diff --git a/pulsar-client-cpp/.gitignore b/pulsar-client-cpp/.gitignore
index 0d8d323e7ff97..2b3f63fbc75d3 100644
--- a/pulsar-client-cpp/.gitignore
+++ b/pulsar-client-cpp/.gitignore
@@ -75,6 +75,7 @@ Makefile
 cmake_install.cmake
 CMakeFiles
 CMakeCache.txt
+.cmake
 
 pulsar-dist
 install_manifest.txt
diff --git a/pulsar-client-cpp/include/pulsar/ConsumerConfiguration.h b/pulsar-client-cpp/include/pulsar/ConsumerConfiguration.h
index b326ca8fb3151..e927ad7c884de 100644
--- a/pulsar-client-cpp/include/pulsar/ConsumerConfiguration.h
+++ b/pulsar-client-cpp/include/pulsar/ConsumerConfiguration.h
@@ -499,6 +499,27 @@ class PULSAR_PUBLIC ConsumerConfiguration {
      */
     bool isAutoAckOldestChunkedMessageOnQueueFull() const;
 
+    /**
+     * If this is enabled, consumer receiver queue size is init as a very small value, 1 by default,
+     * and it will double itself until it reaches the value set by {@link #receiverQueueSize(int)}, if and only if
+     * 1) User calls receive() and there is no messages in receiver queue.
+     * 2) The last message we put in the receiver queue took the last space available in receiver queue.
+     *
+     * This is disabled by default and currentReceiverQueueSize is init as maxReceiverQueueSize.
+     *
+     * The feature should be able to reduce client memory usage.
+     *
+     * Default: false
+     *
+     * @param enabled whether to enable AutoScaledReceiverQueueSize.
+     */
+    ConsumerConfiguration& setAutoScaledReceiverQueueSizeEnabled(bool enabled);
+
+    /**
+     * The associated getter of autoScaledReceiverQueueSizeEnabled
+     */
+    bool isAutoScaledReceiverQueueSizeEnabled() const;
+
     friend class PulsarWrapper;
 
    private:
diff --git a/pulsar-client-cpp/lib/ClientImpl.cc b/pulsar-client-cpp/lib/ClientImpl.cc
index d15e247347a6f..4b504c25a8c85 100644
--- a/pulsar-client-cpp/lib/ClientImpl.cc
+++ b/pulsar-client-cpp/lib/ClientImpl.cc
@@ -92,7 +92,12 @@ ClientImpl::ClientImpl(const std::string& serviceUrl, const ClientConfiguration&
       state_(Open),
       serviceUrl_(serviceUrl),
       clientConfiguration_(detectTls(serviceUrl, clientConfiguration)),
-      memoryLimitController_(clientConfiguration.getMemoryLimit()),
+      memoryLimitController_(
+          clientConfiguration_.getMemoryLimit(),
+          clientConfiguration_.getMemoryLimit() * MEMORY_THRESHOLD_FOR_RECEIVER_QUEUE_SIZE_EXPANSION5,
+          [this]() {
+                     std::for_each(consumers_.begin(), consumers_.end(), [](ConsumerImplBaseWeakPtr consumer) { consumer.lock()->reduceCurrentReceiverQueueSize(); });
+          }),
       ioExecutorProvider_(std::make_shared<ExecutorServiceProvider>(clientConfiguration_.getIOThreads())),
       listenerExecutorProvider_(
           std::make_shared<ExecutorServiceProvider>(clientConfiguration_.getMessageListenerThreads())),
diff --git a/pulsar-client-cpp/lib/ConsumerConfiguration.cc b/pulsar-client-cpp/lib/ConsumerConfiguration.cc
index 2b58835cdbea3..8e9549539c0be 100644
--- a/pulsar-client-cpp/lib/ConsumerConfiguration.cc
+++ b/pulsar-client-cpp/lib/ConsumerConfiguration.cc
@@ -19,6 +19,7 @@
 #include <lib/ConsumerConfigurationImpl.h>
 
 #include <stdexcept>
+#include "pulsar/ConsumerConfiguration.h"
 
 namespace pulsar {
 
@@ -259,5 +260,12 @@ ConsumerConfiguration& ConsumerConfiguration::setAutoAckOldestChunkedMessageOnQu
 bool ConsumerConfiguration::isAutoAckOldestChunkedMessageOnQueueFull() const {
     return impl_->autoAckOldestChunkedMessageOnQueueFull;
 }
+ConsumerConfiguration& ConsumerConfiguration::setAutoScaledReceiverQueueSizeEnabled(bool enabled) {
+    impl_->autoScaledReceiverQueueSizeEnabled = enabled;
+    return *this;
+}
+bool ConsumerConfiguration::isAutoScaledReceiverQueueSizeEnabled() const {
+    return impl_->autoScaledReceiverQueueSizeEnabled;
+}
 
 }  // namespace pulsar
diff --git a/pulsar-client-cpp/lib/ConsumerConfigurationImpl.h b/pulsar-client-cpp/lib/ConsumerConfigurationImpl.h
index 1c13f729b55e0..720fe535bb3dc 100644
--- a/pulsar-client-cpp/lib/ConsumerConfigurationImpl.h
+++ b/pulsar-client-cpp/lib/ConsumerConfigurationImpl.h
@@ -53,6 +53,7 @@ struct ConsumerConfigurationImpl {
     KeySharedPolicy keySharedPolicy;
     size_t maxPendingChunkedMessage{10};
     bool autoAckOldestChunkedMessageOnQueueFull{false};
+    bool autoScaledReceiverQueueSizeEnabled{false};
 };
 }  // namespace pulsar
 #endif /* LIB_CONSUMERCONFIGURATIONIMPL_H_ */
diff --git a/pulsar-client-cpp/lib/ConsumerImpl.cc b/pulsar-client-cpp/lib/ConsumerImpl.cc
index 63d2afc29bcab..57b0b9df95f77 100644
--- a/pulsar-client-cpp/lib/ConsumerImpl.cc
+++ b/pulsar-client-cpp/lib/ConsumerImpl.cc
@@ -31,11 +31,14 @@
 #include "AckGroupingTrackerDisabled.h"
 #include <exception>
 #include <algorithm>
+#include <math.h>
 
 namespace pulsar {
 
 DECLARE_LOG_OBJECT()
 
+const static int INITIAL_RECEIVER_QUEUE_SIZE = 1;
+
 ConsumerImpl::ConsumerImpl(const ClientImplPtr client, const std::string& topic,
                            const std::string& subscriptionName, const ConsumerConfiguration& conf,
                            const ExecutorServicePtr listenerExecutor /* = NULL by default */,
@@ -55,6 +58,7 @@ ConsumerImpl::ConsumerImpl(const ClientImplPtr client, const std::string& topic,
       // This is the initial capacity of the queue
       incomingMessages_(std::max(config_.getReceiverQueueSize(), 1)),
       availablePermits_(0),
+      scaleReceiverQueueHint(false),
       receiverQueueRefillThreshold_(config_.getReceiverQueueSize() / 2),
       consumerId_(client->newConsumerId()),
       consumerName_(config_.getConsumerName()),
@@ -103,6 +107,7 @@ ConsumerImpl::ConsumerImpl(const ClientImplPtr client, const std::string& topic,
     if (conf.isEncryptionEnabled()) {
         msgCrypto_ = std::make_shared<MessageCrypto>(consumerStr_, false);
     }
+    initReceiverQueueSize();
 }
 
 ConsumerImpl::~ConsumerImpl() {
@@ -880,10 +885,12 @@ Optional<MessageId> ConsumerImpl::clearReceiveQueue() {
 void ConsumerImpl::increaseAvailablePermits(const ClientConnectionPtr& currentCnx, int delta) {
     int newAvailablePermits = availablePermits_.fetch_add(delta) + delta;
 
-    while (newAvailablePermits >= receiverQueueRefillThreshold_ && messageListenerRunning_) {
+    while (newAvailablePermits >= getCurrentReceiverQueueSize() / 2 && messageListenerRunning_) {
         if (availablePermits_.compare_exchange_weak(newAvailablePermits, 0)) {
             sendFlowPermitsToBroker(currentCnx, newAvailablePermits);
             break;
+        } else {
+            newAvailablePermits = availablePermits_;
         }
     }
 }
@@ -1385,4 +1392,51 @@ bool ConsumerImpl::isConnected() const {
 
 uint64_t ConsumerImpl::getNumberOfConnectedConsumer() { return isConnected() ? 1 : 0; }
 
-} /* namespace pulsar */
+MemoryLimitController& ConsumerImpl::getMemoryLimitController() {
+    return client_.lock()->getMemoryLimitController();
+}
+
+void ConsumerImpl::reduceCurrentReceiverQueueSize() {
+    if (!config_.isAutoScaledReceiverQueueSizeEnabled()) {
+        return ;
+    }
+    int oldSize = getCurrentReceiverQueueSize();
+    int newSize = std::max(minReceiverQueueSize(), oldSize / 2);
+    if (oldSize > newSize) {
+        setCurrentReceiverQueueSize(newSize);
+    }
+}
+
+void ConsumerImpl::expectMoreIncomingMessages() {
+    if (!config_.isAutoScaledReceiverQueueSizeEnabled()) {
+        return ;
+    }
+    double usage = getMemoryLimitController().currentUsagePercent();
+    if (bool expectedState = true && usage < MEMORY_THRESHOLD_FOR_RECEIVER_QUEUE_SIZE_EXPANSION5
+        && scaleReceiverQueueHint.compare_exchange_strong(expectedState, false)) {
+        int oldSize = getCurrentReceiverQueueSize();
+        int newSize = std::min(config_.getReceiverQueueSize(), oldSize * 2);
+        setCurrentReceiverQueueSize(newSize);
+    }
+}
+void ConsumerImpl::initReceiverQueueSize() {
+    if (config_.isAutoScaledReceiverQueueSizeEnabled()) {
+        int size = minReceiverQueueSize();
+        currentReceiverQueueSize_.exchange(size);
+    } else {
+        currentReceiverQueueSize_.exchange(config_.getReceiverQueueSize());
+    }
+}
+
+void ConsumerImpl::setCurrentReceiverQueueSize(int newSize) {
+    currentReceiverQueueSize_.fetch_xor(newSize);
+}
+
+int ConsumerImpl::getCurrentReceiverQueueSize() {
+    return currentReceiverQueueSize_;
+}
+int ConsumerImpl::minReceiverQueueSize() {
+    int size = std::min(INITIAL_RECEIVER_QUEUE_SIZE, config_.getReceiverQueueSize());
+    return size;
+}
+} /* namespace pulsar */
\ No newline at end of file
diff --git a/pulsar-client-cpp/lib/ConsumerImpl.h b/pulsar-client-cpp/lib/ConsumerImpl.h
index 346d3515ad7f6..ea561ea8f7705 100644
--- a/pulsar-client-cpp/lib/ConsumerImpl.h
+++ b/pulsar-client-cpp/lib/ConsumerImpl.h
@@ -129,6 +129,7 @@ class ConsumerImpl : public ConsumerImplBase,
     void negativeAcknowledge(const MessageId& msgId) override;
     bool isConnected() const override;
     uint64_t getNumberOfConnectedConsumer() override;
+    void reduceCurrentReceiverQueueSize() override;
 
     virtual void disconnectConsumer();
     Result fetchSingleMessageFromBroker(Message& msg);
@@ -141,6 +142,7 @@ class ConsumerImpl : public ConsumerImplBase,
     virtual bool isReadCompacted();
     virtual void hasMessageAvailableAsync(HasMessageAvailableCallback callback);
     virtual void getLastMessageIdAsync(BrokerGetLastMessageIdCallback callback);
+    int getCurrentReceiverQueueSize();
 
    protected:
     // overrided methods from HandlerBase
@@ -157,6 +159,11 @@ class ConsumerImpl : public ConsumerImplBase,
     void handleClose(Result result, ResultCallback callback, ConsumerImplPtr consumer);
     ConsumerStatsBasePtr consumerStatsBasePtr_;
 
+    void setCurrentReceiverQueueSize(int newSize);
+    void expectMoreIncomingMessages();
+    void initReceiverQueueSize();
+    int minReceiverQueueSize();
+
    private:
     bool waitingForZeroQueueSizeMessage;
     bool uncompressMessageIfNeeded(const ClientConnectionPtr& cnx, const proto::MessageIdData& messageIdData,
@@ -173,6 +180,8 @@ class ConsumerImpl : public ConsumerImplBase,
     bool decryptMessageIfNeeded(const ClientConnectionPtr& cnx, const proto::CommandMessage& msg,
                                 const proto::MessageMetadata& metadata, SharedBuffer& payload);
 
+    MemoryLimitController& getMemoryLimitController();
+
     // TODO - Convert these functions to lambda when we move to C++11
     Result receiveHelper(Message& msg);
     Result receiveHelper(Message& msg, int timeout);
@@ -199,6 +208,8 @@ class ConsumerImpl : public ConsumerImplBase,
     UnboundedBlockingQueue<Message> incomingMessages_;
     std::queue<ReceiveCallback> pendingReceives_;
     std::atomic_int availablePermits_;
+    std::atomic_int currentReceiverQueueSize_;
+    std::atomic_bool scaleReceiverQueueHint;
     const int receiverQueueRefillThreshold_;
     uint64_t consumerId_;
     std::string consumerName_;
diff --git a/pulsar-client-cpp/lib/ConsumerImplBase.h b/pulsar-client-cpp/lib/ConsumerImplBase.h
index 693d4da9a3779..ab250cbfbbd0b 100644
--- a/pulsar-client-cpp/lib/ConsumerImplBase.h
+++ b/pulsar-client-cpp/lib/ConsumerImplBase.h
@@ -57,6 +57,7 @@ class ConsumerImplBase {
     virtual void negativeAcknowledge(const MessageId& msgId) = 0;
     virtual bool isConnected() const = 0;
     virtual uint64_t getNumberOfConnectedConsumer() = 0;
+    virtual void reduceCurrentReceiverQueueSize() = 0;
 
    private:
     virtual void setNegativeAcknowledgeEnabledForTesting(bool enabled) = 0;
diff --git a/pulsar-client-cpp/lib/MemoryLimitController.cc b/pulsar-client-cpp/lib/MemoryLimitController.cc
index f55da8e68a4c5..284d53a975924 100644
--- a/pulsar-client-cpp/lib/MemoryLimitController.cc
+++ b/pulsar-client-cpp/lib/MemoryLimitController.cc
@@ -22,7 +22,12 @@
 namespace pulsar {
 
 MemoryLimitController::MemoryLimitController(uint64_t memoryLimit)
-    : memoryLimit_(memoryLimit), currentUsage_(0), mutex_(), condition_() {}
+    : memoryLimit_(memoryLimit), currentUsage_(0), mutex_(), condition_(),
+      triggerThreshold_(0), trigger_(), triggerRunning(false) {}
+
+MemoryLimitController::MemoryLimitController(uint64_t memoryLimit, uint64_t triggerThreshold, Trigger trigger)
+    : memoryLimit_(memoryLimit), currentUsage_(0), mutex_(), condition_(), triggerThreshold_(triggerThreshold),
+      trigger_(trigger), triggerRunning(false) {}
 
 bool MemoryLimitController::tryReserveMemory(uint64_t size) {
     // Avoid CAS operation when size is 0
@@ -40,11 +45,30 @@ bool MemoryLimitController::tryReserveMemory(uint64_t size) {
         }
 
         if (currentUsage_.compare_exchange_strong(current, newUsage)) {
+            checkTrigger(current, newUsage);
             return true;
         }
     }
 }
 
+void MemoryLimitController::forceReserveMemory(uint64_t size) {
+    uint64_t newUsage = currentUsage_.fetch_add(size);
+    checkTrigger(newUsage - size, newUsage);
+}
+
+void MemoryLimitController::checkTrigger(uint64_t preUsage, uint64_t newUsage) {
+    if (newUsage >= triggerThreshold_ && preUsage < triggerThreshold_ && trigger_) {
+        bool expectedState = false;
+        if (triggerRunning.compare_exchange_strong(expectedState, true)) {
+            try {
+                trigger_();
+            } catch (const std::exception exception) {
+            }
+            triggerRunning.exchange(false);
+        }
+    }
+}
+
 bool MemoryLimitController::reserveMemory(uint64_t size) {
     if (!tryReserveMemory(size)) {
         std::unique_lock<std::mutex> lock(mutex_);
@@ -83,4 +107,8 @@ void MemoryLimitController::close() {
     condition_.notify_all();
 }
 
+double MemoryLimitController::currentUsagePercent() const {
+    return 1.0 * currentUsage_ / memoryLimit_;
+}
+
 }  // namespace pulsar
diff --git a/pulsar-client-cpp/lib/MemoryLimitController.h b/pulsar-client-cpp/lib/MemoryLimitController.h
index 38987ea0b68a0..23df2cbfdf2d2 100644
--- a/pulsar-client-cpp/lib/MemoryLimitController.h
+++ b/pulsar-client-cpp/lib/MemoryLimitController.h
@@ -25,14 +25,20 @@
 #include <mutex>
 
 namespace pulsar {
+typedef std::function<void(void)> Trigger;
+
+const static double MEMORY_THRESHOLD_FOR_RECEIVER_QUEUE_SIZE_EXPANSION5 = 0.75;
 
 class MemoryLimitController {
    public:
     explicit MemoryLimitController(uint64_t memoryLimit);
+    MemoryLimitController(uint64_t memoryLimit, uint64_t triggerThreshold, Trigger trigger);
+    void forceReserveMemory(uint64_t size);
     bool tryReserveMemory(uint64_t size);
     bool reserveMemory(uint64_t size);
     void releaseMemory(uint64_t size);
     uint64_t currentUsage() const;
+    double currentUsagePercent() const;
 
     void close();
 
@@ -42,6 +48,10 @@ class MemoryLimitController {
     std::mutex mutex_;
     std::condition_variable condition_;
     bool isClosed_ = false;
+    const uint64_t triggerThreshold_;
+    const Trigger trigger_;
+    std::atomic_bool triggerRunning;
+    void checkTrigger(uint64_t preUsage, uint64_t newUsage);
 };
 
 }  // namespace pulsar
\ No newline at end of file
diff --git a/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.cc b/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.cc
index 0ad9d60681048..c8a00e020f9b4 100644
--- a/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.cc
+++ b/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.cc
@@ -738,3 +738,4 @@ uint64_t MultiTopicsConsumerImpl::getNumberOfConnectedConsumer() {
     });
     return numberOfConnectedConsumer;
 }
+void MultiTopicsConsumerImpl::reduceCurrentReceiverQueueSize() {}
diff --git a/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.h b/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.h
index 98b2f318af95a..c0755624ac1c5 100644
--- a/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.h
+++ b/pulsar-client-cpp/lib/MultiTopicsConsumerImpl.h
@@ -80,6 +80,7 @@ class MultiTopicsConsumerImpl : public ConsumerImplBase,
     void negativeAcknowledge(const MessageId& msgId) override;
     bool isConnected() const override;
     uint64_t getNumberOfConnectedConsumer() override;
+    void reduceCurrentReceiverQueueSize() override;
 
     void handleGetConsumerStats(Result, BrokerConsumerStats, LatchPtr, MultiTopicsBrokerConsumerStatsPtr,
                                 size_t, BrokerConsumerStatsCallback);
diff --git a/pulsar-client-cpp/lib/PartitionedConsumerImpl.cc b/pulsar-client-cpp/lib/PartitionedConsumerImpl.cc
index 23288e2f81108..f4b3ef2c038b7 100644
--- a/pulsar-client-cpp/lib/PartitionedConsumerImpl.cc
+++ b/pulsar-client-cpp/lib/PartitionedConsumerImpl.cc
@@ -642,5 +642,6 @@ uint64_t PartitionedConsumerImpl::getNumberOfConnectedConsumer() {
     }
     return numberOfConnectedConsumer;
 }
+void PartitionedConsumerImpl::reduceCurrentReceiverQueueSize() {}
 
 }  // namespace pulsar
diff --git a/pulsar-client-cpp/lib/PartitionedConsumerImpl.h b/pulsar-client-cpp/lib/PartitionedConsumerImpl.h
index 7fa0ccdd1f4c9..7a822178e737d 100644
--- a/pulsar-client-cpp/lib/PartitionedConsumerImpl.h
+++ b/pulsar-client-cpp/lib/PartitionedConsumerImpl.h
@@ -75,6 +75,7 @@ class PartitionedConsumerImpl : public ConsumerImplBase,
     void negativeAcknowledge(const MessageId& msgId) override;
     bool isConnected() const override;
     uint64_t getNumberOfConnectedConsumer() override;
+    void reduceCurrentReceiverQueueSize() override;
 
     void handleGetConsumerStats(Result, BrokerConsumerStats, LatchPtr, PartitionedBrokerConsumerStatsPtr,
                                 size_t, BrokerConsumerStatsCallback);
diff --git a/pulsar-client-cpp/tests/MemoryLimitControllerTest.cc b/pulsar-client-cpp/tests/MemoryLimitControllerTest.cc
index eb63760eedfba..b00cac6eb00b1 100644
--- a/pulsar-client-cpp/tests/MemoryLimitControllerTest.cc
+++ b/pulsar-client-cpp/tests/MemoryLimitControllerTest.cc
@@ -41,6 +41,31 @@ TEST(MemoryLimitControllerTest, testLimit) {
     ASSERT_EQ(mlc.currentUsage(), 101);
 }
 
+TEST(MemoryLimitControllerTest, testTrigger) {
+    int num = 0;
+    MemoryLimitController mlc(100, 95, [&num]() {
+        num++;
+    });
+
+    mlc.tryReserveMemory(95);
+    ASSERT_EQ(num , 1);
+
+    mlc.releaseMemory(95);
+    ASSERT_EQ(mlc.currentUsage(), 0);
+
+    std::thread t1([&]() {
+        mlc.forceReserveMemory(95);
+    });
+
+    std::thread t2([&]() {
+      mlc.forceReserveMemory(95);
+    });
+
+    t1.join();
+    t2.join();
+    ASSERT_EQ(num, 2);
+}
+
 TEST(MemoryLimitControllerTest, testBlocking) {
     MemoryLimitController mlc(100);