From cf05ea3d9f43e4e84d621e1f9d54cbef552b6e2b Mon Sep 17 00:00:00 2001 From: lichao <lichao@aiotlink.com> Date: 星期二, 18 五月 2021 16:53:28 +0800 Subject: [PATCH] fix center init mutex. --- src/topic_node.cpp | 167 ++++++++++++++++++++++++++++--------------------------- 1 files changed, 86 insertions(+), 81 deletions(-) diff --git a/src/topic_node.cpp b/src/topic_node.cpp index 35228b4..43d748f 100644 --- a/src/topic_node.cpp +++ b/src/topic_node.cpp @@ -28,7 +28,12 @@ namespace { -inline void AddRoute(BHMsgHead &head, const MQId id) { head.add_route()->set_mq_id(id); } +inline void AddRoute(BHMsgHead &head, const ShmSocket &sock) +{ + auto route = head.add_route(); + route->set_mq_id(sock.id()); + route->set_abs_addr(sock.AbsAddr()); +} struct SrcInfo { std::vector<BHAddress> route; @@ -40,7 +45,7 @@ } // namespace TopicNode::TopicNode(SharedMemory &shm) : - shm_(shm), state_(eStateUnregistered) + shm_(shm), state_(eStateUninited) { } @@ -65,69 +70,50 @@ } LOG_DEBUG() << "Node Init, id " << ssn_id_; auto NodeInit = [&]() { - auto SendInitCmd = [&]() { - int64_t init_cmd = ssn_id_ << 4 | EncodeCmd(eCmdNodeInit); - auto end_time = steady_clock::now() + 3s; - bool r = false; - do { - r = ShmMsgQueue::TrySend(shm(), BHTopicCenterAddress(), init_cmd); - } while (!r && steady_clock::now() < end_time); - return r; - }; - if (SendInitCmd()) { - LOG_DEBUG() << "node send init ok"; - auto end_time = steady_clock::now() + 3s; - do { - try { - for (int i = eSockStart; i < eSockEnd; ++i) { - sockets_.emplace_back(new ShmSocket(shm_, false, ssn_id_ + i, kMqLen)); - } - break; - } catch (...) { - sockets_.clear(); - std::this_thread::sleep_for(100ms); - } - } while (steady_clock::now() < end_time); + int64_t init_request = ssn_id_ << 4 | EncodeCmd(eCmdNodeInit); + int64_t reply = 0; + if (BHNodeInit(init_request, reply) && DecodeCmd(reply) == eCmdNodeInitReply) { + int64_t abs_addr = reply >> 4; + sockets_.emplace_back(new ShmSocket(abs_addr, shm_, ssn_id_)); + LOG_DEBUG() << "node init ok"; + } else { + LOG_ERROR() << "Node Init Error"; } }; if (sockets_.empty()) { NodeInit(); } if (!sockets_.empty()) { - LOG_DEBUG() << "node sockets ok"; - auto onNodeCmd = [this](ShmSocket &socket, int64_t &val) { - LOG_DEBUG() << "node recv cmd: " << DecodeCmd(val); - switch (DecodeCmd(val)) { - case eCmdNodeInitReply: { - MsgI msg(val >> 4); - DEFER1(msg.Release()); + auto onMsg = [this](ShmSocket &socket, MsgI &imsg, BHMsgHead &head) { + LOG_DEBUG() << "node recv type: " << head.type(); + switch (head.type()) { + case kMsgTypeProcInit: { + // reuse msg to send proc init. MsgProcInit body; + body.set_extra_mq_num(eSockEnd - eSockStart - 1); auto head = InitMsgHead(GetType(body), info_.proc_id(), ssn_id_); - head.add_route()->set_mq_id(ssn_id_); - if (msg.Fill(head, body)) { - socket.Send(BHTopicCenterAddress(), msg); + AddRoute(head, socket); + if (imsg.Fill(head, body)) { + socket.Send(BHTopicCenterAddress(), imsg); } } break; - default: - break; - } - return true; - }; - - // recv msgs to avoid memory leak. - auto onMsg = [this](ShmSocket &sock, MsgI &imsg, BHMsgHead &head) { - LOG_DEBUG() << "node recv type: " << head.type(); - if (head.type() == kMsgTypeProcInitReply) { + case kMsgTypeProcInitReply: { LOG_DEBUG() << "got proc init reply"; MsgProcInitReply reply; - if (imsg.ParseBody(reply)) { + if (imsg.ParseBody(reply) && IsSuccess(reply.errmsg().errcode())) { + for (auto &addr : reply.extra_mqs()) { + LOG_DEBUG() << "add socket " << addr.abs_addr() << ", id:" << addr.mq_id(); + sockets_.emplace_back(new ShmSocket(addr.abs_addr(), shm(), addr.mq_id())); + } SetProcIndex(reply.proc_index()); + this->state_ = eStateUnregistered; } + } break; + default: break; } return true; }; - SockNode().Start(1, onMsg, onNodeCmd); - LOG_DEBUG() << "sockets ok."; + SockNode().Start(1, onMsg); return true; } return false; @@ -167,19 +153,22 @@ SetLastError(eError, kErrMsgNotInit); return false; } + auto end_time = steady_clock::now() + milliseconds(timeout_ms); + + while (!Valid() && steady_clock::now() < end_time) { + std::this_thread::yield(); + } + if (!Valid()) { + SetLastError(eError, kErrMsgNotInit); + return false; + } auto &sock = SockNode(); MsgRegister body; body.mutable_proc()->Swap(&proc); - auto AddId = [&](const MQId id) { body.add_addrs()->set_mq_id(id); }; - AddId(SockNode().id()); - AddId(SockServer().id()); - AddId(SockClient().id()); - AddId(SockSub().id()); - AddId(SockPub().id()); auto head(InitMsgHead(GetType(body), body.proc().proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); auto CheckResult = [this](MsgI &msg, BHMsgHead &head, MsgCommonReply &rbody) { bool ok = head.type() == kMsgTypeCommonReply && @@ -224,7 +213,7 @@ body.mutable_proc()->Swap(&proc); auto head(InitMsgHead(GetType(body), body.proc().proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); auto CheckResult = [this](MsgI &msg, BHMsgHead &head, MsgCommonReply &rbody) { bool r = head.type() == kMsgTypeCommonReply && @@ -260,7 +249,7 @@ body.mutable_proc()->Swap(&proc); auto head(InitMsgHead(GetType(body), body.proc().proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); if (timeout_ms == 0) { return sock.Send(BHTopicCenterAddress(), head, body); @@ -290,7 +279,7 @@ auto &sock = SockNode(); BHMsgHead head(InitMsgHead(GetType(query), proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); MsgI reply; DEFER1(reply.Release()); @@ -312,7 +301,7 @@ body.mutable_topics()->Swap(&topics); auto head(InitMsgHead(GetType(body), proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); if (timeout_ms == 0) { return sock.Send(BHTopicCenterAddress(), head, body); @@ -341,7 +330,7 @@ for (int i = 0; i < head.route_size() - 1; ++i) { reply_head.add_route()->Swap(head.mutable_route(i)); } - auto remote = head.route().rbegin()->mq_id(); + MQInfo remote = {head.route().rbegin()->mq_id(), head.route().rbegin()->abs_addr()}; sock.Send(remote, reply_head, reply_body); } }; @@ -357,10 +346,17 @@ MsgRequestTopic req; if (!imsg.ParseBody(req)) { return; } - SrcInfo *p = new SrcInfo; - p->route.assign(head.route().begin(), head.route().end()); - p->msg_id = head.msg_id(); - acb(p, *head.mutable_proc_id(), req); + try { + SrcInfo *p = new SrcInfo; + if (!p) { + throw std::runtime_error("no memory."); + } + p->route.assign(head.route().begin(), head.route().end()); + p->msg_id = head.msg_id(); + acb(p, *head.mutable_proc_id(), req); + } catch (std::exception &e) { + LOG_ERROR() << "error server handle msg:" << e.what(); + } }; auto &sock = SockServer(); @@ -381,11 +377,19 @@ if (sock.SyncRecv(imsg, head, timeout_ms) && head.type() == kMsgTypeRequestTopic) { if (imsg.ParseBody(request)) { head.mutable_proc_id()->swap(proc_id); - SrcInfo *p = new SrcInfo; - p->route.assign(head.route().begin(), head.route().end()); - p->msg_id = head.msg_id(); - src_info = p; - return true; + try { + SrcInfo *p = new SrcInfo; + if (!p) { + throw std::runtime_error("no memory."); + } + p->route.assign(head.route().begin(), head.route().end()); + p->msg_id = head.msg_id(); + src_info = p; + return true; + } catch (std::exception &e) { + LOG_ERROR() << "error recv request: " << e.what(); + return false; + } } } return false; @@ -409,7 +413,8 @@ for (unsigned i = 0; i < p->route.size() - 1; ++i) { head.add_route()->Swap(&p->route[i]); } - return sock.Send(p->route.back().mq_id(), head, body); + MQInfo dest = {p->route.back().mq_id(), p->route.back().abs_addr()}; + return sock.Send(dest, head, body); } bool TopicNode::ClientStartWorker(RequestResultCB const &cb, const int nworker) @@ -440,10 +445,10 @@ out_msg_id = msg_id; - auto SendTo = [this, msg_id](const BHAddress &addr, const MsgRequestTopic &req, const RequestResultCB &cb) { + auto SendTo = [this, msg_id](const MQInfo &remote, const MsgRequestTopic &req, const RequestResultCB &cb) { auto &sock = SockClient(); BHMsgHead head(InitMsgHead(GetType(req), proc_id(), ssn(), msg_id)); - AddRoute(head, sock.id()); + AddRoute(head, sock); head.set_topic(req.topic()); if (cb) { @@ -455,15 +460,15 @@ } } }; - return sock.Send(addr.mq_id(), head, req, onRecv); + return sock.Send(remote, head, req, onRecv); } else { - return sock.Send(addr.mq_id(), head, req); + return sock.Send(remote, head, req); } }; try { BHAddress addr; - return (ClientQueryRPCTopic(req.topic(), addr, 3000)) && SendTo(addr, req, cb); + return (ClientQueryRPCTopic(req.topic(), addr, 3000)) && SendTo(MQInfo{addr.mq_id(), addr.abs_addr()}, req, cb); } catch (...) { SetLastError(eError, "internal error."); return false; @@ -484,14 +489,14 @@ if (ClientQueryRPCTopic(request.topic(), addr, timeout_ms)) { LOG_TRACE() << "node: " << SockNode().id() << ", topic dest: " << addr.mq_id(); BHMsgHead head(InitMsgHead(GetType(request), proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); head.set_topic(request.topic()); MsgI reply_msg; DEFER1(reply_msg.Release();); BHMsgHead reply_head; - if (sock.SendAndRecv(addr.mq_id(), head, request, reply_msg, reply_head, timeout_ms) && + if (sock.SendAndRecv({addr.mq_id(), addr.abs_addr()}, head, request, reply_msg, reply_head, timeout_ms) && reply_head.type() == kMsgTypeRequestTopicReply && reply_msg.ParseBody(out_reply)) { reply_head.mutable_proc_id()->swap(out_proc_id); @@ -504,7 +509,7 @@ return false; } -int TopicNode::QueryRPCTopics(const Topic &topic, std::vector<NodeAddress> &addr, const int timeout_ms) +int TopicNode::QueryTopicServers(const Topic &topic, std::vector<NodeAddress> &addr, const int timeout_ms) { int n = 0; MsgQueryTopic query; @@ -532,7 +537,7 @@ return true; } std::vector<NodeAddress> lst; - if (QueryRPCTopics(topic, lst, timeout_ms)) { + if (QueryTopicServers(topic, lst, timeout_ms)) { addr = lst.front().addr(); if (addr.mq_id() != 0) { topic_query_cache_.Store(topic, addr); @@ -555,7 +560,7 @@ try { auto &sock = SockPub(); BHMsgHead head(InitMsgHead(GetType(pub), proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); if (timeout_ms == 0) { return sock.Send(BHTopicBusAddress(), head, pub); @@ -589,7 +594,7 @@ sub.mutable_topics()->Swap(&topics); BHMsgHead head(InitMsgHead(GetType(sub), proc_id(), ssn())); - AddRoute(head, sock.id()); + AddRoute(head, sock); if (timeout_ms == 0) { return sock.Send(BHTopicBusAddress(), head, sub); } else { -- Gitblit v1.8.0