bark
A modern, low latency datadog client for C++
MIT License
Stars
2
Ecosystems:
C++
bark
A modern, low latency datadog stats library for c++
Getting Started
#include "bark/bark.hpp"
auto main() -> int
{
using namespace bark;
// make a none-async client which sends to the local DogStatsD agent
auto client = Client::make_local_client();
// Sending a gauge datagram
client.send(Gauge("gauge.name", 42.0));
// Sending a histogram datagram
client.send(Histogram("histogram.name", 99).with(Tags::from_list({"tag1:hello", "tag2:world"})));
// Sending a count datagram
client.send(Count("count.name", 22).with(SampleRate {0.5}));
// Sending an event datagram
client.send(Event("event.title", "event.value").with(Priority::Normal).with(AlertType::Info));
return 0;
}
We provide a number of Client implementations for different use cases:
| Client | Queue type | Number of active threads | Notes |
|---|---|---|---|
Client |
None | The publishing thread only | Sends metrics on the publishing thread |
SPSCClient |
Single Producer Single Consumer | The publishing thread + 1 | Fixed queue size. Drops new messages when full |
MPMCClient |
Multiple Producer Multiple Consumer | The publishing thread + 1 | Fixed queue size. Drops new messages when full |
AsioClient |
asio::io_context |
The publishing thread + N
|
N is supplied when creating the client |
NoOpClient |
None | (The publishing thread)/None | Useful for testing |
All of these clients implement the IDatadogClient interface. You are not required to use it, but it is there to allow mocking and ease of use for those that need it.
Requirements
We use modern C++ features and as such you are required to use a newish compiler. We test the library in CI with GCC-13 and Clang-18. We are open for PRs adding Mac and Windows CI checks as well.
Benchmarks
We benchmark both the metric creation and serialization time as well as the latency of the main thread to send off a datagram.
See benchmarks/source for the benchmark code.
Clang-18 Results
2024-09-20T12:42:44+00:00
Running ./build/benchmarks/bark_benchmarks
Run on (4 X 3241.38 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x2)
L1 Instruction 32 KiB (x2)
L2 Unified 512 KiB (x2)
L3 Unified 32768 KiB (x1)
Load Average: 3.26, 2.09, 0.92
----------------------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
----------------------------------------------------------------------------------------------------------------------
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_mean 7735 ns 7734 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_median 7719 ns 7719 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_stddev 44.1 ns 44.0 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_cv 0.57 % 0.57 % 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_mean 297 ns 296 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_median 278 ns 278 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_stddev 57.4 ns 57.5 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_cv 19.31 % 19.41 % 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_mean 76.6 ns 76.6 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_median 73.7 ns 73.7 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_stddev 12.0 ns 11.9 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_cv 15.60 % 15.60 % 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_mean 74.6 ns 74.5 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_median 73.1 ns 73.1 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_stddev 3.70 ns 3.71 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_cv 4.95 % 4.97 % 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_mean 7758 ns 7758 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_median 7755 ns 7755 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_stddev 15.3 ns 15.3 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_cv 0.20 % 0.20 % 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_mean 146 ns 146 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_median 143 ns 142 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_stddev 18.6 ns 18.6 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_cv 12.73 % 12.77 % 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_mean 87.3 ns 87.3 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_median 86.4 ns 86.4 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_stddev 5.66 ns 5.61 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_cv 6.48 % 6.43 % 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_mean 96.7 ns 96.6 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_median 93.1 ns 93.1 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_stddev 12.9 ns 12.8 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_cv 13.38 % 13.27 % 32
benchmark_event_serialize/repeats:16_mean 228 ns 228 ns 16
benchmark_event_serialize/repeats:16_median 228 ns 228 ns 16
benchmark_event_serialize/repeats:16_stddev 0.304 ns 0.304 ns 16
benchmark_event_serialize/repeats:16_cv 0.13 % 0.13 % 16
benchmark_metric_serialize<Count>/repeats:16_mean 222 ns 222 ns 16
benchmark_metric_serialize<Count>/repeats:16_median 222 ns 222 ns 16
benchmark_metric_serialize<Count>/repeats:16_stddev 0.403 ns 0.405 ns 16
benchmark_metric_serialize<Count>/repeats:16_cv 0.18 % 0.18 % 16
benchmark_metric_serialize<Gauge>/repeats:16_mean 197 ns 197 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_median 196 ns 196 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_stddev 0.695 ns 0.694 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_cv 0.35 % 0.35 % 16
benchmark_metric_serialize<Histogram>/repeats:16_mean 290 ns 290 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_median 290 ns 290 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_stddev 0.601 ns 0.600 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_cv 0.21 % 0.21 % 16
benchmark_udp_client_send_metric 7342 ns 7342 ns 94870
GCC-13 Results
2024-09-20T12:43:14+00:00
Running ./build/benchmarks/bark_benchmarks
Run on (4 X 3244.6 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x2)
L1 Instruction 32 KiB (x2)
L2 Unified 512 KiB (x2)
L3 Unified 32768 KiB (x1)
Load Average: 1.85, 0.63, 0.24
----------------------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
----------------------------------------------------------------------------------------------------------------------
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_mean 7752 ns 7752 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_median 7722 ns 7722 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_stddev 102 ns 102 ns 32
benchmark_client_send_metric<Client>/iterations:25000/repeats:32_cv 1.31 % 1.31 % 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_mean 255 ns 255 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_median 254 ns 254 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_stddev 46.8 ns 46.8 ns 32
benchmark_client_send_event<AsioClient>/iterations:25000/repeats:32_cv 18.39 % 18.40 % 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_mean 70.9 ns 70.9 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_median 69.0 ns 68.9 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_stddev 9.84 ns 9.84 ns 32
benchmark_client_send_metric<SPSCClient>/iterations:25000/repeats:32_cv 13.87 % 13.88 % 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_mean 66.8 ns 66.8 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_median 65.7 ns 65.6 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_stddev 3.46 ns 3.44 ns 32
benchmark_client_send_metric<MPMCClient>/iterations:25000/repeats:32_cv 5.17 % 5.14 % 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_mean 7690 ns 7689 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_median 7680 ns 7679 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_stddev 32.9 ns 33.0 ns 32
benchmark_client_send_event<Client>/iterations:25000/repeats:32_cv 0.43 % 0.43 % 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_mean 131 ns 131 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_median 128 ns 128 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_stddev 11.7 ns 11.7 ns 32
benchmark_client_send_metric<AsioClient>/iterations:25000/repeats:32_cv 8.98 % 8.99 % 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_mean 78.2 ns 78.1 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_median 77.3 ns 77.3 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_stddev 5.55 ns 5.55 ns 32
benchmark_client_send_event<SPSCClient>/iterations:25000/repeats:32_cv 7.10 % 7.10 % 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_mean 77.3 ns 77.2 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_median 76.1 ns 76.0 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_stddev 3.88 ns 3.89 ns 32
benchmark_client_send_event<MPMCClient>/iterations:25000/repeats:32_cv 5.02 % 5.04 % 32
benchmark_event_serialize/repeats:16_mean 231 ns 231 ns 16
benchmark_event_serialize/repeats:16_median 231 ns 231 ns 16
benchmark_event_serialize/repeats:16_stddev 1.16 ns 1.16 ns 16
benchmark_event_serialize/repeats:16_cv 0.50 % 0.50 % 16
benchmark_metric_serialize<Count>/repeats:16_mean 241 ns 241 ns 16
benchmark_metric_serialize<Count>/repeats:16_median 240 ns 240 ns 16
benchmark_metric_serialize<Count>/repeats:16_stddev 0.901 ns 0.903 ns 16
benchmark_metric_serialize<Count>/repeats:16_cv 0.37 % 0.38 % 16
benchmark_metric_serialize<Gauge>/repeats:16_mean 213 ns 213 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_median 213 ns 213 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_stddev 1.18 ns 1.18 ns 16
benchmark_metric_serialize<Gauge>/repeats:16_cv 0.55 % 0.55 % 16
benchmark_metric_serialize<Histogram>/repeats:16_mean 313 ns 313 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_median 313 ns 313 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_stddev 0.559 ns 0.560 ns 16
benchmark_metric_serialize<Histogram>/repeats:16_cv 0.18 % 0.18 % 16
benchmark_udp_client_send_metric 7332 ns 7332 ns 95439
Building and installing
Create a copy of CMakeUserPresets.template.json called CMakeUserPreset.json. Fix the VCPKG_ROOT environment variable in the new preset file, to point to somewhere on your system where you have vcpkg installed. Now you can run with presets dev etc.
You might want to remove the clang-tidy preset inherit from dev-common while developing to greatly speed up compilation time.