多进程之间需要传输大量数据的时候,比如多进程 RPC 框架的进程之间通信,常用共享内存队列。
但是共享内存队列难免会有 入队+出队 2次 memcpy 。
而且要变长共享内存队列,如果支持多生产者进程+多消费者进程 ,就要处理线程安全方面的问题, 比较麻烦。
process_vm_readv() , process_vm_writev() 是 Linux 3.2 新增的 syscall,用于在多个进程的地址空间之间,高效传输大块数据。
https://www.man7.org/linux/man-pages/man2/process_vm_readv.2.html
https://github.com/open-mpi/ompi/blob/master/opal/mca/btl/sm/btl_sm_get.c#L96
在此, 我提个设想,可以用 process_vm_readv 实现一个多进程内存队列,相比之下,优势是:
下面 demo 代码演示了这个思路。 为了方便,直接用 posix mq 来传输 iovec 数组,也没有考虑多路复用。 真实项目可能需要 eventfd 之内的机制来做 notify。
实测了下,3个进程加起来有 5.5 GB/秒 的速度。
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 | #include <assert.h> #include <fcntl.h> #include <mqueue.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/mman.h> #include <sys/stat.h> #include <sys/types.h> #include <sys/uio.h> #include <sys/wait.h> #include <unistd.h> #include <iostream> #include <sstream> #include <vector> using namespace std; struct MemMessage { pid_t pid = 0; vector<iovec> mem_vec; public: MemMessage() {} string DebugString() const { ostringstream oss; oss << "pid=" << pid; for (const auto& mem : mem_vec) { oss << " iov={" << mem.iov_base << "," << mem.iov_len << "}"; } return oss.str(); } string ToString() const { pid_t from_pid = getpid(); string buff; buff.append((const char*)&from_pid, sizeof(from_pid)); for (const auto& mem : mem_vec) { buff.append((const char*)&mem, sizeof(mem)); } return buff; } bool FromString(const char* buff, size_t buff_len) { size_t pos = 0; if (buff_len < sizeof(pid)) { return false; } pid = *((pid_t*)buff); pos += sizeof(pid); for (; pos + sizeof(iovec) <= buff_len; pos += sizeof(iovec)) { mem_vec.resize(mem_vec.size() + 1); auto& iov = mem_vec.back(); memcpy(reinterpret_cast<char*>(&iov), buff + pos, sizeof(iov)); } return true; } void CopyToMem(string& buff) { size_t sum = 0; for (const auto& iov : mem_vec) { sum += iov.iov_len; } buff.resize(sum, '\0'); iovec local; local.iov_base = &buff[0]; local.iov_len = buff.size(); auto nread = process_vm_readv(pid, &local, 1, &mem_vec[0], mem_vec.size(), 0); // assert(nread >= 0); if (nread <= 0) { cout << getpid() << " process_vm_readv from=" << DebugString() << " sum=" << sum << " len=" << nread << endl; cout << "pid=" << pid << " " << strerror(errno) << endl; } } }; class MemQueue { mqd_t m_mq; struct mq_attr m_attr; public: int Init(const string& queue_name, bool read_write = false) { memset(&m_attr, 0, sizeof(m_attr)); m_attr.mq_flags = 0; m_attr.mq_maxmsg = 10; m_attr.mq_msgsize = 4096; const int oflag = read_write ? O_WRONLY | O_CREAT : O_RDONLY; m_mq = mq_open(queue_name.c_str(), oflag, S_IRWXU, &m_attr); // m_mq = mq_open(queue_name.c_str(), oflag); cout << getpid() << " mq_open=" << queue_name << " oflag=" << oflag << " ret=" << m_mq << endl; } int PushMemMsg(const MemMessage& msg) { string buff = msg.ToString(); return mq_send(m_mq, buff.c_str(), buff.size(), 0); } bool PopMemMsg(MemMessage& msg) { string buff(m_attr.mq_msgsize, '\0'); unsigned prio = 0; const auto ret = mq_receive(m_mq, &buff[0], buff.size(), &prio); if (ret < 0) { return false; } buff.resize(ret); msg.FromString(buff.data(), buff.size()); return true; } }; const static string queue_name = "/test_queue"; enum STATUS { STATUS_WAITING = 0, STATUS_RUNNING = 1, STATUS_EXIT = 2 }; struct ControlCenter { int proc_status; }; ControlCenter* control = nullptr; void child_proc() { const auto ret = fork(); assert(ret >= 0); if (ret != 0) { return; } while (STATUS_WAITING == control->proc_status) { usleep(1000); } MemQueue q; q.Init(queue_name); size_t bytes = 0; while (STATUS_RUNNING == control->proc_status) { MemMessage msg; if (q.PopMemMsg(msg)) { string buff; msg.CopyToMem(buff); bytes += buff.size(); } else { usleep(1000); } } cout << getpid() << " bytes=" << bytes << " exit" << endl; exit(0); } int main(int argc, char* argv[]) { control = (ControlCenter*)mmap(NULL, sizeof(*control), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); control->proc_status = STATUS_WAITING; const static int proc_num = 3; for (int i = 0; i < proc_num; ++i) { child_proc(); } MemQueue q; q.Init(queue_name, true); control->proc_status = STATUS_RUNNING; string str(1024 * 1024 * 10, 'a'); for (int i = 0; i < 10000; ++i) { MemMessage msg; for (int t = 0; t < 3; ++t) { size_t base = rand() % str.size(); size_t size = rand() % (str.size() - base); msg.mem_vec.emplace_back(iovec{&str[base], size}); } const auto ret = q.PushMemMsg(msg); // cout << getpid() << " PushMemMsg=" << msg.DebugString() << " ret=" << ret << endl; // sleep(1); } control->proc_status = STATUS_EXIT; for (int i = 0; i < proc_num; ++i) { int status = 0; wait(&status); } return 0; } |
---|