Merge branch 'master' of https://github.com/divyagayathri-hcl/sonic-s…

…wss-common into zmq
sonic-net · Dec 4, 2024 · 19269ef · 19269ef
2 parents c19af49 + aa1021f
commit 19269ef
Show file tree

Hide file tree

Showing 6 changed files with 234 additions and 25 deletions.
diff --git a/common/redispipeline.h b/common/redispipeline.h
@@ -164,7 +164,7 @@ class RedisPipeline {
             return;
 
         m_channels.insert(channel);
-        m_luaPub += "redis.call('PUBLISH', '" + channel + "', 'G');";
+        m_luaPub += "redis.call('PUBLISH', '" + channel + "', 'G')\n";
         m_shaPub = loadRedisScript(m_luaPub);
     }
 

diff --git a/common/zmqclient.cpp b/common/zmqclient.cpp
@@ -20,8 +20,10 @@ ZmqClient::ZmqClient(const std::string& endpoint)
 {
 }
 
-ZmqClient::ZmqClient(const std::string& endpoint, const std::string& vrf)
+ZmqClient::ZmqClient(const std::string& endpoint, const std::string& vrf, uint32_t waitTimeMs) :
+    m_waitTimeMs(waitTimeMs)
 {
+//    m_waitTimeMs = waitTimeMs;
     initialize(endpoint, vrf);
 }
 
@@ -202,7 +204,31 @@ bool ZmqClient::wait(std::string& dbName,
                      std::vector<std::shared_ptr<KeyOpFieldsValuesTuple>>& kcos)
 {
     SWSS_LOG_ENTER();
+
+    zmq_pollitem_t items [1] = { };
+    items[0].socket = m_socket;
+    items[0].events = ZMQ_POLLIN;
+
     int rc;
+    for (int i = 0; true ; ++i)
+    {
+        rc = zmq_poll(items, 1, (int)m_waitTimeMs);
+        if (rc == 0)
+        {
+            SWSS_LOG_ERROR("zmq_poll timed out");
+            return false;
+        }
+        if (rc > 0)
+        {
+            break;
+        }
+        if (zmq_errno() == EINTR && i <= MQ_MAX_RETRY)
+        {
+            continue;
+        }
+        SWSS_LOG_THROW("zmq_poll failed, zmqerrno: %d", zmq_errno());
+    }
+
     for (int i = 0; true ; ++i)
     {
         rc = zmq_recv(m_socket, m_sendbuffer.data(), m_sendbuffer.size(), 0);
@@ -212,7 +238,8 @@ bool ZmqClient::wait(std::string& dbName,
             {
                 continue;
             }
-            SWSS_LOG_THROW("zmq_recv failed, zmqerrno: %d", zmq_errno());
+            SWSS_LOG_ERROR("zmq_recv failed, zmqerrno: %d", zmq_errno());
+            return false;
         }
         if (rc >= (int)m_sendbuffer.size())
         {

diff --git a/common/zmqclient.h b/common/zmqclient.h
@@ -14,7 +14,7 @@ class ZmqClient
 public:
 
     ZmqClient(const std::string& endpoint);
-    ZmqClient(const std::string& endpoint, const std::string& vrf);
+    ZmqClient(const std::string& endpoint, const std::string& vrf, uint32_t waitTimeMs = 0);
     ~ZmqClient();
 
     bool isConnected();
@@ -42,6 +42,8 @@ class ZmqClient
 
     bool m_connected;
 
+    uint32_t m_waitTimeMs;
+
     std::mutex m_socketMutex;
 
     std::vector<char> m_sendbuffer;

diff --git a/common/zmqserver.cpp b/common/zmqserver.cpp
@@ -21,6 +21,7 @@ ZmqServer::ZmqServer(const std::string& endpoint, const std::string& vrf)
     m_vrf(vrf),
     m_allowZmqPoll(true)
 {
+    connect();
     m_buffer.resize(MQ_RESPONSE_MAX_COUNT);
     m_runThread = true;
     m_mqPollThread = std::make_shared<std::thread>(&ZmqServer::mqPollThread, this);
@@ -33,6 +34,9 @@ ZmqServer::~ZmqServer()
     m_allowZmqPoll = true;
     m_runThread = false;
     m_mqPollThread->join();
+
+    zmq_close(m_socket);
+    zmq_ctx_destroy(m_context);
 }
 
 void ZmqServer::registerMessageHandler(
@@ -87,37 +91,40 @@ void ZmqServer::handleReceivedData(const char* buffer, const size_t size)
     handler->handleReceivedData(kcos);
 }
 
-void ZmqServer::mqPollThread()
+void ZmqServer::connect()
 {
     SWSS_LOG_ENTER();
-    SWSS_LOG_NOTICE("mqPollThread begin");
+    m_context = zmq_ctx_new();
 
-    // Producer/Consumer state table are n:1 mapping, so need use PUSH/PULL pattern http://api.zeromq.org/master:zmq-socket
-    void* context = zmq_ctx_new();;
-    void* socket = zmq_socket(context, ZMQ_PULL);
+    m_socket = zmq_socket(m_context, ZMQ_PULL);
 
     // Increase recv buffer for use all bandwidth:  http://api.zeromq.org/4-2:zmq-setsockopt
     int high_watermark = MQ_WATERMARK;
-    zmq_setsockopt(socket, ZMQ_RCVHWM, &high_watermark, sizeof(high_watermark));
+    zmq_setsockopt(m_socket, ZMQ_RCVHWM, &high_watermark, sizeof(high_watermark));
 
     if (!m_vrf.empty())
-    {
+    {   
         zmq_setsockopt(socket, ZMQ_BINDTODEVICE, m_vrf.c_str(), m_vrf.length());
-    }
+    } 
 
-    int rc = zmq_bind(socket, m_endpoint.c_str());
+    int rc = zmq_bind(m_socket, m_endpoint.c_str());
     if (rc != 0)
     {
-        SWSS_LOG_THROW("zmq_bind failed on endpoint: %s, zmqerrno: %d, message: %s",
+        SWSS_LOG_THROW("zmq_bind failed on endpoint: %s, zmqerrno: %d",
                 m_endpoint.c_str(),
-                zmq_errno(),
-                strerror(zmq_errno()));
+                zmq_errno());
     }
+}
+
+void ZmqServer::mqPollThread()
+{
+    SWSS_LOG_ENTER();
+    SWSS_LOG_NOTICE("mqPollThread begin");
 
     // zmq_poll will use less CPU
     zmq_pollitem_t poll_item;
     poll_item.fd = 0;
-    poll_item.socket = socket;
+    poll_item.socket = m_socket;
     poll_item.events = ZMQ_POLLIN;
     poll_item.revents = 0;
 
@@ -165,10 +172,79 @@ void ZmqServer::mqPollThread()
         handleReceivedData(m_buffer.data(), rc);
     }
 
-    zmq_close(socket);
-    zmq_ctx_destroy(context);
+    while (!m_allowZmqPoll)
+        {
+            usleep(10);
+        }
+    }
 
     SWSS_LOG_NOTICE("mqPollThread end");
 }
 
+void ZmqServer::sendMsg(const std::string& dbName, const std::string& tableName,
+        const std::vector<swss::KeyOpFieldsValuesTuple>& values)
+{
+    int serializedlen = (int)BinarySerializer::serializeBuffer(
+                                                        m_buffer.data(),
+                                                        m_buffer.size(),
+                                                        dbName,
+                                                        tableName,
+                                                        values);
+    SWSS_LOG_DEBUG("sending: %d", serializedlen);
+    int zmq_err = 0;
+    int retry_delay = 10;
+    int rc = 0;
+    for (int i = 0; i <=  MQ_MAX_RETRY; ++i)
+    {
+        rc = zmq_send(m_socket, m_buffer.data(), serializedlen, 0);
+
+        if (rc >= 0)
+        {
+            m_allowZmqPoll = true;
+            SWSS_LOG_DEBUG("zmq sent %d bytes", serializedlen);
+            return;
+        }
+        zmq_err = zmq_errno();
+        // sleep (2 ^ retry time) * 10 ms
+        retry_delay *= 2;
+        if (zmq_err == EINTR
+            || zmq_err== EFSM)
+        {
+            // EINTR: interrupted by signal
+            // EFSM: socket state not ready
+            //       For example when ZMQ socket still not receive reply message from last sended package.
+            //       There was state machine inside ZMQ socket, when the socket is not in ready to send state, this
+            //       error will happen.
+            // for more detail, please check: http://api.zeromq.org/2-1:zmq-send
+            SWSS_LOG_DEBUG("zmq send retry, endpoint: %s, error: %d", m_endpoint.c_str(), zmq_err);
+
+            retry_delay = 0;
+        }
+        else if (zmq_err == EAGAIN)
+        {
+            // EAGAIN: ZMQ is full to need try again
+            SWSS_LOG_WARN("zmq is full, will retry in %d ms, endpoint: %s, error: %d", retry_delay, m_endpoint.c_str(), zmq_err);
+        }
+        else if (zmq_err == ETERM)
+        {
+            auto message =  "zmq connection break, endpoint: " + m_endpoint + ", error: " + to_string(rc);
+            SWSS_LOG_ERROR("%s", message.c_str());
+            throw system_error(make_error_code(errc::connection_reset), message);
+        }
+        else
+        {
+            // for other error, send failed immediately.
+            auto message =  "zmq send failed, endpoint: " + m_endpoint + ", error: " + to_string(rc);
+            SWSS_LOG_ERROR("%s", message.c_str());
+            throw system_error(make_error_code(errc::io_error), message);
+        }
+        usleep(retry_delay * 1000);
+    }
+
+    // failed after retry
+    auto message =  "zmq send failed, endpoint: " + m_endpoint + ", zmqerrno: " + to_string(zmq_err) + ":" + zmq_strerror(zmq_err) + ", msg length:" + to_string(serializedlen);
+    SWSS_LOG_ERROR("%s", message.c_str());
+    throw system_error(make_error_code(errc::io_error), message);
+}
+
 }
diff --git a/common/zmqserver.h b/common/zmqserver.h
@@ -39,7 +39,12 @@ class ZmqServer
                                 const std::string tableName,
                                 ZmqMessageHandler* handler);
 
+    void sendMsg(const std::string& dbName, const std::string& tableName,
+        const std::vector<swss::KeyOpFieldsValuesTuple>& values);
+
 private:
+    void connect();
+
     void handleReceivedData(const char* buffer, const size_t size);
 
     void mqPollThread();
@@ -56,8 +61,14 @@ class ZmqServer
 
     std::string m_vrf;
 
+    void* m_context;
+
+    void* m_socket;
+
     bool m_allowZmqPoll;
 
+    std::vector<char> m_sendbuffer;
+
     std::map<std::string, std::map<std::string, ZmqMessageHandler*>> m_HandlerMap;
 };
 

diff --git a/tests/zmq_state_ut.cpp b/tests/zmq_state_ut.cpp
@@ -288,9 +288,6 @@ static void testMethod(bool producerPersistence)
     // start consumer first, SHM can only have 1 consumer per table.
     thread *consumerThread = new thread(consumerWorker, testTableName, pullEndpoint, !producerPersistence);
 
-    // Wait for the consumer to start.
-    sleep(1);
-
     cout << "Starting " << NUMBER_OF_THREADS << " producers" << endl;
     /* Starting the producer before the producer */
     for (int i = 0; i < NUMBER_OF_THREADS; i++)
@@ -354,9 +351,6 @@ static void testBatchMethod(bool producerPersistence)
     // start consumer first, SHM can only have 1 consumer per table.
     thread *consumerThread = new thread(consumerWorker, testTableName, pullEndpoint, !producerPersistence);
 
-    // Wait for the consumer to start.
-    sleep(1);
-
     cout << "Starting " << NUMBER_OF_THREADS << " producers" << endl;
     /* Starting the producer before the producer */
     for (int i = 0; i < NUMBER_OF_THREADS; i++)
@@ -471,3 +465,102 @@ TEST(ZmqProducerStateTableDeleteAfterSend, test)
     table.getKeys(keys);
     EXPECT_EQ(keys.front(), testKey);
 }
+
+static bool zmq_done = false;
+
+static void zmqConsumerWorker(string tableName, string endpoint, bool dbPersistence)
+{
+    cout << "Consumer thread started: " << tableName << endl;
+    DBConnector db(TEST_DB, 0, true);
+    ZmqServer server(endpoint);
+    ZmqConsumerStateTable c(&db, tableName, server, 128, 0, dbPersistence);
+    Select cs;
+    cs.addSelectable(&c);
+    //validate received data
+    Selectable *selectcs;
+    std::deque<KeyOpFieldsValuesTuple> vkco;
+    int ret = 0;
+    while (!zmq_done)
+    {
+        ret = cs.select(&selectcs, 10, true);
+        if (ret == Select::OBJECT)
+        {
+            c.pops(vkco);
+            std::vector<swss::KeyOpFieldsValuesTuple> values;
+            values.push_back(KeyOpFieldsValuesTuple{"k", SET_COMMAND, std::vector<FieldValueTuple>{FieldValueTuple{"f", "v"}}});
+            server.sendMsg(TEST_DB, tableName, values);
+        }
+    }
+
+    allDataReceived = true;
+    if (dbPersistence)
+    {
+        // wait all persist data write to redis
+        while (c.dbUpdaterQueueSize() > 0)
+        {
+            sleep(1);
+        }
+    }
+
+    cout << "Consumer thread ended: " << tableName << endl;
+}
+
+static void ZmqWithResponse(bool producerPersistence)
+{
+    std::string testTableName = "ZMQ_PROD_CONS_UT";
+    std::string pushEndpoint = "tcp://localhost:1234";
+    std::string pullEndpoint = "tcp://*:1234";
+    // start consumer first, SHM can only have 1 consumer per table.
+    thread *consumerThread = new thread(zmqConsumerWorker, testTableName, pullEndpoint, !producerPersistence);
+
+    // Wait for the consumer to be ready.
+    sleep(1);
+    DBConnector db(TEST_DB, 0, true);
+    ZmqClient client(pushEndpoint, 0, 3000);
+    ZmqProducerStateTable p(&db, testTableName, client, true);
+    std::vector<KeyOpFieldsValuesTuple> kcos;
+    kcos.push_back(KeyOpFieldsValuesTuple{"k", SET_COMMAND, std::vector<FieldValueTuple>{FieldValueTuple{"f", "v"}}});
+    std::vector<std::shared_ptr<KeyOpFieldsValuesTuple>> kcos_p;
+    std::string dbName, tableName;
+    for (int i =0; i < 3; ++i)
+    {
+        p.send(kcos);
+        ASSERT_TRUE(p.wait(dbName, tableName, kcos_p));
+        EXPECT_EQ(dbName, TEST_DB);
+        EXPECT_EQ(tableName, testTableName);
+        ASSERT_EQ(kcos_p.size(), 1);
+        EXPECT_EQ(kfvKey(*kcos_p[0]), "k");
+        EXPECT_EQ(kfvOp(*kcos_p[0]), SET_COMMAND);
+        std::vector<FieldValueTuple> cos = std::vector<FieldValueTuple>{FieldValueTuple{"f", "v"}};
+        EXPECT_EQ(kfvFieldsValues(*kcos_p[0]), cos);
+    }
+
+    zmq_done = true;
+    consumerThread->join();
+    delete consumerThread;
+}
+
+TEST(ZmqWithResponse, test)
+{
+    // test with persist by consumer
+    ZmqWithResponse(false);
+}
+
+TEST(ZmqWithResponseClientError, test)
+{
+    std::string testTableName = "ZMQ_PROD_CONS_UT";
+    std::string pushEndpoint = "tcp://localhost:1234";
+    DBConnector db(TEST_DB, 0, true);
+    ZmqClient client(pushEndpoint, 0, 3000);
+    ZmqProducerStateTable p(&db, testTableName, client, true);
+    std::vector<KeyOpFieldsValuesTuple> kcos;
+    kcos.push_back(KeyOpFieldsValuesTuple{"k", SET_COMMAND, std::vector<FieldValueTuple>{}});
+    std::vector<std::shared_ptr<KeyOpFieldsValuesTuple>> kcos_p;
+    std::string dbName, tableName;
+    p.send(kcos);
+    // Wait will timeout without server reply.
+    EXPECT_FALSE(p.wait(dbName, tableName, kcos_p));
+    // Send will return error without client request.
+    EXPECT_THROW(server.sendMsg(TEST_DB, testTableName, values), std::system_error);
+}
+