Copilot commented on code in PR #3679: URL: https://github.com/apache/brpc/pull/3679#discussion_r4550180111
########## test/fuzzing/fuzz_concurrency.cpp: ########## @@ -0,0 +1,95 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to You under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and +// limitations under the License. + +// Fuzz the auto concurrency limiter (src/brpc/policy/auto_concurrency_limiter.cpp): +// fuzz bytes drive a stream of (error_code, latency_us) responses and +// request ticks through the real gradient-based limiter, including the +// recovery/reset paths exercised by bursts of errors. +// +// OnResponded() stamps samples with the real clock: filling a sample window +// needs >=100us of wall time per sample and remeasurement cannot start +// until ~25s after construction, so coverage would depend on execution +// speed rather than on the input bytes. Samples are therefore fed to +// AddSample() at a synthetic clock that every 8-byte record advances -- +// the same sampling approach as test/brpc_auto_concurrency_limiter_unittest.cpp +// -- which lets an input jump straight across the sample window and the +// remeasure/reset deadlines. +// +// Each 8-byte record is big-endian and decoded as: +// bits 0..7 error_code +// bits 8..15 request-tick flag (bit 8 / 0x100) +// bits 8..31 latency_us (24 bits) +// bits 20..31 also feed OnRequested concurrency (mod kMaxConcurrency) +// bits 40..63 synthetic clock delta (plus 1us floor) + +#include <cstddef> +#include <cstdint> + +#include "butil/time.h" +#include "brpc/policy/auto_concurrency_limiter.h" + +namespace { + +constexpr size_t kRecordBytes = 8; +// Cap for OnRequested()'s current_concurrency argument: wide enough to +// cross typical limiter ceilings without running into huge ints. +constexpr int kMaxConcurrency = 4096; +constexpr int kErrorCodeBits = 8; +constexpr int kLatencyShift = 8; +constexpr uint64_t kLatencyMask = 0xffffffULL; +constexpr int kRequestFlagBit = 8; // bit 8 of the record +constexpr int kConcurrencyShift = 20; +constexpr int kClockDeltaShift = 40; + +} // namespace + +extern "C" int +LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) +{ + if (size < kRecordBytes) { + return 0; + } + + brpc::policy::AutoConcurrencyLimiter limiter; + int64_t now_us = butil::cpuwide_time_us(); Review Comment: Using `butil::cpuwide_time_us()` introduces non-determinism into the fuzzer execution (seed corpus minimization and reproducing crashes can become harder). Since this harness already uses a synthetic clock (`now_us` advanced from input records), consider initializing `now_us` to a fixed value (e.g., `0`) so behavior depends only on input bytes. ########## test/fuzzing/fuzz_concurrency.cpp: ########## @@ -0,0 +1,95 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to You under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and +// limitations under the License. + +// Fuzz the auto concurrency limiter (src/brpc/policy/auto_concurrency_limiter.cpp): +// fuzz bytes drive a stream of (error_code, latency_us) responses and +// request ticks through the real gradient-based limiter, including the +// recovery/reset paths exercised by bursts of errors. +// +// OnResponded() stamps samples with the real clock: filling a sample window +// needs >=100us of wall time per sample and remeasurement cannot start +// until ~25s after construction, so coverage would depend on execution +// speed rather than on the input bytes. Samples are therefore fed to +// AddSample() at a synthetic clock that every 8-byte record advances -- +// the same sampling approach as test/brpc_auto_concurrency_limiter_unittest.cpp +// -- which lets an input jump straight across the sample window and the +// remeasure/reset deadlines. +// +// Each 8-byte record is big-endian and decoded as: +// bits 0..7 error_code +// bits 8..15 request-tick flag (bit 8 / 0x100) Review Comment: The bitfield documentation is internally inconsistent: the request-tick flag is described as `bits 8..15` but the implementation treats it as a single bit (`bit 8`), and those same bits are also part of `latency_us` (`bits 8..31`). Please rewrite this layout to reflect the actual overlapping usage (e.g., explicitly call out that `bit 8` is both the request flag and the LSB of `latency_us`, or separate the fields to avoid overlap). ########## test/fuzzing/fuzz_concurrency.cpp: ########## @@ -0,0 +1,95 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to You under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and +// limitations under the License. + +// Fuzz the auto concurrency limiter (src/brpc/policy/auto_concurrency_limiter.cpp): +// fuzz bytes drive a stream of (error_code, latency_us) responses and +// request ticks through the real gradient-based limiter, including the +// recovery/reset paths exercised by bursts of errors. +// +// OnResponded() stamps samples with the real clock: filling a sample window +// needs >=100us of wall time per sample and remeasurement cannot start +// until ~25s after construction, so coverage would depend on execution +// speed rather than on the input bytes. Samples are therefore fed to +// AddSample() at a synthetic clock that every 8-byte record advances -- +// the same sampling approach as test/brpc_auto_concurrency_limiter_unittest.cpp +// -- which lets an input jump straight across the sample window and the +// remeasure/reset deadlines. +// +// Each 8-byte record is big-endian and decoded as: +// bits 0..7 error_code +// bits 8..15 request-tick flag (bit 8 / 0x100) +// bits 8..31 latency_us (24 bits) +// bits 20..31 also feed OnRequested concurrency (mod kMaxConcurrency) +// bits 40..63 synthetic clock delta (plus 1us floor) + +#include <cstddef> +#include <cstdint> + +#include "butil/time.h" +#include "brpc/policy/auto_concurrency_limiter.h" + +namespace { + +constexpr size_t kRecordBytes = 8; +// Cap for OnRequested()'s current_concurrency argument: wide enough to +// cross typical limiter ceilings without running into huge ints. +constexpr int kMaxConcurrency = 4096; +constexpr int kErrorCodeBits = 8; +constexpr int kLatencyShift = 8; +constexpr uint64_t kLatencyMask = 0xffffffULL; +constexpr int kRequestFlagBit = 8; // bit 8 of the record +constexpr int kConcurrencyShift = 20; +constexpr int kClockDeltaShift = 40; + +} // namespace + +extern "C" int +LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) +{ + if (size < kRecordBytes) { + return 0; + } + + brpc::policy::AutoConcurrencyLimiter limiter; + int64_t now_us = butil::cpuwide_time_us(); + + const size_t n = size / kRecordBytes; + for (size_t i = 0; i < n; ++i) { + uint64_t v = 0; + for (size_t b = 0; b < kRecordBytes; ++b) { + v = (v << 8) | data[i * kRecordBytes + b]; + } + // Advance the synthetic clock by 1us..~16.8s per record. The +1us + // floor keeps the sampled window width strictly positive -- + // UpdateMaxConcurrency() divides by it when computing qps. + now_us += static_cast<int64_t>((v >> kClockDeltaShift) + 1); + const int error_code = + static_cast<int>(v & ((1ULL << kErrorCodeBits) - 1)); + const int64_t latency_us = + static_cast<int64_t>((v >> kLatencyShift) & kLatencyMask); + limiter.AddSample(error_code, latency_us, now_us); + limiter.MaxConcurrency(); + if ((v & (1ULL << kRequestFlagBit)) || (i & 7) == 0) { Review Comment: The `(i & 7) == 0` periodic tick is a magic constant that’s not explained. Please extract `7`/`8` into a named constant (e.g., `kRequestTickPeriod = 8`) and document why a periodic `OnRequested()` call is needed even when the flag isn’t set. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
