Add pthread CPU affinity support

[wenjiecn]

What problem does this PR solve?

Issue Number:#1140

Problem Summary:

What is changed and the side effects?

Changed:
添加gflag
cpu_set：用户需要绑定的cpu集合，默认为空串，表示不开启绑核
当开启绑核时，会解析cpu集合到_cpus数组中
每次调用pthread_create(&_workers[i], NULL, worker_thread, arg);后，会绑定worker_id对应的_cpus[worker_id % _cpus.size()]

Side effects:

• Performance effects:

• Breaking backward compatibility:

---

Check List:

• Please make sure your changes are compilable.
• When providing us with a new feature, it is best to add related tests.
• Please follow Contributor Covenant Code of Conduct.

[wwbmmm]

Maybe you can combine these two flags into one flag. If cpu_set is an empty string, then the thread affinity feature is off.

[wenjiecn]

done

[Copilot]

Pull Request Overview

This PR adds pthread CPU affinity support to enable binding worker threads to specific CPU cores for better performance control. The implementation allows users to control thread-to-CPU binding through configuration flags.

Key Changes:

• Added two gflags: thread_affinity to enable/disable CPU binding and cpu_set to specify target CPUs
• Implemented CPU set parsing with support for range notation (e.g., "0-3,5,6-7")
• Modified worker thread creation to bind threads to CPUs in round-robin fashion when affinity is enabled

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

File	Description
src/bthread/task_control.h	Added CPU affinity helper functions and static _cpus vector to store parsed CPU set
src/bthread/task_control.cpp	Implemented CPU set parsing, thread binding logic, and integrated affinity support into worker thread initialization

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[yanglimingcn]

这个过程能放到startfn这些函数里面吗？

[wenjiecn]

这个过程能放到startfn这些函数里面吗？

@yanglimingcn
startfn方法是给用户自定义的Hook方法
run_tagged_worker_startfn(tag)方法，用户可以自定义将tag[i]下的线程组绑定到自定义的cpus[tag[i]]；
我这里实现一些轮子

[yanglimingcn]

动态增加线程会怎么设置呢？

[wenjiecn]

动态增加线程会怎么设置呢？

@yanglimingcn
动态增加worker_thread的方法是TaskControl::add_workers，我添加了

if (!_cpus.empty()) {
    arg->cpuId = (i + old_concurency) % _cpus.size();
}

无论如何增加，它都是存在_workers里的，所以只要保证_workers[i]对应_cpus[i % _cpus.size()]

[yanglimingcn]

收到

[yanglimingcn]

为什么把这个函数定义为inline的？

[wenjiecn]

因为函数体逻辑简单

[yanglimingcn]

有没有测试过绑核和不绑核的数据？

[wenjiecn]

有没有测试过绑核和不绑核的数据？

@yanglimingcn

不绑核 connection_type=single

./echo_server -minloglevel=4 --ip_port=*.*.*.*:*
./rpc_press -proto="./echo.proto" -method=example.EchoService.Echo -server=*.*.*.*:* -input=.'{"message":"hello"} {"message":"world"}' -qps=0 -timeout_ms=3000 -thread_num=200

[Latency]
avg 531 us
50% 521 us
70% 597 us
90% 757 us
95% 817 us
97% 846 us
99% 907 us
99.9% 1132 us
99.99% 1360 us
max 2250 us

[Latency]
avg 585 us
50% 569 us
70% 649 us
90% 797 us
95% 843 us
97% 877 us
99% 982 us
99.9% 1142 us
99.99% 1291 us
max 2706 us

[Latency]
avg 606 us
50% 590 us
70% 680 us
90% 814 us
95% 880 us
97% 926 us
99% 1033 us
99.9% 1184 us
99.99% 1935 us
max 3452 us

绑核 connection_type=single

numactl --cpunodebind=0 --membind=0 ./echo_server -minloglevel=4 --ip_port=*.*.*.*:* --cpu_set=0-79
numactl --cpunodebind=0 --membind=0 ./rpc_press -proto="./echo.proto" -method=example.EchoService.Echo -server=*.*.*.*:* -input=.'{"message":"hello"} {"message":"world"}' -qps=0 -timeout_ms=3000 -thread_num=200 --cpu_set=0-10
[Latency]
avg 511 us
50% 505 us
70% 576 us
90% 697 us
95% 795 us
97% 843 us
99% 891 us
99.9% 1146 us
99.99% 2124 us
max 7936 us
[Latency]
avg 522 us
50% 516 us
70% 585 us
90% 725 us
95% 807 us
97% 830 us
99% 899 us
99.9% 1145 us
99.99% 1726 us
max 4164 us
[Latency]
avg 516 us
50% 504 us
70% 576 us
90% 703 us
95% 792 us
97% 823 us
99% 871 us
99.9% 1255 us
99.99% 2252 us
max 8437 us

不绑核 connection_type=pooled

./echo_server -minloglevel=4 --ip_port=*.*.*.*:*
./rpc_press -proto="./echo.proto" -method=example.EchoService.Echo -server=*.*.*.*:* -input=.'{"message":"hello"} {"message":"world"}' -qps=0 -timeout_ms=3000 -thread_num=200
[Latency]
avg 1151 us
50% 1135 us
70% 1212 us
90% 1333 us
95% 1403 us
97% 1444 us
99% 1545 us
99.9% 1636 us
99.99% 4142 us
max 5198 us

[Latency]
avg 1322 us
50% 1281 us
70% 1373 us
90% 1534 us
95% 1619 us
97% 1696 us
99% 1767 us
99.9% 1892 us
99.99% 2074 us
max 6483 us

[Latency]
avg 1494 us
50% 1514 us
70% 1590 us
90% 1718 us
95% 1783 us
97% 1817 us
99% 1945 us
99.9% 2074 us
99.99% 3520 us
max 4119 us

绑核 connection_type=pooled

numactl --cpunodebind=0 --membind=0 ./echo_server -minloglevel=4 --ip_port=*.*.*.*:* --cpu_set=0-79
numactl --cpunodebind=0 --membind=0 ./rpc_press -proto="./echo.proto" -method=example.EchoService.Echo -server=*.*.*.*:* -input=.'{"message":"hello"} {"message":"world"}' -qps=0 -timeout_ms=3000 -thread_num=200 --cpu_set=0-10
[Latency]
avg 810 us
50% 789 us
70% 847 us
90% 956 us
95% 1007 us
97% 1048 us
99% 1119 us
99.9% 1209 us
99.99% 1286 us
max 2332 us

[Latency]
avg 809 us
50% 790 us
70% 849 us
90% 966 us
95% 1020 us
97% 1054 us
99% 1120 us
99.9% 1217 us
99.99% 1249 us
max 1425 us

[Latency]
avg 809 us
50% 788 us
70% 851 us
90% 971 us
95% 1011 us
97% 1054 us
99% 1121 us
99.9% 1230 us
99.99% 1260 us
max 1784 us

[yanglimingcn]

在single模式下收益没有pooled下面明显。

[yanglimingcn]

修复一下冲突吧

[yanglimingcn]

LGTM

[wasphin]

LGTM