ctx_flush(&exec_ctx); } + for (auto slice : init_slices) grpc_slice_unref(slice); + for (auto slice : benchmark_slices) grpc_slice_unref(slice); grpc_chttp2_hpack_parser_destroy(&exec_ctx, &p); grpc_exec_ctx_finish(&exec_ctx); track_counters.Finish(state); @@ -262,7 +337,7 @@ static void BM_HpackParserParseHeader(benchmark::State &state) { namespace hpack_parser_fixtures { -static grpc_slice MakeSlice(std::initializer_list<uint8_t> bytes) { +static grpc_slice MakeSlice(std::vector<uint8_t> bytes) { grpc_slice s = grpc_slice_malloc(bytes.size()); uint8_t *p = GRPC_SLICE_START_PTR(s); for (auto b : bytes) { @@ -346,6 +421,64 @@ class NonIndexedElem { } }; +class NonIndexedBinaryElem1 { + public: + static std::vector<grpc_slice> GetInitSlices() { return {}; } + static std::vector<grpc_slice> GetBenchmarkSlices() { + return {MakeSlice( + {0x00, 0x07, 'a', 'b', 'c', '-', 'b', 'i', 'n', 0x82, 0xf7, 0xb3})}; + } +}; + +class NonIndexedBinaryElem3 { + public: + static std::vector<grpc_slice> GetInitSlices() { return {}; } + static std::vector<grpc_slice> GetBenchmarkSlices() { + return {MakeSlice({0x00, 0x07, 'a', 'b', 'c', '-', 'b', 'i', 'n', 0x84, + 0x7f, 0x4e, 0x29, 0x3f})}; + } +}; + +class NonIndexedBinaryElem10 { + public: + static std::vector<grpc_slice> GetInitSlices() { return {}; } + static std::vector<grpc_slice> GetBenchmarkSlices() { + return {MakeSlice({0x00, 0x07, 'a', 'b', 'c', '-', 'b', + 'i', 'n', 0x8b, 0x71, 0x0c, 0xa5, 0x81, + 0x73, 0x7b, 0x47, 0x13, 0xe9, 0xf7, 0xe3})}; + } +}; + +class NonIndexedBinaryElem31 { + public: + static std::vector<grpc_slice> GetInitSlices() { return {}; } + static std::vector<grpc_slice> GetBenchmarkSlices() { + return {MakeSlice({0x00, 0x07, 'a', 'b', 'c', '-', 'b', 'i', 'n', + 0xa3, 0x92, 0x43, 0x7f, 0xbe, 0x7c, 0xea, 0x6f, 0xf3, + 0x3d, 0xa7, 0xa7, 0x67, 0xfb, 0xe2, 0x82, 0xf7, 0xf2, + 0x8f, 0x1f, 0x9d, 0xdf, 0xf1, 0x7e, 0xb3, 0xef, 0xb2, + 0x8f, 0x53, 0x77, 0xce, 0x0c, 0x13, 0xe3, 0xfd, 0x87})}; + } +}; + +class NonIndexedBinaryElem100 { + public: + static std::vector<grpc_slice> GetInitSlices() { return {}; } + static std::vector<grpc_slice> GetBenchmarkSlices() { + return {MakeSlice( + {0x00, 0x07, 'a', 'b', 'c', '-', 'b', 'i', 'n', 0xeb, 0x1d, 0x4d, + 0xe8, 0x96, 0x8c, 0x14, 0x20, 0x06, 0xc1, 0xc3, 0xdf, 0x6e, 0x1f, 0xef, + 0xde, 0x2f, 0xde, 0xb7, 0xf2, 0xfe, 0x6d, 0xd4, 0xe4, 0x7d, 0xf5, 0x55, + 0x46, 0x52, 0x3d, 0x91, 0xf2, 0xd4, 0x6f, 0xca, 0x34, 0xcd, 0xd9, 0x39, + 0xbd, 0x03, 0x27, 0xe3, 0x9c, 0x74, 0xcc, 0x17, 0x34, 0xed, 0xa6, 0x6a, + 0x77, 0x73, 0x10, 0xcd, 0x8e, 0x4e, 0x5c, 0x7c, 0x72, 0x39, 0xd8, 0xe6, + 0x78, 0x6b, 0xdb, 0xa5, 0xb7, 0xab, 0xe7, 0x46, 0xae, 0x21, 0xab, 0x7f, + 0x01, 0x89, 0x13, 0xd7, 0xca, 0x17, 0x6e, 0xcb, 0xd6, 0x79, 0x71, 0x68, + 0xbf, 0x8a, 0x3f, 0x32, 0xe8, 0xba, 0xf5, 0xbe, 0xb3, 0xbc, 0xde, 0x28, + 0xc7, 0xcf, 0x62, 0x7a, 0x58, 0x2c, 0xcf, 0x4d, 0xe3})}; + } +}; + class RepresentativeClientInitialMetadata { public: static std::vector<grpc_slice> GetInitSlices() { @@ -437,6 +570,11 @@ BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, IndexedSingleInternedElem); BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, AddIndexedSingleInternedElem); BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, KeyIndexedSingleInternedElem); BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedElem); +BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedBinaryElem1); +BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedBinaryElem3); +BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedBinaryElem10); +BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedBinaryElem31); +BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, NonIndexedBinaryElem100); BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, RepresentativeClientInitialMetadata); BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, @@ -446,4 +584,23 @@ BENCHMARK_TEMPLATE(BM_HpackParserParseHeader, } // namespace hpack_parser_fixtures +static void BM_Base16SomeStuff(benchmark::State &state) { + uint8_t *bytes = new uint8_t[state.range(0)]; + for (int i = 0; i < state.range(0); i++) { + bytes[i] = static_cast<uint8_t>(rand()); + } + uint8_t *encoded = new uint8_t[state.range(0) * 2]; + static const uint8_t hex[] = "0123456789abcdef"; + while (state.KeepRunning()) { + for (int i = 0; i < state.range(0); i++) { + encoded[2 * i + 0] = hex[encoded[i] >> 8]; + encoded[2 * i + 1] = hex[encoded[i] & 0xf]; + } + } + delete[] encoded; + delete[] bytes; + state.SetBytesProcessed(state.iterations() * state.range(0)); +} +BENCHMARK(BM_Base16SomeStuff)->Range(1, 4096); + BENCHMARK_MAIN(); diff --git a/test/cpp/microbenchmarks/bm_chttp2_transport.cc b/test/cpp/microbenchmarks/bm_chttp2_transport.cc index 265a5a6b22..1611250850 100644 --- a/test/cpp/microbenchmarks/bm_chttp2_transport.cc +++ b/test/cpp/microbenchmarks/bm_chttp2_transport.cc @@ -207,7 +207,7 @@ class Stream { static_cast<grpc_stream *>(stream_), closure); } - void Op(grpc_transport_stream_op *op) { + void Op(grpc_transport_stream_op_batch *op) { grpc_transport_perform_stream_op(f_->exec_ctx(), f_->transport(), static_cast<grpc_stream *>(stream_), op); } @@ -305,10 +305,16 @@ static void BM_StreamCreateSendInitialMetadataDestroy(benchmark::State &state) { TrackCounters track_counters; Fixture f(grpc::ChannelArguments(), true); Stream s(&f); - grpc_transport_stream_op op; + grpc_transport_stream_op_batch op; + grpc_transport_stream_op_batch_payload op_payload; std::unique_ptr<Closure> start; std::unique_ptr<Closure> done; + auto reset_op = [&]() { + memset(&op, 0, sizeof(op)); + op.payload = &op_payload; + }; + grpc_metadata_batch b; grpc_metadata_batch_init(&b); b.deadline = gpr_inf_future(GPR_CLOCK_MONOTONIC); @@ -324,14 +330,16 @@ static void BM_StreamCreateSendInitialMetadataDestroy(benchmark::State &state) { start = MakeClosure([&](grpc_exec_ctx *exec_ctx, grpc_error *error) { if (!state.KeepRunning()) return; s.Init(state); - memset(&op, 0, sizeof(op)); + reset_op(); op.on_complete = done.get(); - op.send_initial_metadata = &b; + op.send_initial_metadata = true; + op.payload->send_initial_metadata.send_initial_metadata = &b; s.Op(&op); }); done = MakeClosure([&](grpc_exec_ctx *exec_ctx, grpc_error *error) { - memset(&op, 0, sizeof(op)); - op.cancel_error = GRPC_ERROR_CANCELLED; + reset_op(); + op.cancel_stream = true; + op.payload->cancel_stream.cancel_error = GRPC_ERROR_CANCELLED; s.Op(&op); s.DestroyThen(start.get()); }); @@ -348,11 +356,16 @@ static void BM_TransportEmptyOp(benchmark::State &state) { Fixture f(grpc::ChannelArguments(), true); Stream s(&f); s.Init(state); - grpc_transport_stream_op op; + grpc_transport_stream_op_batch op; + grpc_transport_stream_op_batch_payload op_payload; + auto reset_op = [&]() { + memset(&op, 0, sizeof(op)); + op.payload = &op_payload; + }; std::unique_ptr<Closure> c = MakeClosure([&](grpc_exec_ctx *exec_ctx, grpc_error *error) { if (!state.KeepRunning()) return; - memset(&op, 0, sizeof(op)); + reset_op(); op.on_complete = c.get(); s.Op(&op); }); @@ -370,7 +383,12 @@ static void BM_TransportStreamSend(benchmark::State &state) { Fixture f(grpc::ChannelArguments(), true); Stream s(&f); s.Init(state); - grpc_transport_stream_op op; + grpc_transport_stream_op_batch op; + grpc_transport_stream_op_batch_payload op_payload; + auto reset_op = [&]() { + memset(&op, 0, sizeof(op)); + op.payload = &op_payload; + }; grpc_slice_buffer_stream send_stream; grpc_slice_buffer send_buffer; grpc_slice_buffer_init(&send_buffer); @@ -397,20 +415,23 @@ static void BM_TransportStreamSend(benchmark::State &state) { s.chttp2_stream()->outgoing_window_delta = 1024 * 1024 * 1024; f.chttp2_transport()->outgoing_window = 1024 * 1024 * 1024; grpc_slice_buffer_stream_init(&send_stream, &send_buffer, 0); - memset(&op, 0, sizeof(op)); + reset_op(); op.on_complete = c.get(); - op.send_message = &send_stream.base; + op.send_message = true; + op.payload->send_message.send_message = &send_stream.base; s.Op(&op); }); - memset(&op, 0, sizeof(op)); - op.send_initial_metadata = &b; + reset_op(); + op.send_initial_metadata = true; + op.payload->send_initial_metadata.send_initial_metadata = &b; op.on_complete = c.get(); s.Op(&op); f.FlushExecCtx(); - memset(&op, 0, sizeof(op)); - op.cancel_error = GRPC_ERROR_CANCELLED; + reset_op(); + op.cancel_stream = true; + op.payload->cancel_stream.cancel_error = GRPC_ERROR_CANCELLED; s.Op(&op); s.DestroyThen( MakeOnceClosure([](grpc_exec_ctx *exec_ctx, grpc_error *error) {})); @@ -483,10 +504,16 @@ static void BM_TransportStreamRecv(benchmark::State &state) { Fixture f(grpc::ChannelArguments(), true); Stream s(&f); s.Init(state); - grpc_transport_stream_op op; + grpc_transport_stream_op_batch_payload op_payload; + grpc_transport_stream_op_batch op; grpc_byte_stream *recv_stream; grpc_slice incoming_data = CreateIncomingDataSlice(state.range(0), 16384); + auto reset_op = [&]() { + memset(&op, 0, sizeof(op)); + op.payload = &op_payload; + }; + grpc_metadata_batch b; grpc_metadata_batch_init(&b); grpc_metadata_batch b_recv; @@ -518,10 +545,11 @@ static void BM_TransportStreamRecv(benchmark::State &state) { s.chttp2_stream()->incoming_window_delta = 1024 * 1024 * 1024; f.chttp2_transport()->incoming_window = 1024 * 1024 * 1024; received = 0; - memset(&op, 0, sizeof(op)); + reset_op(); op.on_complete = do_nothing.get(); - op.recv_message = &recv_stream; - op.recv_message_ready = drain_start.get(); + op.recv_message = true; + op.payload->recv_message.recv_message = &recv_stream; + op.payload->recv_message.recv_message_ready = drain_start.get(); s.Op(&op); f.PushInput(grpc_slice_ref(incoming_data)); }); @@ -554,9 +582,13 @@ static void BM_TransportStreamRecv(benchmark::State &state) { grpc_closure_run(exec_ctx, drain.get(), GRPC_ERROR_NONE); }); - memset(&op, 0, sizeof(op)); - op.send_initial_metadata = &b; - op.recv_initial_metadata = &b_recv; + reset_op(); + op.send_initial_metadata = true; + op.payload->send_initial_metadata.send_initial_metadata = &b; + op.recv_initial_metadata = true; + op.payload->recv_initial_metadata.recv_initial_metadata = &b_recv; + op.payload->recv_initial_metadata.recv_initial_metadata_ready = + do_nothing.get(); op.on_complete = c.get(); s.Op(&op); f.PushInput(SLICE_FROM_BUFFER( @@ -573,8 +605,9 @@ static void BM_TransportStreamRecv(benchmark::State &state) { "\x10\x14grpc-accept-encoding\x15identity,deflate,gzip")); f.FlushExecCtx(); - memset(&op, 0, sizeof(op)); - op.cancel_error = GRPC_ERROR_CANCELLED; + reset_op(); + op.cancel_stream = true; + op.payload->cancel_stream.cancel_error = GRPC_ERROR_CANCELLED; s.Op(&op); s.DestroyThen( MakeOnceClosure([](grpc_exec_ctx *exec_ctx, grpc_error *error) {})); diff --git a/test/cpp/microbenchmarks/bm_closure.cc b/test/cpp/microbenchmarks/bm_closure.cc index 28a385b6c1..d52fe4ee30 100644 --- a/test/cpp/microbenchmarks/bm_closure.cc +++ b/test/cpp/microbenchmarks/bm_closure.cc @@ -33,7 +33,9 @@ /* Test various closure related operations */ +#include <benchmark/benchmark.h> #include <grpc/grpc.h> +#include <sstream> extern "C" { #include "src/core/lib/iomgr/closure.h" @@ -43,7 +45,6 @@ extern "C" { } #include "test/cpp/microbenchmarks/helpers.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" auto& force_library_initialization = Library::get(); diff --git a/test/cpp/microbenchmarks/bm_cq.cc b/test/cpp/microbenchmarks/bm_cq.cc index 91e6a85101..38ac9d2705 100644 --- a/test/cpp/microbenchmarks/bm_cq.cc +++ b/test/cpp/microbenchmarks/bm_cq.cc @@ -34,12 +34,11 @@ /* This benchmark exists to ensure that the benchmark integration is * working */ +#include <benchmark/benchmark.h> #include <grpc++/completion_queue.h> #include <grpc++/impl/grpc_library.h> #include <grpc/grpc.h> - #include "test/cpp/microbenchmarks/helpers.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" extern "C" { #include "src/core/lib/surface/completion_queue.h" @@ -59,6 +58,17 @@ static void BM_CreateDestroyCpp(benchmark::State& state) { } BENCHMARK(BM_CreateDestroyCpp); +/* Create cq using a different constructor */ +static void BM_CreateDestroyCpp2(benchmark::State& state) { + TrackCounters track_counters; + while (state.KeepRunning()) { + grpc_completion_queue* core_cq = grpc_completion_queue_create(NULL); + CompletionQueue cq(core_cq); + } + track_counters.Finish(state); +} +BENCHMARK(BM_CreateDestroyCpp2); + static void BM_CreateDestroyCore(benchmark::State& state) { TrackCounters track_counters; while (state.KeepRunning()) { diff --git a/test/cpp/microbenchmarks/bm_error.cc b/test/cpp/microbenchmarks/bm_error.cc index 00e1a08cab..ea9777bbe6 100644 --- a/test/cpp/microbenchmarks/bm_error.cc +++ b/test/cpp/microbenchmarks/bm_error.cc @@ -33,6 +33,7 @@ /* Test various operations on grpc_error */ +#include <benchmark/benchmark.h> #include <memory> extern "C" { @@ -41,7 +42,6 @@ extern "C" { } #include "test/cpp/microbenchmarks/helpers.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" auto& force_library_initialization = Library::get(); diff --git a/test/cpp/microbenchmarks/bm_fullstack_streaming_ping_pong.cc b/test/cpp/microbenchmarks/bm_fullstack_streaming_ping_pong.cc index 00e37f7912..c536e15a2c 100644 --- a/test/cpp/microbenchmarks/bm_fullstack_streaming_ping_pong.cc +++ b/test/cpp/microbenchmarks/bm_fullstack_streaming_ping_pong.cc @@ -33,14 +33,13 @@ /* Benchmark gRPC end2end in various configurations */ +#include <benchmark/benchmark.h> #include <sstream> - #include "src/core/lib/profiling/timers.h" #include "src/cpp/client/create_channel_internal.h" #include "src/proto/grpc/testing/echo.grpc.pb.h" #include "test/cpp/microbenchmarks/fullstack_context_mutators.h" #include "test/cpp/microbenchmarks/fullstack_fixtures.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" namespace grpc { namespace testing { @@ -240,6 +239,173 @@ static void BM_StreamingPingPongMsgs(benchmark::State& state) { state.SetBytesProcessed(msg_size * state.iterations() * 2); } +// Repeatedly makes Streaming Bidi calls (exchanging a configurable number of +// messages in each call) in a loop on a single channel. Different from +// BM_StreamingPingPong we are using stream coalescing api, e.g. WriteLast, +// WriteAndFinish, set_initial_metadata_corked. These apis aim at saving +// sendmsg syscalls for streaming by coalescing 1. initial metadata with first +// message; 2. final streaming message with trailing metadata. +// +// First parmeter (i.e state.range(0)): Message size (in bytes) to use +// Second parameter (i.e state.range(1)): Number of ping pong messages. +// Note: One ping-pong means two messages (one from client to server and +// the other from server to client): +// Third parameter (i.e state.range(2)): Switch between using WriteAndFinish +// API and WriteLast API for server. +template <class Fixture, class ClientContextMutator, class ServerContextMutator> +static void BM_StreamingPingPongWithCoalescingApi(benchmark::State& state) { + const int msg_size = state.range(0); + const int max_ping_pongs = state.range(1); + // This options is used to test out server API: WriteLast and WriteAndFinish + // respectively, since we can not use both of them on server side at the same + // time. Value 1 means we are testing out the WriteAndFinish API, and + // otherwise we are testing out the WriteLast API. + const int write_and_finish = state.range(2); + + EchoTestService::AsyncService service; + std::unique_ptr<Fixture> fixture(new Fixture(&service)); + { + EchoResponse send_response; + EchoResponse recv_response; + EchoRequest send_request; + EchoRequest recv_request; + + if (msg_size > 0) { + send_request.set_message(std::string(msg_size, 'a')); + send_response.set_message(std::string(msg_size, 'b')); + } + + std::unique_ptr<EchoTestService::Stub> stub( + EchoTestService::NewStub(fixture->channel())); + + while (state.KeepRunning()) { + ServerContext svr_ctx; + ServerContextMutator svr_ctx_mut(&svr_ctx); + ServerAsyncReaderWriter<EchoResponse, EchoRequest> response_rw(&svr_ctx); + service.RequestBidiStream(&svr_ctx, &response_rw, fixture->cq(), + fixture->cq(), tag(0)); + + ClientContext cli_ctx; + ClientContextMutator cli_ctx_mut(&cli_ctx); + cli_ctx.set_initial_metadata_corked(true); + // tag:1 here will never comes up, since we are not performing any op due + // to initial metadata coalescing. + auto request_rw = stub->AsyncBidiStream(&cli_ctx, fixture->cq(), tag(1)); + + void* t; + bool ok; + int need_tags; + + // Send 'max_ping_pongs' number of ping pong messages + int ping_pong_cnt = 0; + while (ping_pong_cnt < max_ping_pongs) { + if (ping_pong_cnt == max_ping_pongs - 1) { + request_rw->WriteLast(send_request, WriteOptions(), tag(2)); + } else { + request_rw->Write(send_request, tag(2)); // Start client send + } + + need_tags = (1 << 2) | (1 << 3) | (1 << 4) | (1 << 5); + + if (ping_pong_cnt == 0) { + // wait for the server call structure (call_hook, etc.) to be + // initialized (async stream between client side and server side + // established). It is necessary when client init metadata is + // coalesced + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + while ((int)(intptr_t)t != 0) { + // In some cases tag:2 comes before tag:0 (write tag comes out + // first), this while loop is to make sure get tag:0. + int i = (int)(intptr_t)t; + GPR_ASSERT(need_tags & (1 << i)); + need_tags &= ~(1 << i); + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + } + } + + response_rw.Read(&recv_request, tag(3)); // Start server recv + request_rw->Read(&recv_response, tag(4)); // Start client recv + + while (need_tags) { + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + GPR_ASSERT(ok); + int i = (int)(intptr_t)t; + + // If server recv is complete, start the server send operation + if (i == 3) { + if (ping_pong_cnt == max_ping_pongs - 1) { + if (write_and_finish == 1) { + response_rw.WriteAndFinish(send_response, WriteOptions(), + Status::OK, tag(5)); + } else { + response_rw.WriteLast(send_response, WriteOptions(), tag(5)); + // WriteLast buffers the write, so neither server write op nor + // client read op will finish inside the while loop. + need_tags &= ~(1 << 4); + need_tags &= ~(1 << 5); + } + } else { + response_rw.Write(send_response, tag(5)); + } + } + + GPR_ASSERT(need_tags & (1 << i)); + need_tags &= ~(1 << i); + } + + ping_pong_cnt++; + } + + if (max_ping_pongs == 0) { + need_tags = (1 << 6) | (1 << 7) | (1 << 8); + } else { + if (write_and_finish == 1) { + need_tags = (1 << 8); + } else { + // server's buffered write and the client's read of the buffered write + // tags should come up. + need_tags = (1 << 4) | (1 << 5) | (1 << 7) | (1 << 8); + } + } + + // No message write or initial metadata write happened yet. + if (max_ping_pongs == 0) { + request_rw->WritesDone(tag(6)); + // wait for server call data structure(call_hook, etc.) to be + // initialized, since initial metadata is corked. + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + while ((int)(intptr_t)t != 0) { + int i = (int)(intptr_t)t; + GPR_ASSERT(need_tags & (1 << i)); + need_tags &= ~(1 << i); + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + } + response_rw.Finish(Status::OK, tag(7)); + } else { + if (write_and_finish != 1) { + response_rw.Finish(Status::OK, tag(7)); + } + } + + Status recv_status; + request_rw->Finish(&recv_status, tag(8)); + + while (need_tags) { + GPR_ASSERT(fixture->cq()->Next(&t, &ok)); + int i = (int)(intptr_t)t; + GPR_ASSERT(need_tags & (1 << i)); + need_tags &= ~(1 << i); + } + + GPR_ASSERT(recv_status.ok()); + } + } + + fixture->Finish(state); + fixture.reset(); + state.SetBytesProcessed(msg_size * state.iterations() * max_ping_pongs * 2); +} + /******************************************************************************* * CONFIGURATIONS */ @@ -270,6 +436,30 @@ BENCHMARK_TEMPLATE(BM_StreamingPingPongMsgs, InProcessCHTTP2, NoOpMutator, BENCHMARK_TEMPLATE(BM_StreamingPingPongMsgs, TCP, NoOpMutator, NoOpMutator) ->Range(0, 128 * 1024 * 1024); +// Generate Args for StreamingPingPongWithCoalescingApi benchmarks. Currently +// generates args for only "small streams" (i.e streams with 0, 1 or 2 messages) +static void StreamingPingPongWithCoalescingApiArgs( + benchmark::internal::Benchmark* b) { + int msg_size = 0; + + b->Args( + {0, 0, 0}); // spl case: 0 ping-pong msgs (msg_size doesn't matter here) + b->Args( + {0, 0, 1}); // spl case: 0 ping-pong msgs (msg_size doesn't matter here) + + for (msg_size = 0; msg_size <= 128 * 1024 * 1024; + msg_size == 0 ? msg_size++ : msg_size *= 8) { + b->Args({msg_size, 1, 0}); + b->Args({msg_size, 2, 0}); + b->Args({msg_size, 1, 1}); + b->Args({msg_size, 2, 1}); + } +} + +BENCHMARK_TEMPLATE(BM_StreamingPingPongWithCoalescingApi, InProcessCHTTP2, + NoOpMutator, NoOpMutator) + ->Apply(StreamingPingPongWithCoalescingApiArgs); + } // namespace testing } // namespace grpc diff --git a/test/cpp/microbenchmarks/bm_fullstack_streaming_pump.cc b/test/cpp/microbenchmarks/bm_fullstack_streaming_pump.cc index dc0e7d769a..5c1eb1165b 100644 --- a/test/cpp/microbenchmarks/bm_fullstack_streaming_pump.cc +++ b/test/cpp/microbenchmarks/bm_fullstack_streaming_pump.cc @@ -33,14 +33,13 @@ /* Benchmark gRPC end2end in various configurations */ +#include <benchmark/benchmark.h> #include <sstream> - #include "src/core/lib/profiling/timers.h" #include "src/cpp/client/create_channel_internal.h" #include "src/proto/grpc/testing/echo.grpc.pb.h" #include "test/cpp/microbenchmarks/fullstack_context_mutators.h" #include "test/cpp/microbenchmarks/fullstack_fixtures.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" namespace grpc { namespace testing { diff --git a/test/cpp/microbenchmarks/bm_fullstack_trickle.cc b/test/cpp/microbenchmarks/bm_fullstack_trickle.cc index 5011f06368..c563f28b55 100644 --- a/test/cpp/microbenchmarks/bm_fullstack_trickle.cc +++ b/test/cpp/microbenchmarks/bm_fullstack_trickle.cc @@ -33,12 +33,12 @@ /* Benchmark gRPC end2end in various configurations */ +#include <benchmark/benchmark.h> #include "src/core/lib/profiling/timers.h" #include "src/cpp/client/create_channel_internal.h" #include "src/proto/grpc/testing/echo.grpc.pb.h" #include "test/cpp/microbenchmarks/fullstack_context_mutators.h" #include "test/cpp/microbenchmarks/fullstack_fixtures.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" extern "C" { #include "src/core/ext/transport/chttp2/transport/chttp2_transport.h" #include "src/core/ext/transport/chttp2/transport/internal.h" diff --git a/test/cpp/microbenchmarks/bm_fullstack_unary_ping_pong.cc b/test/cpp/microbenchmarks/bm_fullstack_unary_ping_pong.cc index e51d272b10..615b05b7c7 100644 --- a/test/cpp/microbenchmarks/bm_fullstack_unary_ping_pong.cc +++ b/test/cpp/microbenchmarks/bm_fullstack_unary_ping_pong.cc @@ -33,14 +33,13 @@ /* Benchmark gRPC end2end in various configurations */ +#include <benchmark/benchmark.h> #include <sstream> - #include "src/core/lib/profiling/timers.h" #include "src/cpp/client/create_channel_internal.h" #include "src/proto/grpc/testing/echo.grpc.pb.h" #include "test/cpp/microbenchmarks/fullstack_context_mutators.h" #include "test/cpp/microbenchmarks/fullstack_fixtures.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" namespace grpc { namespace testing { diff --git a/test/cpp/microbenchmarks/bm_metadata.cc b/test/cpp/microbenchmarks/bm_metadata.cc index 34874b57f5..7029f369ad 100644 --- a/test/cpp/microbenchmarks/bm_metadata.cc +++ b/test/cpp/microbenchmarks/bm_metadata.cc @@ -33,17 +33,15 @@ /* Test out various metadata handling primitives */ +#include <benchmark/benchmark.h> #include <grpc/grpc.h> extern "C" { -#include "src/core/lib/slice/slice_internal.h" #include "src/core/lib/transport/metadata.h" #include "src/core/lib/transport/static_metadata.h" -#include "src/core/lib/transport/transport.h" } #include "test/cpp/microbenchmarks/helpers.h" -#include "third_party/benchmark/include/benchmark/benchmark.h" auto& force_library_initialization = Library::get(); @@ -65,19 +63,6 @@ static void BM_SliceFromCopied(benchmark::State& state) { } BENCHMARK(BM_SliceFromCopied); -static void BM_SliceFromStreamOwnedBuffer(benchmark::State& state) { - grpc_stream_refcount r; - GRPC_STREAM_REF_INIT(&r, 1, NULL, NULL, "test"); - char buffer[64]; - grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT; - while (state.KeepRunning()) { - grpc_slice_unref_internal(&exec_ctx, grpc_slice_from_stream_owned_buffer( - &r, buffer, sizeof(buffer))); - } - grpc_exec_ctx_finish(&exec_ctx); -} -BENCHMARK(BM_SliceFromStreamOwnedBuffer); - static void BM_SliceIntern(benchmark::State& state) { TrackCounters track_counters; gpr_slice slice = grpc_slice_from_static_string("abc"); diff --git a/test/cpp/microbenchmarks/helpers.h b/test/cpp/microbenchmarks/helpers.h index f44b7cf83a..49ed517b1d 100644 --- a/test/cpp/microbenchmarks/helpers.h +++ b/test/cpp/microbenchmarks/helpers.h @@ -41,8 +41,8 @@ extern "C" { #include "test/core/util/memory_counters.h" } +#include <benchmark/benchmark.h> #include <grpc++/impl/grpc_library.h> -#include "third_party/benchmark/include/benchmark/benchmark.h" class Library { public: diff --git a/test/cpp/microbenchmarks/noop-benchmark.cc b/test/cpp/microbenchmarks/noop-benchmark.cc index 99fa6d5f6e..7372ad04f2 100644 --- a/test/cpp/microbenchmarks/noop-benchmark.cc +++ b/test/cpp/microbenchmarks/noop-benchmark.cc @@ -34,7 +34,7 @@ /* This benchmark exists to ensure that the benchmark integration is * working */ -#include "third_party/benchmark/include/benchmark/benchmark.h" +#include <benchmark/benchmark.h> static void BM_NoOp(benchmark::State& state) { while (state.KeepRunning()) { |