cppcache/src/TcrConnectionManager.cpp - geode-native - Git at Google

 /*
  * Licensed to the Apache Software Foundation (ASF) under one or more
  * contributor license agreements.  See the NOTICE file distributed with
  * this work for additional information regarding copyright ownership.
  * The ASF licenses this file to You under the Apache License, Version 2.0
  * (the "License"); you may not use this file except in compliance with
  * the License.  You may obtain a copy of the License at
  *
  *      http://www.apache.org/licenses/LICENSE-2.0
  *
  * Unless required by applicable law or agreed to in writing, software
  * distributed under the License is distributed on an "AS IS" BASIS,
  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */

 #include "TcrConnectionManager.hpp"

 #include <chrono>
 #include <set>
 #include <thread>

 #include <geode/SystemProperties.hpp>

 #include "CacheImpl.hpp"
 #include "ExpiryHandler_T.hpp"
 #include "ExpiryTaskManager.hpp"
 #include "RemoteQueryService.hpp"
 #include "ServerLocation.hpp"
 #include "TcrConnection.hpp"
 #include "TcrEndpoint.hpp"
 #include "TcrHADistributionManager.hpp"
 #include "ThinClientBaseDM.hpp"
 #include "ThinClientCacheDistributionManager.hpp"
 #include "ThinClientHARegion.hpp"
 #include "ThinClientLocatorHelper.hpp"
 #include "ThinClientRedundancyManager.hpp"
 #include "ThinClientRegion.hpp"
 #include "Utils.hpp"
 #include "util/exception.hpp"

 namespace apache {
 namespace geode {
 namespace client {

 volatile bool TcrConnectionManager::TEST_DURABLE_CLIENT_CRASH = false;

 const char *TcrConnectionManager::NC_Redundancy = "NC Redundancy";
 const char *TcrConnectionManager::NC_Failover = "NC Failover";
 const char *TcrConnectionManager::NC_CleanUp = "NC CleanUp";

 TcrConnectionManager::TcrConnectionManager(CacheImpl *cache)
     : m_cache(cache),
       m_initGuard(false),
       m_failoverSema(0),
       m_failoverTask(nullptr),
       m_cleanupSema(0),
       m_cleanupTask(nullptr),
       m_pingTaskId(-1),
       m_servermonitorTaskId(-1),
       // Create the queues with flag to not delete the objects
       m_notifyCleanupSemaList(false),
       m_redundancySema(0),
       m_redundancyTask(nullptr),
       m_isDurable(false),
       m_isNetDown(false) {
   m_redundancyManager = std::unique_ptr<ThinClientRedundancyManager>(
       new ThinClientRedundancyManager(this));
 }

 ExpiryTaskManager::id_type TcrConnectionManager::getPingTaskId() {
   return m_pingTaskId;
 }
 void TcrConnectionManager::init(bool isPool) {
   if (!m_initGuard) {
     m_initGuard = true;
   } else {
     return;
   }
   auto &props = m_cache->getDistributedSystem().getSystemProperties();
   m_isDurable = !props.durableClientId().empty();
   auto pingInterval = (props.pingInterval() / 2);
   if (!isPool) {
     ACE_Event_Handler *connectionChecker =
         new ExpiryHandler_T<TcrConnectionManager>(
             this, &TcrConnectionManager::checkConnection);
     m_pingTaskId = m_cache->getExpiryTaskManager().scheduleExpiryTask(
         connectionChecker, std::chrono::seconds(10), pingInterval, false);
     LOGFINE(
         "TcrConnectionManager::TcrConnectionManager Registered ping "
         "task with id = %ld, interval = %ld",
         m_pingTaskId, pingInterval.count());
   }

   m_redundancyManager->m_HAenabled = false;

   startFailoverAndCleanupThreads(isPool);
 }

 void TcrConnectionManager::startFailoverAndCleanupThreads(bool isPool) {
   if (!isPool && (m_failoverTask == nullptr || m_cleanupTask == nullptr)) {
     std::lock_guard<decltype(m_distMngrsLock)> _guard(m_distMngrsLock);
     if (!m_failoverTask) {
       m_failoverTask = std::unique_ptr<Task<TcrConnectionManager>>(
           new Task<TcrConnectionManager>(this, &TcrConnectionManager::failover,
                                          NC_Failover));
       m_failoverTask->start();
     }
     if (!m_cleanupTask) {
       if (m_redundancyManager->m_HAenabled) {
         m_redundancyManager->startPeriodicAck();
       }
       m_cleanupTask = std::unique_ptr<Task<TcrConnectionManager>>(
           new Task<TcrConnectionManager>(this, &TcrConnectionManager::cleanup,
                                          NC_CleanUp));
       m_cleanupTask->start();
     }
   }
 }

 void TcrConnectionManager::close() {
   LOGFINE("TcrConnectionManager is closing");

   if (m_pingTaskId > 0) {
     m_cache->getExpiryTaskManager().cancelTask(m_pingTaskId);
   }

   if (m_failoverTask != nullptr) {
     m_failoverTask->stopNoblock();
     m_failoverSema.release();
     m_failoverTask->wait();
     m_failoverTask = nullptr;
   }

   LOGFINE("TcrConnectionManager is closed");
 }

 void TcrConnectionManager::readyForEvents() {
   m_redundancyManager->readyForEvents();
 }

 TcrConnectionManager::~TcrConnectionManager() {
   if (m_cleanupTask != nullptr) {
     m_cleanupTask->stopNoblock();
     m_cleanupSema.release();
     m_cleanupTask->wait();
     // Clean notification lists if something remains in there; see bug #250
     cleanNotificationLists();
     m_cleanupTask = nullptr;

     // sanity cleanup of any remaining endpoints with warning; see bug #298
     //  cleanup of endpoints, when regions are destroyed via notification
     {
       auto &&guard = m_endpoints.make_lock();
       if (m_endpoints.size() > 0) {
         LOGFINE("TCCM: endpoints remain in destructor");
       }
     }
   }
   TcrConnectionManager::TEST_DURABLE_CLIENT_CRASH = false;
 }

 void TcrConnectionManager::connect(
     ThinClientBaseDM *distMng, std::vector<TcrEndpoint *> &endpoints,
     const std::unordered_set<std::string> &endpointStrs) {
   std::lock_guard<decltype(m_distMngrsLock)> guardDistMngrs(m_distMngrsLock);
   {
     auto &&endpointsGuard = m_endpoints.make_lock();
     int32_t numEndPoints = static_cast<int32_t>(endpointStrs.size());

     if (numEndPoints == 0) {
       LOGFINE(
           "TcrConnectionManager::connect(): Empty endpointstr vector "
           "passed to TCCM, will initialize endpoints list with all available "
           "endpoints (%zu).",
           m_endpoints.size());
       for (const auto &currItr : m_endpoints) {
         auto ep = currItr.second;
         ep->setNumRegions(ep->numRegions() + 1);
         LOGFINER(
             "TCCM 2: incremented region reference count for endpoint %s "
             "to %d",
             ep->name().c_str(), ep->numRegions());
         endpoints.push_back(ep.get());
       }
     } else {
       for (const auto &iter : endpointStrs) {
         auto ep = addRefToTcrEndpoint(iter, distMng);
         endpoints.push_back(ep);
       }
     }
   }

   m_distMngrs.push_back(distMng);

   // If a region/DM is joining after the marker has been
   // received then trigger it's marker flag.
   if (m_redundancyManager->m_globalProcessedMarker) {
     TcrHADistributionManager *tcrHADM =
         dynamic_cast<TcrHADistributionManager *>(distMng);
     if (tcrHADM != nullptr) {
       ThinClientHARegion *tcrHARegion =
           dynamic_cast<ThinClientHARegion *>(tcrHADM->m_region);
       tcrHARegion->setProcessedMarker();
     }
   }
 }

 TcrEndpoint *TcrConnectionManager::addRefToTcrEndpoint(std::string endpointName,
                                                        ThinClientBaseDM *dm) {
   std::shared_ptr<TcrEndpoint> ep;

   auto &&guard = m_endpoints.make_lock();
   const auto &find = m_endpoints.find(endpointName);
   if (find == m_endpoints.end()) {
     // this endpoint does not exist
     ep = std::make_shared<TcrEndpoint>(endpointName, m_cache, m_failoverSema,
                                        m_cleanupSema, m_redundancySema, dm,
                                        false);
     m_endpoints.emplace(endpointName, ep);
   } else {
     ep = find->second;
   }
   ep->setNumRegions(ep->numRegions() + 1);

   LOGFINER("TCCM: incremented region reference count for endpoint %s to %d",
            ep->name().c_str(), ep->numRegions());

   return ep.get();
 }

 void TcrConnectionManager::disconnect(ThinClientBaseDM *distMng,
                                       std::vector<TcrEndpoint *> &endpoints,
                                       bool keepEndpoints) {
   std::lock_guard<decltype(m_distMngrsLock)> guardDistMngrs(m_distMngrsLock);
   {
     auto &&guard = m_endpoints.make_lock();
     for (const auto &ep : endpoints) {
       removeRefToEndpoint(ep, keepEndpoints);
     }
   }

   m_distMngrs.remove(distMng);
 }

 bool TcrConnectionManager::removeRefToEndpoint(TcrEndpoint *ep,
                                                bool keepEndpoint) {
   bool hasRemovedEndpoint = false;

   if (keepEndpoint && (ep->numRegions() == 1)) {
     return false;
   }
   ep->setNumRegions(ep->numRegions() - 1);

   LOGFINER("TCCM: decremented region reference count for endpoint %s to %d",
            ep->name().c_str(), ep->numRegions());

   if (0 == ep->numRegions()) {
     // this endpoint no longer used
     m_endpoints.erase(ep->name());
     LOGFINE("delete endpoint %s", ep->name().c_str());
     _GEODE_SAFE_DELETE(ep);
     hasRemovedEndpoint = true;
   }
   return hasRemovedEndpoint;
 }

 int TcrConnectionManager::processEventIdMap(const ACE_Time_Value &currTime,
                                             const void *) {
   return m_redundancyManager->processEventIdMap(currTime, nullptr);
 }

 int TcrConnectionManager::checkConnection(const ACE_Time_Value &,
                                           const void *) {
   auto &&guard = m_endpoints.make_lock();
   for (const auto &currItr : m_endpoints) {
     if (currItr.second->connected() && !m_isNetDown) {
       currItr.second->pingServer();
     }
   }
   return 0;
 }

 int TcrConnectionManager::checkRedundancy(const ACE_Time_Value &,
                                           const void *) {
   m_redundancySema.release();
   return 0;
 }

 void TcrConnectionManager::failover(std::atomic<bool> &isRunning) {
   LOGFINE("TcrConnectionManager: starting failover thread");
   while (isRunning) {
     m_failoverSema.acquire();
     if (isRunning && !m_isNetDown) {
       try {
         std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
         for (const auto &it : m_distMngrs) {
           it->failover();
         }
         while (m_failoverSema.tryacquire() != -1) {
           ;
         }
       } catch (const Exception &e) {
         LOGERROR(e.what());
       } catch (const std::exception &e) {
         LOGERROR(e.what());
       } catch (...) {
         LOGERROR(
             "Unexpected exception while failing over to a "
             "different endpoint");
       }
     }
   }
   LOGFINE("TcrConnectionManager: ending failover thread");
 }

 void TcrConnectionManager::getAllEndpoints(
     std::vector<TcrEndpoint *> &endpoints) {
   auto &&guard = m_endpoints.make_lock();
   for (const auto &currItr : m_endpoints) {
     endpoints.push_back(currItr.second.get());
   }
 }

 int32_t TcrConnectionManager::getNumEndPoints() {
   return static_cast<int32_t>(m_endpoints.size());
 }

 GfErrType TcrConnectionManager::registerInterestAllRegions(
     TcrEndpoint *ep, const TcrMessage *request, TcrMessageReply *reply) {
   // Preconditions:
   // 1. m_distMngrs.size() > 1 (query distribution manager + 1 or more
   // TcrHADistributionManagers).

   GfErrType err = GF_NOERR;
   GfErrType opErr = GF_NOERR;
   std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
   std::list<ThinClientBaseDM *>::iterator begin = m_distMngrs.begin();
   std::list<ThinClientBaseDM *>::iterator end = m_distMngrs.end();
   for (std::list<ThinClientBaseDM *>::iterator it = begin; it != end; ++it) {
     TcrHADistributionManager *tcrHADM =
         dynamic_cast<TcrHADistributionManager *>(*it);
     if (tcrHADM != nullptr) {
       if ((opErr = tcrHADM->registerInterestForRegion(ep, request, reply)) !=
           GF_NOERR) {
         if (err == GF_NOERR) {
           err = opErr;
         }
       }
     }
   }
   return err;
 }
 GfErrType TcrConnectionManager::sendSyncRequestCq(TcrMessage &request,
                                                   TcrMessageReply &reply) {
   LOGDEBUG("TcrConnectionManager::sendSyncRequestCq");
   GfErrType err = GF_NOERR;
   // Preconditions:
   // 1. m_distMngrs.size() > 1 (query distribution manager + 1 or more
   // TcrHADistributionManagers).

   std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
   std::list<ThinClientBaseDM *>::iterator begin = m_distMngrs.begin();
   std::list<ThinClientBaseDM *>::iterator end = m_distMngrs.end();
   for (std::list<ThinClientBaseDM *>::iterator it = begin; it != end; ++it) {
     TcrHADistributionManager *tcrHADM =
         dynamic_cast<TcrHADistributionManager *>(*it);
     if (tcrHADM != nullptr) {
       return tcrHADM->sendSyncRequestCq(request, reply);
     }
   }
   return err;
 }

 void TcrConnectionManager::initializeHAEndpoints(const char *endpointsStr) {
   std::unordered_set<std::string> endpointsList;
   Utils::parseEndpointNamesString(endpointsStr, endpointsList);
   for (std::unordered_set<std::string>::iterator iter = endpointsList.begin();
        iter != endpointsList.end(); ++iter) {
     addRefToTcrEndpoint(*iter);
   }
   // Postconditions:
   // 1. endpointsList.size() > 0
 }

 void TcrConnectionManager::removeHAEndpoints() {
   auto &&guard = m_endpoints.make_lock();
   auto currItr = m_endpoints.begin();
   while (currItr != m_endpoints.end()) {
     if (removeRefToEndpoint(currItr->second.get())) {
       currItr = m_endpoints.begin();
     } else {
       currItr++;
     }
   }
 }

 void TcrConnectionManager::netDown() {
   m_isNetDown = true;

   //  sleep for 15 seconds to allow ping and redundancy threads to pause.
   std::this_thread::sleep_for(std::chrono::seconds(15));

   {
     auto &&guard = m_endpoints.make_lock();
     for (auto &currItr : m_endpoints) {
       currItr.second->setConnectionStatus(false);
     }
   }

   m_redundancyManager->netDown();
 }

 /* Need to do a get on unknown key after calling this Fn to restablish all
  * connection */
 void TcrConnectionManager::revive() {
   m_isNetDown = false;

   //  sleep for 15 seconds to allow redundancy thread to reestablish
   //  connections.
   std::this_thread::sleep_for(std::chrono::seconds(15));
 }

 void TcrConnectionManager::redundancy(std::atomic<bool> &isRunning) {
   LOGFINE("Starting subscription maintain redundancy thread.");
   while (isRunning) {
     m_redundancySema.acquire();
     if (isRunning && !m_isNetDown) {
       m_redundancyManager->maintainRedundancyLevel();
       while (m_redundancySema.tryacquire() != -1) {
         ;
       }
     }
   }
   LOGFINE("Ending subscription maintain redundancy thread.");
 }

 void TcrConnectionManager::addNotificationForDeletion(
     Task<TcrEndpoint> *notifyReceiver, TcrConnection *notifyConnection,
     ACE_Semaphore &notifyCleanupSema) {
   std::lock_guard<decltype(m_notificationLock)> guard(m_notificationLock);
   m_connectionReleaseList.put(notifyConnection);
   m_receiverReleaseList.put(notifyReceiver);
   m_notifyCleanupSemaList.put(&notifyCleanupSema);
 }

 void TcrConnectionManager::cleanup(std::atomic<bool> &isRunning) {
   LOGFINE("TcrConnectionManager: starting cleanup thread");
   do {
     //  If we block on acquire, the queue must be empty (precondition).
     if (m_receiverReleaseList.size() == 0) {
       LOGDEBUG(
           "TcrConnectionManager::cleanup(): waiting to acquire cleanup "
           "semaphore.");
       m_cleanupSema.acquire();
     }
     cleanNotificationLists();

     while (m_cleanupSema.tryacquire() != -1) {
       ;
     }

   } while (isRunning);

   LOGFINE("TcrConnectionManager: ending cleanup thread");
   //  Postcondition - all notification channels should be cleaned up by the end
   //  of this function.
 }

 void TcrConnectionManager::cleanNotificationLists() {
   Task<TcrEndpoint> *notifyReceiver;
   TcrConnection *notifyConnection;
   ACE_Semaphore *notifyCleanupSema;

   while (true) {
     {
       std::lock_guard<decltype(m_notificationLock)> guard(m_notificationLock);
       notifyReceiver = m_receiverReleaseList.get();
       if (!notifyReceiver) break;
       notifyConnection = m_connectionReleaseList.get();
       notifyCleanupSema = m_notifyCleanupSemaList.get();
     }
     notifyReceiver->wait();
     //_GEODE_SAFE_DELETE(notifyReceiver);
     _GEODE_SAFE_DELETE(notifyConnection);
     notifyCleanupSema->release();
   }
 }

 void TcrConnectionManager::processMarker() {
   // also set the static bool m_processedMarker for makePrimary messages
   m_redundancyManager->m_globalProcessedMarker = true;
 }

 //  TESTING: Durable clients - return queue status of endpoing. Not thread safe.
 bool TcrConnectionManager::getEndpointStatus(const std::string &endpoint) {
   for (auto &currItr : m_endpoints) {
     auto ep = currItr.second;
     const std::string epName = ep->name();
     if (epName == endpoint) return ep->getServerQueueStatusTEST();
   }
   return false;
 }

 GfErrType TcrConnectionManager::sendSyncRequestCq(
     TcrMessage &request, TcrMessageReply &reply,
     TcrHADistributionManager *theHADM) {
   return m_redundancyManager->sendSyncRequestCq(request, reply, theHADM);
 }

 GfErrType TcrConnectionManager::sendSyncRequestRegisterInterest(
     TcrMessage &request, TcrMessageReply &reply, bool attemptFailover,
     TcrEndpoint *endpoint, TcrHADistributionManager *theHADM,
     ThinClientRegion *region) {
   return m_redundancyManager->sendSyncRequestRegisterInterest(
       request, reply, attemptFailover, endpoint, theHADM, region);
 }
 }  // namespace client
 }  // namespace geode
 }  // namespace apache
	/*
	* Licensed to the Apache Software Foundation (ASF) under one or more
	* contributor license agreements. See the NOTICE file distributed with
	* this work for additional information regarding copyright ownership.
	* The ASF licenses this file to You under the Apache License, Version 2.0
	* (the "License"); you may not use this file except in compliance with
	* the License. You may obtain a copy of the License at
	*
	* http://www.apache.org/licenses/LICENSE-2.0
	*
	* Unless required by applicable law or agreed to in writing, software
	* distributed under the License is distributed on an "AS IS" BASIS,
	* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	* See the License for the specific language governing permissions and
	* limitations under the License.
	*/

	#include "TcrConnectionManager.hpp"

	#include <chrono>
	#include <set>
	#include <thread>

	#include <geode/SystemProperties.hpp>

	#include "CacheImpl.hpp"
	#include "ExpiryHandler_T.hpp"
	#include "ExpiryTaskManager.hpp"
	#include "RemoteQueryService.hpp"
	#include "ServerLocation.hpp"
	#include "TcrConnection.hpp"
	#include "TcrEndpoint.hpp"
	#include "TcrHADistributionManager.hpp"
	#include "ThinClientBaseDM.hpp"
	#include "ThinClientCacheDistributionManager.hpp"
	#include "ThinClientHARegion.hpp"
	#include "ThinClientLocatorHelper.hpp"
	#include "ThinClientRedundancyManager.hpp"
	#include "ThinClientRegion.hpp"
	#include "Utils.hpp"
	#include "util/exception.hpp"

	namespace apache {
	namespace geode {
	namespace client {

	volatile bool TcrConnectionManager::TEST_DURABLE_CLIENT_CRASH = false;

	const char *TcrConnectionManager::NC_Redundancy = "NC Redundancy";
	const char *TcrConnectionManager::NC_Failover = "NC Failover";
	const char *TcrConnectionManager::NC_CleanUp = "NC CleanUp";

	TcrConnectionManager::TcrConnectionManager(CacheImpl *cache)
	: m_cache(cache),
	m_initGuard(false),
	m_failoverSema(0),
	m_failoverTask(nullptr),
	m_cleanupSema(0),
	m_cleanupTask(nullptr),
	m_pingTaskId(-1),
	m_servermonitorTaskId(-1),
	// Create the queues with flag to not delete the objects
	m_notifyCleanupSemaList(false),
	m_redundancySema(0),
	m_redundancyTask(nullptr),
	m_isDurable(false),
	m_isNetDown(false) {
	m_redundancyManager = std::unique_ptr<ThinClientRedundancyManager>(
	new ThinClientRedundancyManager(this));
	}

	ExpiryTaskManager::id_type TcrConnectionManager::getPingTaskId() {
	return m_pingTaskId;
	}
	void TcrConnectionManager::init(bool isPool) {
	if (!m_initGuard) {
	m_initGuard = true;
	} else {
	return;
	}
	auto &props = m_cache->getDistributedSystem().getSystemProperties();
	m_isDurable = !props.durableClientId().empty();
	auto pingInterval = (props.pingInterval() / 2);
	if (!isPool) {
	ACE_Event_Handler *connectionChecker =
	new ExpiryHandler_T<TcrConnectionManager>(
	this, &TcrConnectionManager::checkConnection);
	m_pingTaskId = m_cache->getExpiryTaskManager().scheduleExpiryTask(
	connectionChecker, std::chrono::seconds(10), pingInterval, false);
	LOGFINE(
	"TcrConnectionManager::TcrConnectionManager Registered ping "
	"task with id = %ld, interval = %ld",
	m_pingTaskId, pingInterval.count());
	}

	m_redundancyManager->m_HAenabled = false;

	startFailoverAndCleanupThreads(isPool);
	}

	void TcrConnectionManager::startFailoverAndCleanupThreads(bool isPool) {
	if (!isPool && (m_failoverTask == nullptr \|\| m_cleanupTask == nullptr)) {
	std::lock_guard<decltype(m_distMngrsLock)> _guard(m_distMngrsLock);
	if (!m_failoverTask) {
	m_failoverTask = std::unique_ptr<Task<TcrConnectionManager>>(
	new Task<TcrConnectionManager>(this, &TcrConnectionManager::failover,
	NC_Failover));
	m_failoverTask->start();
	}
	if (!m_cleanupTask) {
	if (m_redundancyManager->m_HAenabled) {
	m_redundancyManager->startPeriodicAck();
	}
	m_cleanupTask = std::unique_ptr<Task<TcrConnectionManager>>(
	new Task<TcrConnectionManager>(this, &TcrConnectionManager::cleanup,
	NC_CleanUp));
	m_cleanupTask->start();
	}
	}
	}

	void TcrConnectionManager::close() {
	LOGFINE("TcrConnectionManager is closing");

	if (m_pingTaskId > 0) {
	m_cache->getExpiryTaskManager().cancelTask(m_pingTaskId);
	}

	if (m_failoverTask != nullptr) {
	m_failoverTask->stopNoblock();
	m_failoverSema.release();
	m_failoverTask->wait();
	m_failoverTask = nullptr;
	}

	LOGFINE("TcrConnectionManager is closed");
	}

	void TcrConnectionManager::readyForEvents() {
	m_redundancyManager->readyForEvents();
	}

	TcrConnectionManager::~TcrConnectionManager() {
	if (m_cleanupTask != nullptr) {
	m_cleanupTask->stopNoblock();
	m_cleanupSema.release();
	m_cleanupTask->wait();
	// Clean notification lists if something remains in there; see bug #250
	cleanNotificationLists();
	m_cleanupTask = nullptr;

	// sanity cleanup of any remaining endpoints with warning; see bug #298
	// cleanup of endpoints, when regions are destroyed via notification
	{
	auto &&guard = m_endpoints.make_lock();
	if (m_endpoints.size() > 0) {
	LOGFINE("TCCM: endpoints remain in destructor");
	}
	}
	}
	TcrConnectionManager::TEST_DURABLE_CLIENT_CRASH = false;
	}

	void TcrConnectionManager::connect(
	ThinClientBaseDM distMng, std::vector<TcrEndpoint > &endpoints,
	const std::unordered_set<std::string> &endpointStrs) {
	std::lock_guard<decltype(m_distMngrsLock)> guardDistMngrs(m_distMngrsLock);
	{
	auto &&endpointsGuard = m_endpoints.make_lock();
	int32_t numEndPoints = static_cast<int32_t>(endpointStrs.size());

	if (numEndPoints == 0) {
	LOGFINE(
	"TcrConnectionManager::connect(): Empty endpointstr vector "
	"passed to TCCM, will initialize endpoints list with all available "
	"endpoints (%zu).",
	m_endpoints.size());
	for (const auto &currItr : m_endpoints) {
	auto ep = currItr.second;
	ep->setNumRegions(ep->numRegions() + 1);
	LOGFINER(
	"TCCM 2: incremented region reference count for endpoint %s "
	"to %d",
	ep->name().c_str(), ep->numRegions());
	endpoints.push_back(ep.get());
	}
	} else {
	for (const auto &iter : endpointStrs) {
	auto ep = addRefToTcrEndpoint(iter, distMng);
	endpoints.push_back(ep);
	}
	}
	}

	m_distMngrs.push_back(distMng);

	// If a region/DM is joining after the marker has been
	// received then trigger it's marker flag.
	if (m_redundancyManager->m_globalProcessedMarker) {
	TcrHADistributionManager *tcrHADM =
	dynamic_cast<TcrHADistributionManager *>(distMng);
	if (tcrHADM != nullptr) {
	ThinClientHARegion *tcrHARegion =
	dynamic_cast<ThinClientHARegion *>(tcrHADM->m_region);
	tcrHARegion->setProcessedMarker();
	}
	}
	}

	TcrEndpoint *TcrConnectionManager::addRefToTcrEndpoint(std::string endpointName,
	ThinClientBaseDM *dm) {
	std::shared_ptr<TcrEndpoint> ep;

	auto &&guard = m_endpoints.make_lock();
	const auto &find = m_endpoints.find(endpointName);
	if (find == m_endpoints.end()) {
	// this endpoint does not exist
	ep = std::make_shared<TcrEndpoint>(endpointName, m_cache, m_failoverSema,
	m_cleanupSema, m_redundancySema, dm,
	false);
	m_endpoints.emplace(endpointName, ep);
	} else {
	ep = find->second;
	}
	ep->setNumRegions(ep->numRegions() + 1);

	LOGFINER("TCCM: incremented region reference count for endpoint %s to %d",
	ep->name().c_str(), ep->numRegions());

	return ep.get();
	}

	void TcrConnectionManager::disconnect(ThinClientBaseDM *distMng,
	std::vector<TcrEndpoint *> &endpoints,
	bool keepEndpoints) {
	std::lock_guard<decltype(m_distMngrsLock)> guardDistMngrs(m_distMngrsLock);
	{
	auto &&guard = m_endpoints.make_lock();
	for (const auto &ep : endpoints) {
	removeRefToEndpoint(ep, keepEndpoints);
	}
	}

	m_distMngrs.remove(distMng);
	}

	bool TcrConnectionManager::removeRefToEndpoint(TcrEndpoint *ep,
	bool keepEndpoint) {
	bool hasRemovedEndpoint = false;

	if (keepEndpoint && (ep->numRegions() == 1)) {
	return false;
	}
	ep->setNumRegions(ep->numRegions() - 1);

	LOGFINER("TCCM: decremented region reference count for endpoint %s to %d",
	ep->name().c_str(), ep->numRegions());

	if (0 == ep->numRegions()) {
	// this endpoint no longer used
	m_endpoints.erase(ep->name());
	LOGFINE("delete endpoint %s", ep->name().c_str());
	_GEODE_SAFE_DELETE(ep);
	hasRemovedEndpoint = true;
	}
	return hasRemovedEndpoint;
	}

	int TcrConnectionManager::processEventIdMap(const ACE_Time_Value &currTime,
	const void *) {
	return m_redundancyManager->processEventIdMap(currTime, nullptr);
	}

	int TcrConnectionManager::checkConnection(const ACE_Time_Value &,
	const void *) {
	auto &&guard = m_endpoints.make_lock();
	for (const auto &currItr : m_endpoints) {
	if (currItr.second->connected() && !m_isNetDown) {
	currItr.second->pingServer();
	}
	}
	return 0;
	}

	int TcrConnectionManager::checkRedundancy(const ACE_Time_Value &,
	const void *) {
	m_redundancySema.release();
	return 0;
	}

	void TcrConnectionManager::failover(std::atomic<bool> &isRunning) {
	LOGFINE("TcrConnectionManager: starting failover thread");
	while (isRunning) {
	m_failoverSema.acquire();
	if (isRunning && !m_isNetDown) {
	try {
	std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
	for (const auto &it : m_distMngrs) {
	it->failover();
	}
	while (m_failoverSema.tryacquire() != -1) {
	;
	}
	} catch (const Exception &e) {
	LOGERROR(e.what());
	} catch (const std::exception &e) {
	LOGERROR(e.what());
	} catch (...) {
	LOGERROR(
	"Unexpected exception while failing over to a "
	"different endpoint");
	}
	}
	}
	LOGFINE("TcrConnectionManager: ending failover thread");
	}

	void TcrConnectionManager::getAllEndpoints(
	std::vector<TcrEndpoint *> &endpoints) {
	auto &&guard = m_endpoints.make_lock();
	for (const auto &currItr : m_endpoints) {
	endpoints.push_back(currItr.second.get());
	}
	}

	int32_t TcrConnectionManager::getNumEndPoints() {
	return static_cast<int32_t>(m_endpoints.size());
	}

	GfErrType TcrConnectionManager::registerInterestAllRegions(
	TcrEndpoint ep, const TcrMessage request, TcrMessageReply *reply) {
	// Preconditions:
	// 1. m_distMngrs.size() > 1 (query distribution manager + 1 or more
	// TcrHADistributionManagers).

	GfErrType err = GF_NOERR;
	GfErrType opErr = GF_NOERR;
	std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
	std::list<ThinClientBaseDM *>::iterator begin = m_distMngrs.begin();
	std::list<ThinClientBaseDM *>::iterator end = m_distMngrs.end();
	for (std::list<ThinClientBaseDM *>::iterator it = begin; it != end; ++it) {
	TcrHADistributionManager *tcrHADM =
	dynamic_cast<TcrHADistributionManager >(it);
	if (tcrHADM != nullptr) {
	if ((opErr = tcrHADM->registerInterestForRegion(ep, request, reply)) !=
	GF_NOERR) {
	if (err == GF_NOERR) {
	err = opErr;
	}
	}
	}
	}
	return err;
	}
	GfErrType TcrConnectionManager::sendSyncRequestCq(TcrMessage &request,
	TcrMessageReply &reply) {
	LOGDEBUG("TcrConnectionManager::sendSyncRequestCq");
	GfErrType err = GF_NOERR;
	// Preconditions:
	// 1. m_distMngrs.size() > 1 (query distribution manager + 1 or more
	// TcrHADistributionManagers).

	std::lock_guard<decltype(m_distMngrsLock)> guard(m_distMngrsLock);
	std::list<ThinClientBaseDM *>::iterator begin = m_distMngrs.begin();
	std::list<ThinClientBaseDM *>::iterator end = m_distMngrs.end();
	for (std::list<ThinClientBaseDM *>::iterator it = begin; it != end; ++it) {
	TcrHADistributionManager *tcrHADM =
	dynamic_cast<TcrHADistributionManager >(it);
	if (tcrHADM != nullptr) {
	return tcrHADM->sendSyncRequestCq(request, reply);
	}
	}
	return err;
	}

	void TcrConnectionManager::initializeHAEndpoints(const char *endpointsStr) {
	std::unordered_set<std::string> endpointsList;
	Utils::parseEndpointNamesString(endpointsStr, endpointsList);
	for (std::unordered_set<std::string>::iterator iter = endpointsList.begin();
	iter != endpointsList.end(); ++iter) {
	addRefToTcrEndpoint(*iter);
	}
	// Postconditions:
	// 1. endpointsList.size() > 0
	}

	void TcrConnectionManager::removeHAEndpoints() {
	auto &&guard = m_endpoints.make_lock();
	auto currItr = m_endpoints.begin();
	while (currItr != m_endpoints.end()) {
	if (removeRefToEndpoint(currItr->second.get())) {
	currItr = m_endpoints.begin();
	} else {
	currItr++;
	}
	}
	}

	void TcrConnectionManager::netDown() {
	m_isNetDown = true;

	// sleep for 15 seconds to allow ping and redundancy threads to pause.
	std::this_thread::sleep_for(std::chrono::seconds(15));

	{
	auto &&guard = m_endpoints.make_lock();
	for (auto &currItr : m_endpoints) {
	currItr.second->setConnectionStatus(false);
	}
	}

	m_redundancyManager->netDown();
	}

	/* Need to do a get on unknown key after calling this Fn to restablish all
	* connection */
	void TcrConnectionManager::revive() {
	m_isNetDown = false;

	// sleep for 15 seconds to allow redundancy thread to reestablish
	// connections.
	std::this_thread::sleep_for(std::chrono::seconds(15));
	}

	void TcrConnectionManager::redundancy(std::atomic<bool> &isRunning) {
	LOGFINE("Starting subscription maintain redundancy thread.");
	while (isRunning) {
	m_redundancySema.acquire();
	if (isRunning && !m_isNetDown) {
	m_redundancyManager->maintainRedundancyLevel();
	while (m_redundancySema.tryacquire() != -1) {
	;
	}
	}
	}
	LOGFINE("Ending subscription maintain redundancy thread.");
	}

	void TcrConnectionManager::addNotificationForDeletion(
	Task<TcrEndpoint> notifyReceiver, TcrConnection notifyConnection,
	ACE_Semaphore &notifyCleanupSema) {
	std::lock_guard<decltype(m_notificationLock)> guard(m_notificationLock);
	m_connectionReleaseList.put(notifyConnection);
	m_receiverReleaseList.put(notifyReceiver);
	m_notifyCleanupSemaList.put(&notifyCleanupSema);
	}

	void TcrConnectionManager::cleanup(std::atomic<bool> &isRunning) {
	LOGFINE("TcrConnectionManager: starting cleanup thread");
	do {
	// If we block on acquire, the queue must be empty (precondition).
	if (m_receiverReleaseList.size() == 0) {
	LOGDEBUG(
	"TcrConnectionManager::cleanup(): waiting to acquire cleanup "
	"semaphore.");
	m_cleanupSema.acquire();
	}
	cleanNotificationLists();

	while (m_cleanupSema.tryacquire() != -1) {
	;
	}

	} while (isRunning);

	LOGFINE("TcrConnectionManager: ending cleanup thread");
	// Postcondition - all notification channels should be cleaned up by the end
	// of this function.
	}

	void TcrConnectionManager::cleanNotificationLists() {
	Task<TcrEndpoint> *notifyReceiver;
	TcrConnection *notifyConnection;
	ACE_Semaphore *notifyCleanupSema;

	while (true) {
	{
	std::lock_guard<decltype(m_notificationLock)> guard(m_notificationLock);
	notifyReceiver = m_receiverReleaseList.get();
	if (!notifyReceiver) break;
	notifyConnection = m_connectionReleaseList.get();
	notifyCleanupSema = m_notifyCleanupSemaList.get();
	}
	notifyReceiver->wait();
	//_GEODE_SAFE_DELETE(notifyReceiver);
	_GEODE_SAFE_DELETE(notifyConnection);
	notifyCleanupSema->release();
	}
	}

	void TcrConnectionManager::processMarker() {
	// also set the static bool m_processedMarker for makePrimary messages
	m_redundancyManager->m_globalProcessedMarker = true;
	}

	// TESTING: Durable clients - return queue status of endpoing. Not thread safe.
	bool TcrConnectionManager::getEndpointStatus(const std::string &endpoint) {
	for (auto &currItr : m_endpoints) {
	auto ep = currItr.second;
	const std::string epName = ep->name();
	if (epName == endpoint) return ep->getServerQueueStatusTEST();
	}
	return false;
	}

	GfErrType TcrConnectionManager::sendSyncRequestCq(
	TcrMessage &request, TcrMessageReply &reply,
	TcrHADistributionManager *theHADM) {
	return m_redundancyManager->sendSyncRequestCq(request, reply, theHADM);
	}

	GfErrType TcrConnectionManager::sendSyncRequestRegisterInterest(
	TcrMessage &request, TcrMessageReply &reply, bool attemptFailover,
	TcrEndpoint endpoint, TcrHADistributionManager theHADM,
	ThinClientRegion *region) {
	return m_redundancyManager->sendSyncRequestRegisterInterest(
	request, reply, attemptFailover, endpoint, theHADM, region);
	}
	} // namespace client
	} // namespace geode
	} // namespace apache