Retry connection after getting disconnected from server. (#168)

* Set TCP keep alive and reset connection configurations for devices * update version * updated onConnectionResumed method * remove tcp keep alive * Updated log statements * updated version.
awslabs · Sep 3, 2021 · 679afbc · 679afbc
1 parent 1988dbc
commit 679afbc
Show file tree

Hide file tree

Showing 4 changed files with 54 additions and 11 deletions.
diff --git a/source/SharedCrtResourceManager.cpp b/source/SharedCrtResourceManager.cpp
@@ -21,8 +21,9 @@ using namespace Aws::Iot::DeviceClient::Logging;
 
 constexpr int SharedCrtResourceManager::DEFAULT_WAIT_TIME_SECONDS;
 
-bool SharedCrtResourceManager::initialize(const PlainConfig &config)
+bool SharedCrtResourceManager::initialize(const PlainConfig &config, vector<Feature *> *featuresList)
 {
+    features = featuresList;
     initializeAllocator();
     initialized = buildClient(config) == SharedCrtResourceManager::SUCCESS;
     return initialized;
@@ -167,6 +168,7 @@ int SharedCrtResourceManager::establishConnection(const PlainConfig &config)
     clientConfigBuilder.WithCertificateAuthority(config.rootCa->c_str());
     clientConfigBuilder.WithSdkName(SharedCrtResourceManager::BINARY_NAME);
     clientConfigBuilder.WithSdkVersion(SharedCrtResourceManager::BINARY_VERSION);
+
     auto clientConfig = clientConfigBuilder.Build();
 
     if (!clientConfig)
@@ -222,11 +224,43 @@ int SharedCrtResourceManager::establishConnection(const PlainConfig &config)
         }
     };
 
+    /*
+     * Invoked when connection is interrupted.
+     */
+    auto OnConnectionInterrupted = [&](Mqtt::MqttConnection &, int errorCode) {
+        {
+            if (errorCode)
+            {
+                LOGM_ERROR(
+                    TAG,
+                    "MQTT Connection interrupted with error: `%s`. Device Client will retry connection until it is "
+                    "successfully connected to the core. ",
+                    ErrorDebugString(errorCode));
+            }
+        }
+    };
+
+    /*
+     * Invoked when connection is resumed.
+     */
+    auto OnConnectionResumed = [&](Mqtt::MqttConnection &, int returnCode, bool) {
+        {
+            LOGM_INFO(TAG, "MQTT connection resumed with return code: %d", returnCode);
+        }
+    };
+
     connection->OnConnectionCompleted = move(onConnectionCompleted);
     connection->OnDisconnect = move(onDisconnect);
+    connection->OnConnectionInterrupted = move(OnConnectionInterrupted);
+    connection->OnConnectionResumed = move(OnConnectionResumed);
 
     LOGM_INFO(TAG, "Establishing MQTT connection with client id %s...", config.thingName->c_str());
-    if (!connection->Connect(config.thingName->c_str(), true, 0))
+    if (!connection->SetReconnectTimeout(15, 240))
+    {
+        LOG_ERROR(TAG, "Device Client is not able to set reconnection settings. Device Client will retry again.");
+        return RETRY;
+    }
+    if (!connection->Connect(config.thingName->c_str(), false))
     {
         LOGM_ERROR(TAG, "MQTT Connection failed with error: %s", ErrorDebugString(connection->LastError()));
         return RETRY;
@@ -306,3 +340,12 @@ void SharedCrtResourceManager::disconnect()
         LOG_ERROR(TAG, "MQTT Connection failed to disconnect");
     }
 }
+
+void SharedCrtResourceManager::startDeviceClientFeatures()
+{
+    LOG_INFO(TAG, "Starting Device Client features.");
+    for (auto *feature : *features)
+    {
+        feature->start();
+    }
+}
diff --git a/source/SharedCrtResourceManager.h b/source/SharedCrtResourceManager.h
@@ -4,6 +4,7 @@
 #ifndef DEVICE_CLIENT_SHAREDCRTRESOURCEMANAGER_H
 #define DEVICE_CLIENT_SHAREDCRTRESOURCEMANAGER_H
 
+#include "Feature.h"
 #include "config/Config.h"
 
 #include <atomic>
@@ -41,6 +42,7 @@ namespace Aws
                 std::unique_ptr<Aws::Iot::MqttClient> mqttClient;
                 std::shared_ptr<Crt::Mqtt::MqttConnection> connection;
                 struct aws_allocator *allocator;
+                std::vector<Feature *> *features;
 
                 bool locateCredentials(const PlainConfig &config);
                 int buildClient(const PlainConfig &config);
@@ -50,9 +52,10 @@ namespace Aws
                 static const int SUCCESS = 0;
                 static const int RETRY = 1;
                 static const int ABORT = 2;
-                bool initialize(const PlainConfig &config);
+                bool initialize(const PlainConfig &config, std::vector<Feature *> *features);
                 void initializeAWSHttpLib();
                 int establishConnection(const PlainConfig &config);
+                void startDeviceClientFeatures();
                 std::shared_ptr<Crt::Mqtt::MqttConnection> getConnection();
                 Aws::Crt::Io::EventLoopGroup *getEventLoopGroup();
                 struct aws_allocator *getAllocator();

diff --git a/source/main.cpp b/source/main.cpp
@@ -283,7 +283,7 @@ int main(int argc, char *argv[])
     shared_ptr<DefaultClientBaseNotifier> listener =
         shared_ptr<DefaultClientBaseNotifier>(new DefaultClientBaseNotifier);
     resourceManager = shared_ptr<SharedCrtResourceManager>(new SharedCrtResourceManager);
-    if (!resourceManager.get()->initialize(config.config))
+    if (!resourceManager.get()->initialize(config.config, &features))
     {
         LOGM_ERROR(
             TAG,
@@ -300,7 +300,7 @@ int main(int argc, char *argv[])
     {
 
         /*
-         * Establish MQTT connection using claim certificates and private key to provision device/thing.
+         * Establish MQTT connection using claim certificates and private key to provision the device/thing.
          */
 #    if !defined(DISABLE_MQTT)
         attemptConnection();
@@ -316,7 +316,7 @@ int main(int argc, char *argv[])
         {
             LOGM_ERROR(
                 TAG,
-                "*** %s: Failed to Provision thing or Validate newly created resources. "
+                "*** %s: Failed to Provision thing or validate newly created resources. "
                 "Please verify your AWS IoT credentials, "
                 "configuration, Fleet Provisioning Template, claim certificate and policy used. ***",
                 DC_FATAL_ERROR);
@@ -428,10 +428,7 @@ int main(int argc, char *argv[])
 #    endif
 #endif
 
-    for (auto &feature : features)
-    {
-        feature->start();
-    }
+    resourceManager->startDeviceClientFeatures();
     featuresReadWriteLock.unlock(); // UNLOCK
 
     // Now allow this thread to sleep until it's interrupted by a signal

diff --git a/source/util/Retry.cpp b/source/util/Retry.cpp
@@ -35,7 +35,7 @@ bool Retry::exponentialBackoff(
 
     if (config.maxRetries < 0)
     {
-        LOG_DEBUG(TAG, "Retryable function will retry until success");
+        LOG_DEBUG(TAG, "Retryable function starting, it will retry until success");
     }
 
     bool successful = false;