foundationdb/fdbserver/TagPartitionedLogSystem.actor.cpp

/*
 * TagPartitionedLogSystem.actor.cpp
 *
 * This source file is part of the FoundationDB open source project
 *
 * Copyright 2013-2018 Apple Inc. and the FoundationDB project authors
 *
 * Licensed 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 "flow/ActorCollection.h"
#include "fdbserver/LogSystem.h"
#include "fdbserver/ServerDBInfo.h"
#include "fdbserver/DBCoreState.h"
#include "fdbserver/WaitFailure.h"
#include "fdbclient/SystemData.h"
#include "fdbrpc/simulator.h"
#include "fdbrpc/Replication.h"
#include "fdbrpc/ReplicationUtils.h"
#include "fdbserver/Knobs.h"
#include "fdbserver/RecoveryState.h"
#include "fdbserver/LogProtocolMessage.h"
#include "flow/actorcompiler.h" // This must be the last #include.

ACTOR Future<Version> minVersionWhenReady(Future<Void> f, std::vector<Future<TLogCommitReply>> replies) {
	wait(f);
	Version minVersion = std::numeric_limits<Version>::max();
	for (auto& reply : replies) {
		if (reply.isReady() && !reply.isError()) {
			minVersion = std::min(minVersion, reply.get().version);
		}
	}
	return minVersion;
}

// TagPartitionedLogSystem info in old epoch
struct OldLogData {
	std::vector<Reference<LogSet>> tLogs;
	int32_t logRouterTags;
	int32_t txsTags; // The number of txsTags, which may change across generations.
	Version epochBegin, epochEnd;
	std::set<int8_t> pseudoLocalities;
	LogEpoch epoch;

	OldLogData() : epochBegin(0), epochEnd(0), logRouterTags(0), txsTags(0), epoch(0) {}

	// Constructor for T of OldTLogConf and OldTLogCoreData
	template <class T>
	explicit OldLogData(const T& conf)
	  : logRouterTags(conf.logRouterTags), txsTags(conf.txsTags), epochBegin(conf.epochBegin), epochEnd(conf.epochEnd),
	    pseudoLocalities(conf.pseudoLocalities), epoch(conf.epoch) {
		tLogs.resize(conf.tLogs.size());
		for (int j = 0; j < conf.tLogs.size(); j++) {
			auto logSet = makeReference<LogSet>(conf.tLogs[j]);
			tLogs[j] = logSet;
		}
	}
};

struct LogLockInfo {
	Version epochEnd;
	bool isCurrent;
	Reference<LogSet> logSet;
	std::vector<Future<TLogLockResult>> replies;

	LogLockInfo() : epochEnd(std::numeric_limits<Version>::max()), isCurrent(false) {}
};

LogSet::LogSet(const TLogSet& tLogSet)
  : tLogWriteAntiQuorum(tLogSet.tLogWriteAntiQuorum), tLogReplicationFactor(tLogSet.tLogReplicationFactor),
    tLogLocalities(tLogSet.tLogLocalities), tLogVersion(tLogSet.tLogVersion), tLogPolicy(tLogSet.tLogPolicy),
    isLocal(tLogSet.isLocal), locality(tLogSet.locality), startVersion(tLogSet.startVersion),
    satelliteTagLocations(tLogSet.satelliteTagLocations) {
	for (const auto& log : tLogSet.tLogs) {
		logServers.push_back(makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(log));
	}
	for (const auto& log : tLogSet.logRouters) {
		logRouters.push_back(makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(log));
	}
	for (const auto& log : tLogSet.backupWorkers) {
		backupWorkers.push_back(makeReference<AsyncVar<OptionalInterface<BackupInterface>>>(log));
	}
	filterLocalityDataForPolicy(tLogPolicy, &tLogLocalities);
	updateLocalitySet(tLogLocalities);
}

LogSet::LogSet(const CoreTLogSet& coreSet)
  : tLogWriteAntiQuorum(coreSet.tLogWriteAntiQuorum), tLogReplicationFactor(coreSet.tLogReplicationFactor),
    tLogLocalities(coreSet.tLogLocalities), tLogVersion(coreSet.tLogVersion), tLogPolicy(coreSet.tLogPolicy),
    isLocal(coreSet.isLocal), locality(coreSet.locality), startVersion(coreSet.startVersion),
    satelliteTagLocations(coreSet.satelliteTagLocations) {
	for (const auto& log : coreSet.tLogs) {
		logServers.push_back(
		    makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(OptionalInterface<TLogInterface>(log)));
	}
	// Do NOT recover coreSet.backupWorkers, because master will recruit new ones.
	filterLocalityDataForPolicy(tLogPolicy, &tLogLocalities);
	updateLocalitySet(tLogLocalities);
}

TLogSet::TLogSet(const LogSet& rhs)
  : tLogWriteAntiQuorum(rhs.tLogWriteAntiQuorum), tLogReplicationFactor(rhs.tLogReplicationFactor),
    tLogLocalities(rhs.tLogLocalities), tLogVersion(rhs.tLogVersion), tLogPolicy(rhs.tLogPolicy), isLocal(rhs.isLocal),
    locality(rhs.locality), startVersion(rhs.startVersion), satelliteTagLocations(rhs.satelliteTagLocations) {
	for (const auto& tlog : rhs.logServers) {
		tLogs.push_back(tlog->get());
	}
	for (const auto& logRouter : rhs.logRouters) {
		logRouters.push_back(logRouter->get());
	}
	for (const auto& worker : rhs.backupWorkers) {
		backupWorkers.push_back(worker->get());
	}
}

OldTLogConf::OldTLogConf(const OldLogData& oldLogData)
  : logRouterTags(oldLogData.logRouterTags), txsTags(oldLogData.txsTags), epochBegin(oldLogData.epochBegin),
    epochEnd(oldLogData.epochEnd), pseudoLocalities(oldLogData.pseudoLocalities), epoch(oldLogData.epoch) {
	for (const Reference<LogSet>& logSet : oldLogData.tLogs) {
		tLogs.emplace_back(*logSet);
	}
}

CoreTLogSet::CoreTLogSet(const LogSet& logset)
  : tLogWriteAntiQuorum(logset.tLogWriteAntiQuorum), tLogReplicationFactor(logset.tLogReplicationFactor),
    tLogLocalities(logset.tLogLocalities), tLogPolicy(logset.tLogPolicy), isLocal(logset.isLocal),
    locality(logset.locality), startVersion(logset.startVersion), satelliteTagLocations(logset.satelliteTagLocations),
    tLogVersion(logset.tLogVersion) {
	for (const auto& log : logset.logServers) {
		tLogs.push_back(log->get().id());
	}
	// Do NOT store logset.backupWorkers, because master will recruit new ones.
}

OldTLogCoreData::OldTLogCoreData(const OldLogData& oldData)
  : logRouterTags(oldData.logRouterTags), txsTags(oldData.txsTags), epochBegin(oldData.epochBegin),
    epochEnd(oldData.epochEnd), pseudoLocalities(oldData.pseudoLocalities), epoch(oldData.epoch) {
	for (const Reference<LogSet>& logSet : oldData.tLogs) {
		if (logSet->logServers.size()) {
			tLogs.emplace_back(*logSet);
		}
	}
}

struct TagPartitionedLogSystem : ILogSystem, ReferenceCounted<TagPartitionedLogSystem> {
	const UID dbgid;
	LogSystemType logSystemType;
	std::vector<Reference<LogSet>> tLogs; // LogSets in different locations: primary, satellite, or remote
	int expectedLogSets;
	int logRouterTags;
	int txsTags;
	UID recruitmentID;
	int repopulateRegionAntiQuorum;
	bool stopped;
	std::set<int8_t> pseudoLocalities; // Represent special localities that will be mapped to tagLocalityLogRouter
	const LogEpoch epoch;
	LogEpoch oldestBackupEpoch;

	// new members
	std::map<Tag, Version> pseudoLocalityPopVersion;
	Future<Void> rejoins;
	Future<Void> recoveryComplete;
	Future<Void> remoteRecovery;
	Future<Void> remoteRecoveryComplete;
	std::vector<LogLockInfo> lockResults;
	AsyncVar<bool> recoveryCompleteWrittenToCoreState;
	bool remoteLogsWrittenToCoreState;
	bool hasRemoteServers;
	AsyncTrigger backupWorkerChanged;
	std::set<UID> removedBackupWorkers; // Workers that are removed before setting them.

	Optional<Version> recoverAt;
	Optional<Version> recoveredAt;
	Version knownCommittedVersion;
	Version backupStartVersion = invalidVersion; // max(tLogs[0].startVersion, previous epochEnd).
	LocalityData locality;
	// For each currently running popFromLog actor, outstandingPops is
	// (logID, tag)->(max popped version, durableKnownCommittedVersion).
	// Why do we need durableKnownCommittedVersion? knownCommittedVersion gives the lower bound of what data
	// will need to be copied into the next generation to restore the replication factor.
	// Guess: It probably serves as a minimum version of what data should be on a TLog in the next generation and
	// sending a pop for anything less than durableKnownCommittedVersion for the TLog will be absurd.
	std::map<std::pair<UID, Tag>, std::pair<Version, Version>> outstandingPops;

	Optional<PromiseStream<Future<Void>>> addActor;
	ActorCollection popActors;
	std::vector<OldLogData> oldLogData; // each element has the log info. in one old epoch.
	AsyncTrigger logSystemConfigChanged;

	TagPartitionedLogSystem(UID dbgid,
	                        LocalityData locality,
	                        LogEpoch e,
	                        Optional<PromiseStream<Future<Void>>> addActor = Optional<PromiseStream<Future<Void>>>())
	  : dbgid(dbgid), logSystemType(LogSystemType::empty), expectedLogSets(0), logRouterTags(0), txsTags(0),
	    repopulateRegionAntiQuorum(0), epoch(e), oldestBackupEpoch(0), recoveryCompleteWrittenToCoreState(false),
	    locality(locality), remoteLogsWrittenToCoreState(false), hasRemoteServers(false), stopped(false),
	    addActor(addActor), popActors(false) {}

	void stopRejoins() final { rejoins = Future<Void>(); }

	void addref() final { ReferenceCounted<TagPartitionedLogSystem>::addref(); }

	void delref() final { ReferenceCounted<TagPartitionedLogSystem>::delref(); }

	std::string describe() const final {
		std::string result;
		for (int i = 0; i < tLogs.size(); i++) {
			result += format("%d: ", i);
			for (int j = 0; j < tLogs[i]->logServers.size(); j++) {
				result += tLogs[i]->logServers[j]->get().id().toString() +
				          ((j == tLogs[i]->logServers.size() - 1) ? " " : ", ");
			}
		}
		return result;
	}

	UID getDebugID() const final { return dbgid; }

	void addPseudoLocality(int8_t locality) {
		ASSERT(locality < 0);
		pseudoLocalities.insert(locality);
		for (uint16_t i = 0; i < logRouterTags; i++) {
			pseudoLocalityPopVersion[Tag(locality, i)] = 0;
		}
	}

	Tag getPseudoPopTag(Tag tag, ProcessClass::ClassType type) const final {
		switch (type) {
		case ProcessClass::LogRouterClass:
			if (tag.locality == tagLocalityLogRouter) {
				ASSERT(pseudoLocalities.count(tagLocalityLogRouterMapped) > 0);
				tag.locality = tagLocalityLogRouterMapped;
			}
			break;

		case ProcessClass::BackupClass:
			if (tag.locality == tagLocalityLogRouter) {
				ASSERT(pseudoLocalities.count(tagLocalityBackup) > 0);
				tag.locality = tagLocalityBackup;
			}
			break;

		default: // This should be an error at caller site.
			break;
		}
		return tag;
	}

	bool hasPseudoLocality(int8_t locality) const final { return pseudoLocalities.count(locality) > 0; }

	// Return the min version of all pseudoLocalities, i.e., logRouter and backupTag
	Version popPseudoLocalityTag(Tag tag, Version upTo) final {
		ASSERT(isPseudoLocality(tag.locality) && hasPseudoLocality(tag.locality));

		Version& localityVersion = pseudoLocalityPopVersion[tag];
		localityVersion = std::max(localityVersion, upTo);
		Version minVersion = localityVersion;
		// Why do we need to use the minimum popped version among all tags? Reason: for example,
		// 2 pseudo tags pop 100 or 150, respectively. It's only safe to pop min(100, 150),
		// because [101,150) is needed by another pseudo tag.
		for (const int8_t locality : pseudoLocalities) {
			minVersion = std::min(minVersion, pseudoLocalityPopVersion[Tag(locality, tag.id)]);
		}
		// TraceEvent("TLogPopPseudoTag", dbgid).detail("Tag", tag.toString()).detail("Version", upTo).detail("PopVersion", minVersion);
		return minVersion;
	}

	static Future<Void> recoverAndEndEpoch(Reference<AsyncVar<Reference<ILogSystem>>> const& outLogSystem,
	                                       UID const& dbgid,
	                                       DBCoreState const& oldState,
	                                       FutureStream<TLogRejoinRequest> const& rejoins,
	                                       LocalityData const& locality,
	                                       bool* forceRecovery) {
		return epochEnd(outLogSystem, dbgid, oldState, rejoins, locality, forceRecovery);
	}

	static Reference<ILogSystem> fromLogSystemConfig(UID const& dbgid,
	                                                 LocalityData const& locality,
	                                                 LogSystemConfig const& lsConf,
	                                                 bool excludeRemote,
	                                                 bool useRecoveredAt,
	                                                 Optional<PromiseStream<Future<Void>>> addActor) {
		ASSERT(lsConf.logSystemType == LogSystemType::tagPartitioned ||
		       (lsConf.logSystemType == LogSystemType::empty && !lsConf.tLogs.size()));
		// ASSERT(lsConf.epoch == epoch);  //< FIXME
		auto logSystem = makeReference<TagPartitionedLogSystem>(dbgid, locality, lsConf.epoch, addActor);

		logSystem->tLogs.reserve(lsConf.tLogs.size());
		logSystem->expectedLogSets = lsConf.expectedLogSets;
		logSystem->logRouterTags = lsConf.logRouterTags;
		logSystem->txsTags = lsConf.txsTags;
		logSystem->recruitmentID = lsConf.recruitmentID;
		logSystem->stopped = lsConf.stopped;
		if (useRecoveredAt) {
			logSystem->recoveredAt = lsConf.recoveredAt;
		}
		logSystem->pseudoLocalities = lsConf.pseudoLocalities;
		for (const TLogSet& tLogSet : lsConf.tLogs) {
			if (!excludeRemote || tLogSet.isLocal) {
				logSystem->tLogs.push_back(makeReference<LogSet>(tLogSet));
			}
		}

		for (const auto& oldTlogConf : lsConf.oldTLogs) {
			logSystem->oldLogData.emplace_back(oldTlogConf);
			//TraceEvent("BWFromLSConf")
			//    .detail("Epoch", logSystem->oldLogData.back().epoch)
			//    .detail("Version", logSystem->oldLogData.back().epochEnd);
		}

		logSystem->logSystemType = lsConf.logSystemType;
		logSystem->oldestBackupEpoch = lsConf.oldestBackupEpoch;
		return logSystem;
	}

	static Reference<ILogSystem> fromOldLogSystemConfig(UID const& dbgid,
	                                                    LocalityData const& locality,
	                                                    LogSystemConfig const& lsConf) {
		ASSERT(lsConf.logSystemType == LogSystemType::tagPartitioned ||
		       (lsConf.logSystemType == LogSystemType::empty && !lsConf.tLogs.size()));
		// ASSERT(lsConf.epoch == epoch);  //< FIXME
		const LogEpoch e = lsConf.oldTLogs.size() > 0 ? lsConf.oldTLogs[0].epoch : 0;
		auto logSystem = makeReference<TagPartitionedLogSystem>(dbgid, locality, e);

		if (lsConf.oldTLogs.size()) {
			for (const TLogSet& tLogSet : lsConf.oldTLogs[0].tLogs) {
				logSystem->tLogs.push_back(makeReference<LogSet>(tLogSet));
			}
			logSystem->logRouterTags = lsConf.oldTLogs[0].logRouterTags;
			logSystem->txsTags = lsConf.oldTLogs[0].txsTags;
			// logSystem->epochEnd = lsConf.oldTLogs[0].epochEnd;

			for (int i = 1; i < lsConf.oldTLogs.size(); i++) {
				logSystem->oldLogData.emplace_back(lsConf.oldTLogs[i]);
			}
		}
		logSystem->logSystemType = lsConf.logSystemType;
		logSystem->stopped = true;
		logSystem->pseudoLocalities = lsConf.pseudoLocalities;

		return logSystem;
	}

	// Convert TagPartitionedLogSystem to DBCoreState and override input newState as return value
	void toCoreState(DBCoreState& newState) final {
		if (recoveryComplete.isValid() && recoveryComplete.isError())
			throw recoveryComplete.getError();

		if (remoteRecoveryComplete.isValid() && remoteRecoveryComplete.isError())
			throw remoteRecoveryComplete.getError();

		newState.tLogs.clear();
		newState.logRouterTags = logRouterTags;
		newState.txsTags = txsTags;
		newState.pseudoLocalities = pseudoLocalities;
		for (const auto& t : tLogs) {
			if (t->logServers.size()) {
				newState.tLogs.emplace_back(*t);
				newState.tLogs.back().tLogLocalities.clear();
				for (const auto& log : t->logServers) {
					newState.tLogs.back().tLogLocalities.push_back(log->get().interf().filteredLocality);
				}
			}
		}

		newState.oldTLogData.clear();
		if (!recoveryComplete.isValid() || !recoveryComplete.isReady() ||
		    (repopulateRegionAntiQuorum == 0 &&
		     (!remoteRecoveryComplete.isValid() || !remoteRecoveryComplete.isReady())) ||
		    epoch != oldestBackupEpoch) {
			for (const auto& oldData : oldLogData) {
				newState.oldTLogData.emplace_back(oldData);
				TraceEvent("BWToCore")
				    .detail("Epoch", newState.oldTLogData.back().epoch)
				    .detail("TotalTags", newState.oldTLogData.back().logRouterTags)
				    .detail("BeginVersion", newState.oldTLogData.back().epochBegin)
				    .detail("EndVersion", newState.oldTLogData.back().epochEnd);
			}
		}

		newState.logSystemType = logSystemType;
	}

	bool remoteStorageRecovered() final { return remoteRecoveryComplete.isValid() && remoteRecoveryComplete.isReady(); }

	Future<Void> onCoreStateChanged() final {
		std::vector<Future<Void>> changes;
		changes.push_back(Never());
		if (recoveryComplete.isValid() && !recoveryComplete.isReady()) {
			changes.push_back(recoveryComplete);
		}
		if (remoteRecovery.isValid() && !remoteRecovery.isReady()) {
			changes.push_back(remoteRecovery);
		}
		if (remoteRecoveryComplete.isValid() && !remoteRecoveryComplete.isReady()) {
			changes.push_back(remoteRecoveryComplete);
		}
		changes.push_back(backupWorkerChanged.onTrigger()); // changes to oldestBackupEpoch
		return waitForAny(changes);
	}

	void coreStateWritten(DBCoreState const& newState) final {
		if (!newState.oldTLogData.size()) {
			recoveryCompleteWrittenToCoreState.set(true);
		}
		for (auto& t : newState.tLogs) {
			if (!t.isLocal) {
				TraceEvent("RemoteLogsWritten", dbgid);
				remoteLogsWrittenToCoreState = true;
				break;
			}
		}
	}

	Future<Void> onError() final { return onError_internal(this); }

	ACTOR static Future<Void> onError_internal(TagPartitionedLogSystem* self) {
		// Never returns normally, but throws an error if the subsystem stops working
		loop {
			std::vector<Future<Void>> failed;
			std::vector<Future<Void>> backupFailed(1, Never());
			std::vector<Future<Void>> changes;

			for (auto& it : self->tLogs) {
				for (auto& t : it->logServers) {
					if (t->get().present()) {
						failed.push_back(waitFailureClient(t->get().interf().waitFailure,
						                                   SERVER_KNOBS->TLOG_TIMEOUT,
						                                   -SERVER_KNOBS->TLOG_TIMEOUT /
						                                       SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
						                                   /*trace=*/true));
					} else {
						changes.push_back(t->onChange());
					}
				}
				for (auto& t : it->logRouters) {
					if (t->get().present()) {
						failed.push_back(waitFailureClient(t->get().interf().waitFailure,
						                                   SERVER_KNOBS->TLOG_TIMEOUT,
						                                   -SERVER_KNOBS->TLOG_TIMEOUT /
						                                       SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
						                                   /*trace=*/true));
					} else {
						changes.push_back(t->onChange());
					}
				}
				for (const auto& worker : it->backupWorkers) {
					if (worker->get().present()) {
						backupFailed.push_back(waitFailureClient(worker->get().interf().waitFailure,
						                                         SERVER_KNOBS->BACKUP_TIMEOUT,
						                                         -SERVER_KNOBS->BACKUP_TIMEOUT /
						                                             SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
						                                         /*trace=*/true));
					} else {
						changes.push_back(worker->onChange());
					}
				}
			}

			if (!self->recoveryCompleteWrittenToCoreState.get()) {
				for (auto& old : self->oldLogData) {
					for (auto& it : old.tLogs) {
						for (auto& t : it->logRouters) {
							if (t->get().present()) {
								failed.push_back(waitFailureClient(t->get().interf().waitFailure,
								                                   SERVER_KNOBS->TLOG_TIMEOUT,
								                                   -SERVER_KNOBS->TLOG_TIMEOUT /
								                                       SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
								                                   /*trace=*/true));
							} else {
								changes.push_back(t->onChange());
							}
						}
					}
					// Monitor changes of backup workers for old epochs.
					for (const auto& worker : old.tLogs[0]->backupWorkers) {
						if (worker->get().present()) {
							backupFailed.push_back(waitFailureClient(worker->get().interf().waitFailure,
							                                         SERVER_KNOBS->BACKUP_TIMEOUT,
							                                         -SERVER_KNOBS->BACKUP_TIMEOUT /
							                                             SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
							                                         /*trace=*/true));
						} else {
							changes.push_back(worker->onChange());
						}
					}
				}
			}

			if (self->hasRemoteServers && (!self->remoteRecovery.isReady() || self->remoteRecovery.isError())) {
				changes.push_back(self->remoteRecovery);
			}

			changes.push_back(self->recoveryCompleteWrittenToCoreState.onChange());
			changes.push_back(self->backupWorkerChanged.onTrigger());

			ASSERT(failed.size() >= 1);
			wait(quorum(changes, 1) || tagError<Void>(quorum(failed, 1), master_tlog_failed()) ||
			     tagError<Void>(quorum(backupFailed, 1), master_backup_worker_failed()));
		}
	}

	ACTOR static Future<Void> pushResetChecker(Reference<ConnectionResetInfo> self, NetworkAddress addr) {
		self->slowReplies = 0;
		self->fastReplies = 0;
		wait(delay(SERVER_KNOBS->PUSH_STATS_INTERVAL));
		TraceEvent("SlowPushStats")
		    .detail("PeerAddress", addr)
		    .detail("SlowReplies", self->slowReplies)
		    .detail("FastReplies", self->fastReplies);
		if (self->slowReplies >= SERVER_KNOBS->PUSH_STATS_SLOW_AMOUNT &&
		    self->slowReplies / double(self->slowReplies + self->fastReplies) >= SERVER_KNOBS->PUSH_STATS_SLOW_RATIO) {
			FlowTransport::transport().resetConnection(addr);
			self->lastReset = now();
		}
		return Void();
	}

	ACTOR static Future<TLogCommitReply> recordPushMetrics(Reference<ConnectionResetInfo> self,
	                                                       NetworkAddress addr,
	                                                       Future<TLogCommitReply> in) {
		state double startTime = now();
		TLogCommitReply t = wait(in);
		if (now() - self->lastReset > SERVER_KNOBS->PUSH_RESET_INTERVAL) {
			if (now() - startTime > SERVER_KNOBS->PUSH_MAX_LATENCY) {
				if (self->resetCheck.isReady()) {
					self->resetCheck = pushResetChecker(self, addr);
				}
				self->slowReplies++;
			} else {
				self->fastReplies++;
			}
		}
		return t;
	}

	Future<Version> push(Version prevVersion,
	                     Version version,
	                     Version knownCommittedVersion,
	                     Version minKnownCommittedVersion,
	                     LogPushData& data,
	                     SpanID const& spanContext,
	                     Optional<UID> debugID) final {
		// FIXME: Randomize request order as in LegacyLogSystem?
		vector<Future<Void>> quorumResults;
		vector<Future<TLogCommitReply>> allReplies;
		int location = 0;
		Span span("TPLS:push"_loc, spanContext);
		for (auto& it : tLogs) {
			if (it->isLocal && it->logServers.size()) {
				if (it->connectionResetTrackers.size() == 0) {
					for (int i = 0; i < it->logServers.size(); i++) {
						it->connectionResetTrackers.push_back(makeReference<ConnectionResetInfo>());
					}
				}
				vector<Future<Void>> tLogCommitResults;
				for (int loc = 0; loc < it->logServers.size(); loc++) {
					Standalone<StringRef> msg = data.getMessages(location);
					allReplies.push_back(recordPushMetrics(
					    it->connectionResetTrackers[loc],
					    it->logServers[loc]->get().interf().address(),
					    it->logServers[loc]->get().interf().commit.getReply(TLogCommitRequest(spanContext,
					                                                                          msg.arena(),
					                                                                          prevVersion,
					                                                                          version,
					                                                                          knownCommittedVersion,
					                                                                          minKnownCommittedVersion,
					                                                                          msg,
					                                                                          debugID),
					                                                        TaskPriority::ProxyTLogCommitReply)));
					Future<Void> commitSuccess = success(allReplies.back());
					addActor.get().send(commitSuccess);
					tLogCommitResults.push_back(commitSuccess);
					location++;
				}
				quorumResults.push_back(quorum(tLogCommitResults, tLogCommitResults.size() - it->tLogWriteAntiQuorum));
			}
		}

		return minVersionWhenReady(waitForAll(quorumResults), allReplies);
	}

	Reference<IPeekCursor> peekAll(UID dbgid, Version begin, Version end, Tag tag, bool parallelGetMore) {
		int bestSet = 0;
		std::vector<Reference<LogSet>> localSets;
		Version lastBegin = 0;
		bool foundSpecial = false;
		for (auto& log : tLogs) {
			if (log->locality == tagLocalitySpecial || log->locality == tagLocalityUpgraded) {
				foundSpecial = true;
			}
			if (log->isLocal && log->logServers.size() &&
			    (log->locality == tagLocalitySpecial || log->locality == tagLocalityUpgraded ||
			     log->locality == tag.locality || tag == txsTag || tag.locality == tagLocalityTxs ||
			     tag.locality == tagLocalityLogRouter ||
			     ((tag.locality == tagLocalityUpgraded || tag == cacheTag) && log->locality != tagLocalitySatellite))) {
				lastBegin = std::max(lastBegin, log->startVersion);
				localSets.push_back(log);
				if (log->locality != tagLocalitySatellite) {
					bestSet = localSets.size() - 1;
				}
			}
		}

		if (!localSets.size()) {
			lastBegin = end;
		}

		if (begin >= lastBegin && localSets.size()) {
			TraceEvent("TLogPeekAllCurrentOnly", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end)
			    .detail("BestLogs", localSets[bestSet]->logServerString());
			return makeReference<ILogSystem::SetPeekCursor>(
			    localSets, bestSet, localSets[bestSet]->bestLocationFor(tag), tag, begin, end, parallelGetMore);
		} else {
			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			std::vector<LogMessageVersion> epochEnds;

			if (lastBegin < end && localSets.size()) {
				TraceEvent("TLogPeekAllAddingCurrent", dbgid)
				    .detail("Tag", tag.toString())
				    .detail("Begin", begin)
				    .detail("End", end)
				    .detail("BestLogs", localSets[bestSet]->logServerString());
				cursors.push_back(makeReference<ILogSystem::SetPeekCursor>(localSets,
				                                                           bestSet,
				                                                           localSets[bestSet]->bestLocationFor(tag),
				                                                           tag,
				                                                           lastBegin,
				                                                           end,
				                                                           parallelGetMore));
			}
			for (int i = 0; begin < lastBegin; i++) {
				if (i == oldLogData.size()) {
					if (tag == txsTag || tag.locality == tagLocalityTxs || tag == cacheTag) {
						break;
					}
					TraceEvent("TLogPeekAllDead", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end)
					    .detail("LastBegin", lastBegin)
					    .detail("OldLogDataSize", oldLogData.size());
					return makeReference<ILogSystem::ServerPeekCursor>(
					    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(),
					    tag,
					    begin,
					    getPeekEnd(),
					    false,
					    false);
				}

				int bestOldSet = 0;
				std::vector<Reference<LogSet>> localOldSets;
				Version thisBegin = begin;
				bool thisSpecial = false;
				for (auto& log : oldLogData[i].tLogs) {
					if (log->locality == tagLocalitySpecial || log->locality == tagLocalityUpgraded) {
						thisSpecial = true;
					}
					if (log->isLocal && log->logServers.size() &&
					    (log->locality == tagLocalitySpecial || log->locality == tagLocalityUpgraded ||
					     log->locality == tag.locality || tag == txsTag || tag.locality == tagLocalityTxs ||
					     tag.locality == tagLocalityLogRouter ||
					     ((tag.locality == tagLocalityUpgraded || tag == cacheTag) &&
					      log->locality != tagLocalitySatellite))) {
						thisBegin = std::max(thisBegin, log->startVersion);
						localOldSets.push_back(log);
						if (log->locality != tagLocalitySatellite) {
							bestOldSet = localOldSets.size() - 1;
						}
					}
				}

				if (!localOldSets.size()) {
					TraceEvent("TLogPeekAllNoLocalSets", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end)
					    .detail("LastBegin", lastBegin);
					if (!cursors.size() && !foundSpecial) {
						continue;
					}
					return makeReference<ILogSystem::ServerPeekCursor>(
					    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(),
					    tag,
					    begin,
					    getPeekEnd(),
					    false,
					    false);
				}
				if (thisSpecial) {
					foundSpecial = true;
				}

				if (thisBegin < lastBegin) {
					if (thisBegin < end) {
						TraceEvent("TLogPeekAllAddingOld", dbgid)
						    .detail("Tag", tag.toString())
						    .detail("Begin", begin)
						    .detail("End", end)
						    .detail("BestLogs", localOldSets[bestOldSet]->logServerString())
						    .detail("LastBegin", lastBegin)
						    .detail("ThisBegin", thisBegin);
						cursors.push_back(
						    makeReference<ILogSystem::SetPeekCursor>(localOldSets,
						                                             bestOldSet,
						                                             localOldSets[bestOldSet]->bestLocationFor(tag),
						                                             tag,
						                                             thisBegin,
						                                             std::min(lastBegin, end),
						                                             parallelGetMore));
						epochEnds.push_back(LogMessageVersion(std::min(lastBegin, end)));
					}
					lastBegin = thisBegin;
				}
			}

			return makeReference<ILogSystem::MultiCursor>(cursors, epochEnds);
		}
	}

	Reference<IPeekCursor> peekRemote(UID dbgid, Version begin, Optional<Version> end, Tag tag, bool parallelGetMore) {
		int bestSet = -1;
		Version lastBegin = recoveredAt.present() ? recoveredAt.get() + 1 : 0;
		for (int t = 0; t < tLogs.size(); t++) {
			if (tLogs[t]->isLocal) {
				lastBegin = std::max(lastBegin, tLogs[t]->startVersion);
			}

			if (tLogs[t]->logRouters.size()) {
				ASSERT(bestSet == -1);
				bestSet = t;
			}
		}
		if (bestSet == -1) {
			TraceEvent("TLogPeekRemoteNoBestSet", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end.present() ? end.get() : getPeekEnd());
			return makeReference<ILogSystem::ServerPeekCursor>(Reference<AsyncVar<OptionalInterface<TLogInterface>>>(),
			                                                   tag,
			                                                   begin,
			                                                   getPeekEnd(),
			                                                   false,
			                                                   parallelGetMore);
		}
		if (begin >= lastBegin) {
			TraceEvent("TLogPeekRemoteBestOnly", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end.present() ? end.get() : getPeekEnd())
			    .detail("BestSet", bestSet)
			    .detail("BestSetStart", lastBegin)
			    .detail("LogRouterIds", tLogs[bestSet]->logRouterString());
			return makeReference<ILogSystem::BufferedCursor>(
			    tLogs[bestSet]->logRouters, tag, begin, end.present() ? end.get() + 1 : getPeekEnd(), parallelGetMore);
		} else {
			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			std::vector<LogMessageVersion> epochEnds;
			TraceEvent("TLogPeekRemoteAddingBest", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end.present() ? end.get() : getPeekEnd())
			    .detail("BestSet", bestSet)
			    .detail("BestSetStart", lastBegin)
			    .detail("LogRouterIds", tLogs[bestSet]->logRouterString());
			cursors.push_back(makeReference<ILogSystem::BufferedCursor>(tLogs[bestSet]->logRouters,
			                                                            tag,
			                                                            lastBegin,
			                                                            end.present() ? end.get() + 1 : getPeekEnd(),
			                                                            parallelGetMore));
			int i = 0;
			while (begin < lastBegin) {
				if (i == oldLogData.size()) {
					TraceEvent("TLogPeekRemoteDead", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end.present() ? end.get() : getPeekEnd())
					    .detail("LastBegin", lastBegin)
					    .detail("OldLogDataSize", oldLogData.size());
					return makeReference<ILogSystem::ServerPeekCursor>(
					    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(),
					    tag,
					    begin,
					    getPeekEnd(),
					    false,
					    parallelGetMore);
				}

				int bestOldSet = -1;
				Version thisBegin = begin;
				for (int t = 0; t < oldLogData[i].tLogs.size(); t++) {
					if (oldLogData[i].tLogs[t]->isLocal) {
						thisBegin = std::max(thisBegin, oldLogData[i].tLogs[t]->startVersion);
					}

					if (oldLogData[i].tLogs[t]->logRouters.size()) {
						ASSERT(bestOldSet == -1);
						bestOldSet = t;
					}
				}
				if (bestOldSet == -1) {
					TraceEvent("TLogPeekRemoteNoOldBestSet", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end.present() ? end.get() : getPeekEnd());
					return makeReference<ILogSystem::ServerPeekCursor>(
					    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(),
					    tag,
					    begin,
					    getPeekEnd(),
					    false,
					    parallelGetMore);
				}

				if (thisBegin < lastBegin) {
					TraceEvent("TLogPeekRemoteAddingOldBest", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end.present() ? end.get() : getPeekEnd())
					    .detail("BestOldSet", bestOldSet)
					    .detail("LogRouterIds", oldLogData[i].tLogs[bestOldSet]->logRouterString())
					    .detail("LastBegin", lastBegin)
					    .detail("ThisBegin", thisBegin)
					    .detail("BestStartVer", oldLogData[i].tLogs[bestOldSet]->startVersion);
					cursors.push_back(makeReference<ILogSystem::BufferedCursor>(
					    oldLogData[i].tLogs[bestOldSet]->logRouters, tag, thisBegin, lastBegin, parallelGetMore));
					epochEnds.emplace_back(lastBegin);
					lastBegin = thisBegin;
				}
				i++;
			}

			return makeReference<ILogSystem::MultiCursor>(cursors, epochEnds);
		}
	}

	Reference<IPeekCursor> peek(UID dbgid, Version begin, Optional<Version> end, Tag tag, bool parallelGetMore) final {
		if (!tLogs.size()) {
			TraceEvent("TLogPeekNoLogSets", dbgid).detail("Tag", tag.toString()).detail("Begin", begin);
			return makeReference<ILogSystem::ServerPeekCursor>(
			    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(), tag, begin, getPeekEnd(), false, false);
		}

		if (tag.locality == tagLocalityRemoteLog) {
			return peekRemote(dbgid, begin, end, tag, parallelGetMore);
		} else {
			return peekAll(dbgid, begin, getPeekEnd(), tag, parallelGetMore);
		}
	}

	Reference<IPeekCursor> peek(UID dbgid,
	                            Version begin,
	                            Optional<Version> end,
	                            std::vector<Tag> tags,
	                            bool parallelGetMore) final {
		if (tags.empty()) {
			TraceEvent("TLogPeekNoTags", dbgid).detail("Begin", begin);
			return makeReference<ILogSystem::ServerPeekCursor>(
			    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(), invalidTag, begin, getPeekEnd(), false, false);
		}

		if (tags.size() == 1) {
			return peek(dbgid, begin, end, tags[0], parallelGetMore);
		}

		std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
		cursors.reserve(tags.size());
		for (auto tag : tags) {
			cursors.push_back(peek(dbgid, begin, end, tag, parallelGetMore));
		}
		return makeReference<ILogSystem::BufferedCursor>(cursors,
		                                                 begin,
		                                                 end.present() ? end.get() + 1 : getPeekEnd(),
		                                                 true,
		                                                 tLogs[0]->locality == tagLocalityUpgraded,
		                                                 false);
	}

	Reference<IPeekCursor> peekLocal(UID dbgid,
	                                 Tag tag,
	                                 Version begin,
	                                 Version end,
	                                 bool useMergePeekCursors,
	                                 int8_t peekLocality = tagLocalityInvalid) {
		if (tag.locality >= 0 || tag.locality == tagLocalityUpgraded || tag.locality == tagLocalitySpecial) {
			peekLocality = tag.locality;
		}
		ASSERT(peekLocality >= 0 || peekLocality == tagLocalityUpgraded || tag.locality == tagLocalitySpecial);

		int bestSet = -1;
		bool foundSpecial = false;
		int logCount = 0;
		for (int t = 0; t < tLogs.size(); t++) {
			if (tLogs[t]->logServers.size() && tLogs[t]->locality != tagLocalitySatellite) {
				logCount++;
			}
			if (tLogs[t]->logServers.size() &&
			    (tLogs[t]->locality == tagLocalitySpecial || tLogs[t]->locality == tagLocalityUpgraded ||
			     tLogs[t]->locality == peekLocality || peekLocality == tagLocalityUpgraded ||
			     peekLocality == tagLocalitySpecial)) {
				if (tLogs[t]->locality == tagLocalitySpecial || tLogs[t]->locality == tagLocalityUpgraded) {
					foundSpecial = true;
				}
				bestSet = t;
				break;
			}
		}
		if (bestSet == -1) {
			TraceEvent("TLogPeekLocalNoBestSet", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end)
			    .detail("LogCount", logCount);
			if (useMergePeekCursors || logCount > 1) {
				throw worker_removed();
			} else {
				return makeReference<ILogSystem::ServerPeekCursor>(
				    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(), tag, begin, getPeekEnd(), false, false);
			}
		}

		if (begin >= tLogs[bestSet]->startVersion) {
			TraceEvent("TLogPeekLocalBestOnly", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", begin)
			    .detail("End", end)
			    .detail("BestSet", bestSet)
			    .detail("BestSetStart", tLogs[bestSet]->startVersion)
			    .detail("LogId", tLogs[bestSet]->logServers[tLogs[bestSet]->bestLocationFor(tag)]->get().id());
			if (useMergePeekCursors) {
				return makeReference<ILogSystem::MergedPeekCursor>(tLogs[bestSet]->logServers,
				                                                   tLogs[bestSet]->bestLocationFor(tag),
				                                                   tLogs[bestSet]->logServers.size() + 1 -
				                                                       tLogs[bestSet]->tLogReplicationFactor,
				                                                   tag,
				                                                   begin,
				                                                   end,
				                                                   true,
				                                                   tLogs[bestSet]->tLogLocalities,
				                                                   tLogs[bestSet]->tLogPolicy,
				                                                   tLogs[bestSet]->tLogReplicationFactor);
			} else {
				return makeReference<ILogSystem::ServerPeekCursor>(
				    tLogs[bestSet]->logServers[tLogs[bestSet]->bestLocationFor(tag)], tag, begin, end, false, false);
			}
		} else {
			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			std::vector<LogMessageVersion> epochEnds;

			if (tLogs[bestSet]->startVersion < end) {
				TraceEvent("TLogPeekLocalAddingBest", dbgid)
				    .detail("Tag", tag.toString())
				    .detail("Begin", begin)
				    .detail("End", end)
				    .detail("BestSet", bestSet)
				    .detail("BestSetStart", tLogs[bestSet]->startVersion)
				    .detail("LogId", tLogs[bestSet]->logServers[tLogs[bestSet]->bestLocationFor(tag)]->get().id());
				if (useMergePeekCursors) {
					cursors.push_back(makeReference<ILogSystem::MergedPeekCursor>(
					    tLogs[bestSet]->logServers,
					    tLogs[bestSet]->bestLocationFor(tag),
					    tLogs[bestSet]->logServers.size() + 1 - tLogs[bestSet]->tLogReplicationFactor,
					    tag,
					    tLogs[bestSet]->startVersion,
					    end,
					    true,
					    tLogs[bestSet]->tLogLocalities,
					    tLogs[bestSet]->tLogPolicy,
					    tLogs[bestSet]->tLogReplicationFactor));
				} else {
					cursors.push_back(makeReference<ILogSystem::ServerPeekCursor>(
					    tLogs[bestSet]->logServers[tLogs[bestSet]->bestLocationFor(tag)],
					    tag,
					    tLogs[bestSet]->startVersion,
					    end,
					    false,
					    false));
				}
			}
			Version lastBegin = tLogs[bestSet]->startVersion;
			for (int i = 0; begin < lastBegin; i++) {
				if (i == oldLogData.size()) {
					if ((tag == txsTag || tag.locality == tagLocalityTxs) && cursors.size()) {
						break;
					}
					TraceEvent("TLogPeekLocalDead", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end)
					    .detail("LastBegin", lastBegin)
					    .detail("OldLogDataSize", oldLogData.size());
					throw worker_removed();
				}

				int bestOldSet = -1;
				logCount = 0;
				bool nextFoundSpecial = false;
				for (int t = 0; t < oldLogData[i].tLogs.size(); t++) {
					if (oldLogData[i].tLogs[t]->logServers.size() &&
					    oldLogData[i].tLogs[t]->locality != tagLocalitySatellite) {
						logCount++;
					}
					if (oldLogData[i].tLogs[t]->logServers.size() &&
					    (oldLogData[i].tLogs[t]->locality == tagLocalitySpecial ||
					     oldLogData[i].tLogs[t]->locality == tagLocalityUpgraded ||
					     oldLogData[i].tLogs[t]->locality == peekLocality || peekLocality == tagLocalityUpgraded ||
					     peekLocality == tagLocalitySpecial)) {
						if (oldLogData[i].tLogs[t]->locality == tagLocalitySpecial ||
						    oldLogData[i].tLogs[t]->locality == tagLocalityUpgraded) {
							nextFoundSpecial = true;
						}
						if (foundSpecial && !oldLogData[i].tLogs[t]->isLocal) {
							TraceEvent("TLogPeekLocalRemoteBeforeSpecial", dbgid)
							    .detail("Tag", tag.toString())
							    .detail("Begin", begin)
							    .detail("End", end)
							    .detail("LastBegin", lastBegin)
							    .detail("OldLogDataSize", oldLogData.size())
							    .detail("Idx", i);
							throw worker_removed();
						}
						bestOldSet = t;
						break;
					}
				}

				if (bestOldSet == -1) {
					TraceEvent("TLogPeekLocalNoBestSet", dbgid)
					    .detail("Tag", tag.toString())
					    .detail("Begin", begin)
					    .detail("End", end)
					    .detail("LastBegin", lastBegin)
					    .detail("OldLogDataSize", oldLogData.size())
					    .detail("Idx", i)
					    .detail("LogRouterTags", oldLogData[i].logRouterTags)
					    .detail("LogCount", logCount)
					    .detail("FoundSpecial", foundSpecial);
					if (oldLogData[i].logRouterTags == 0 || logCount > 1 || foundSpecial) {
						throw worker_removed();
					}
					continue;
				}

				foundSpecial = nextFoundSpecial;

				Version thisBegin = std::max(oldLogData[i].tLogs[bestOldSet]->startVersion, begin);
				if (thisBegin < lastBegin) {
					if (thisBegin < end) {
						TraceEvent("TLogPeekLocalAddingOldBest", dbgid)
						    .detail("Tag", tag.toString())
						    .detail("Begin", begin)
						    .detail("End", end)
						    .detail("BestOldSet", bestOldSet)
						    .detail("LogServers", oldLogData[i].tLogs[bestOldSet]->logServerString())
						    .detail("ThisBegin", thisBegin)
						    .detail("LastBegin", lastBegin);
						// detail("LogId",
						// oldLogData[i].tLogs[bestOldSet]->logServers[tLogs[bestOldSet]->bestLocationFor( tag
						// )]->get().id());
						cursors.push_back(makeReference<ILogSystem::MergedPeekCursor>(
						    oldLogData[i].tLogs[bestOldSet]->logServers,
						    oldLogData[i].tLogs[bestOldSet]->bestLocationFor(tag),
						    oldLogData[i].tLogs[bestOldSet]->logServers.size() + 1 -
						        oldLogData[i].tLogs[bestOldSet]->tLogReplicationFactor,
						    tag,
						    thisBegin,
						    std::min(lastBegin, end),
						    useMergePeekCursors,
						    oldLogData[i].tLogs[bestOldSet]->tLogLocalities,
						    oldLogData[i].tLogs[bestOldSet]->tLogPolicy,
						    oldLogData[i].tLogs[bestOldSet]->tLogReplicationFactor));
						epochEnds.emplace_back(std::min(lastBegin, end));
					}
					lastBegin = thisBegin;
				}
			}

			return makeReference<ILogSystem::MultiCursor>(cursors, epochEnds);
		}
	}

	Reference<IPeekCursor> peekTxs(UID dbgid,
	                               Version begin,
	                               int8_t peekLocality,
	                               Version localEnd,
	                               bool canDiscardPopped) final {
		Version end = getEnd();
		if (!tLogs.size()) {
			TraceEvent("TLogPeekTxsNoLogs", dbgid);
			return makeReference<ILogSystem::ServerPeekCursor>(
			    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(), txsTag, begin, end, false, false);
		}
		TraceEvent("TLogPeekTxs", dbgid)
		    .detail("Begin", begin)
		    .detail("End", end)
		    .detail("LocalEnd", localEnd)
		    .detail("PeekLocality", peekLocality)
		    .detail("CanDiscardPopped", canDiscardPopped);

		int maxTxsTags = txsTags;
		bool needsOldTxs = tLogs[0]->tLogVersion < TLogVersion::V4;
		for (auto& it : oldLogData) {
			maxTxsTags = std::max<int>(maxTxsTags, it.txsTags);
			needsOldTxs = needsOldTxs || it.tLogs[0]->tLogVersion < TLogVersion::V4;
		}

		if (peekLocality < 0 || localEnd == invalidVersion || localEnd <= begin) {
			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			cursors.reserve(maxTxsTags);
			for (int i = 0; i < maxTxsTags; i++) {
				cursors.push_back(peekAll(dbgid, begin, end, Tag(tagLocalityTxs, i), true));
			}
			// SOMEDAY: remove once upgrades from 6.2 are no longer supported
			if (needsOldTxs) {
				cursors.push_back(peekAll(dbgid, begin, end, txsTag, true));
			}

			return makeReference<ILogSystem::BufferedCursor>(cursors, begin, end, false, false, canDiscardPopped);
		}

		try {
			if (localEnd >= end) {
				std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
				cursors.reserve(maxTxsTags);
				for (int i = 0; i < maxTxsTags; i++) {
					cursors.push_back(peekLocal(dbgid, Tag(tagLocalityTxs, i), begin, end, true, peekLocality));
				}
				// SOMEDAY: remove once upgrades from 6.2 are no longer supported
				if (needsOldTxs) {
					cursors.push_back(peekLocal(dbgid, txsTag, begin, end, true, peekLocality));
				}

				return makeReference<ILogSystem::BufferedCursor>(cursors, begin, end, false, false, canDiscardPopped);
			}

			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			std::vector<LogMessageVersion> epochEnds;

			cursors.resize(2);

			std::vector<Reference<ILogSystem::IPeekCursor>> localCursors;
			std::vector<Reference<ILogSystem::IPeekCursor>> allCursors;
			for (int i = 0; i < maxTxsTags; i++) {
				localCursors.push_back(peekLocal(dbgid, Tag(tagLocalityTxs, i), begin, localEnd, true, peekLocality));
				allCursors.push_back(peekAll(dbgid, localEnd, end, Tag(tagLocalityTxs, i), true));
			}
			// SOMEDAY: remove once upgrades from 6.2 are no longer supported
			if (needsOldTxs) {
				localCursors.push_back(peekLocal(dbgid, txsTag, begin, localEnd, true, peekLocality));
				allCursors.push_back(peekAll(dbgid, localEnd, end, txsTag, true));
			}

			cursors[1] = makeReference<ILogSystem::BufferedCursor>(
			    localCursors, begin, localEnd, false, false, canDiscardPopped);
			cursors[0] = makeReference<ILogSystem::BufferedCursor>(allCursors, localEnd, end, false, false, false);
			epochEnds.emplace_back(localEnd);

			return makeReference<ILogSystem::MultiCursor>(cursors, epochEnds);
		} catch (Error& e) {
			if (e.code() == error_code_worker_removed) {
				std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
				cursors.reserve(maxTxsTags);
				for (int i = 0; i < maxTxsTags; i++) {
					cursors.push_back(peekAll(dbgid, begin, end, Tag(tagLocalityTxs, i), true));
				}
				// SOMEDAY: remove once upgrades from 6.2 are no longer supported
				if (needsOldTxs) {
					cursors.push_back(peekAll(dbgid, begin, end, txsTag, true));
				}

				return makeReference<ILogSystem::BufferedCursor>(cursors, begin, end, false, false, canDiscardPopped);
			}
			throw;
		}
	}

	Reference<IPeekCursor> peekSingle(UID dbgid,
	                                  Version begin,
	                                  Tag tag,
	                                  std::vector<std::pair<Version, Tag>> history) final {
		while (history.size() && begin >= history.back().first) {
			history.pop_back();
		}

		if (history.size() == 0) {
			TraceEvent("TLogPeekSingleNoHistory", dbgid).detail("Tag", tag.toString()).detail("Begin", begin);
			return peekLocal(dbgid, tag, begin, getPeekEnd(), false);
		} else {
			std::vector<Reference<ILogSystem::IPeekCursor>> cursors;
			std::vector<LogMessageVersion> epochEnds;

			TraceEvent("TLogPeekSingleAddingLocal", dbgid)
			    .detail("Tag", tag.toString())
			    .detail("Begin", history[0].first);
			cursors.push_back(peekLocal(dbgid, tag, history[0].first, getPeekEnd(), false));

			for (int i = 0; i < history.size(); i++) {
				TraceEvent("TLogPeekSingleAddingOld", dbgid)
				    .detail("Tag", tag.toString())
				    .detail("HistoryTag", history[i].second.toString())
				    .detail("Begin", i + 1 == history.size() ? begin : std::max(history[i + 1].first, begin))
				    .detail("End", history[i].first);
				cursors.push_back(peekLocal(dbgid,
				                            history[i].second,
				                            i + 1 == history.size() ? begin : std::max(history[i + 1].first, begin),
				                            history[i].first,
				                            false));
				epochEnds.emplace_back(history[i].first);
			}

			return makeReference<ILogSystem::MultiCursor>(cursors, epochEnds);
		}
	}

	// LogRouter or BackupWorker use this function to obtain a cursor for peeking tlogs of a generation (i.e., epoch).
	// Specifically, the epoch is determined by looking up "dbgid" in tlog sets of generations.
	// The returned cursor can peek data at the "tag" from the given "begin" version to that epoch's end version or
	// the recovery version for the latest old epoch. For the current epoch, the cursor has no end version.
	Reference<IPeekCursor> peekLogRouter(UID dbgid, Version begin, Tag tag) final {
		bool found = false;
		for (const auto& log : tLogs) {
			found = log->hasLogRouter(dbgid) || log->hasBackupWorker(dbgid);
			if (found) {
				break;
			}
		}
		if (found) {
			if (stopped) {
				std::vector<Reference<LogSet>> localSets;
				int bestPrimarySet = 0;
				int bestSatelliteSet = -1;
				for (auto& log : tLogs) {
					if (log->isLocal && log->logServers.size()) {
						TraceEvent("TLogPeekLogRouterLocalSet", dbgid)
						    .detail("Tag", tag.toString())
						    .detail("Begin", begin)
						    .detail("LogServers", log->logServerString());
						localSets.push_back(log);
						if (log->locality == tagLocalitySatellite) {
							bestSatelliteSet = localSets.size() - 1;
						} else {
							bestPrimarySet = localSets.size() - 1;
						}
					}
				}
				int bestSet = bestPrimarySet;
				if (SERVER_KNOBS->LOG_ROUTER_PEEK_FROM_SATELLITES_PREFERRED && bestSatelliteSet != -1 &&
				    tLogs[bestSatelliteSet]->tLogVersion >= TLogVersion::V4) {
					bestSet = bestSatelliteSet;
				}

				TraceEvent("TLogPeekLogRouterSets", dbgid).detail("Tag", tag.toString()).detail("Begin", begin);
				// FIXME: do this merge on one of the logs in the other data center to avoid sending multiple copies
				// across the WAN
				return makeReference<ILogSystem::SetPeekCursor>(
				    localSets, bestSet, localSets[bestSet]->bestLocationFor(tag), tag, begin, getPeekEnd(), true);
			} else {
				int bestPrimarySet = -1;
				int bestSatelliteSet = -1;
				for (int i = 0; i < tLogs.size(); i++) {
					const auto& log = tLogs[i];
					if (log->logServers.size() && log->isLocal) {
						if (log->locality == tagLocalitySatellite) {
							bestSatelliteSet = i;
							break;
						} else {
							if (bestPrimarySet == -1)
								bestPrimarySet = i;
						}
					}
				}
				int bestSet = bestPrimarySet;
				if (SERVER_KNOBS->LOG_ROUTER_PEEK_FROM_SATELLITES_PREFERRED && bestSatelliteSet != -1 &&
				    tLogs[bestSatelliteSet]->tLogVersion >= TLogVersion::V4) {
					bestSet = bestSatelliteSet;
				}
				const auto& log = tLogs[bestSet];
				TraceEvent("TLogPeekLogRouterBestOnly", dbgid)
				    .detail("Tag", tag.toString())
				    .detail("Begin", begin)
				    .detail("LogId", log->logServers[log->bestLocationFor(tag)]->get().id());
				return makeReference<ILogSystem::ServerPeekCursor>(
				    log->logServers[log->bestLocationFor(tag)], tag, begin, getPeekEnd(), false, true);
			}
		}
		bool firstOld = true;
		for (const auto& old : oldLogData) {
			found = false;
			for (const auto& log : old.tLogs) {
				found = log->hasLogRouter(dbgid) || log->hasBackupWorker(dbgid);
				if (found) {
					break;
				}
			}
			if (found) {
				int bestPrimarySet = 0;
				int bestSatelliteSet = -1;
				std::vector<Reference<LogSet>> localSets;
				for (auto& log : old.tLogs) {
					if (log->isLocal && log->logServers.size()) {
						TraceEvent("TLogPeekLogRouterOldLocalSet", dbgid)
						    .detail("Tag", tag.toString())
						    .detail("Begin", begin)
						    .detail("LogServers", log->logServerString());
						localSets.push_back(log);
						if (log->locality == tagLocalitySatellite) {
							bestSatelliteSet = localSets.size() - 1;
						} else {
							bestPrimarySet = localSets.size() - 1;
						}
					}
				}
				int bestSet = bestPrimarySet;
				if (SERVER_KNOBS->LOG_ROUTER_PEEK_FROM_SATELLITES_PREFERRED && bestSatelliteSet != -1 &&
				    old.tLogs[bestSatelliteSet]->tLogVersion >= TLogVersion::V4) {
					bestSet = bestSatelliteSet;
				}

				TraceEvent("TLogPeekLogRouterOldSets", dbgid)
				    .detail("Tag", tag.toString())
				    .detail("Begin", begin)
				    .detail("OldEpoch", old.epochEnd)
				    .detail("RecoveredAt", recoveredAt.present() ? recoveredAt.get() : -1)
				    .detail("FirstOld", firstOld);
				// FIXME: do this merge on one of the logs in the other data center to avoid sending multiple copies
				// across the WAN
				return makeReference<ILogSystem::SetPeekCursor>(
				    localSets,
				    bestSet,
				    localSets[bestSet]->bestLocationFor(tag),
				    tag,
				    begin,
				    firstOld && recoveredAt.present() ? recoveredAt.get() + 1 : old.epochEnd,
				    true);
			}
			firstOld = false;
		}
		return makeReference<ILogSystem::ServerPeekCursor>(
		    Reference<AsyncVar<OptionalInterface<TLogInterface>>>(), tag, begin, getPeekEnd(), false, false);
	}

	Version getKnownCommittedVersion() final {
		Version result = invalidVersion;
		for (auto& it : lockResults) {
			auto versions = TagPartitionedLogSystem::getDurableVersion(dbgid, it);
			if (versions.present()) {
				result = std::max(result, versions.get().first);
			}
		}
		return result;
	}

	Future<Void> onKnownCommittedVersionChange() final {
		std::vector<Future<Void>> result;
		for (auto& it : lockResults) {
			result.push_back(TagPartitionedLogSystem::getDurableVersionChanged(it));
		}
		if (!result.size()) {
			return Never();
		}
		return waitForAny(result);
	}

	void popLogRouter(Version upTo,
	                  Tag tag,
	                  Version durableKnownCommittedVersion,
	                  int8_t popLocality) { // FIXME: do not need to pop all generations of old logs
		if (!upTo)
			return;
		for (auto& t : tLogs) {
			if (t->locality == popLocality) {
				for (auto& log : t->logRouters) {
					Version prev = outstandingPops[std::make_pair(log->get().id(), tag)].first;
					if (prev < upTo)
						outstandingPops[std::make_pair(log->get().id(), tag)] =
						    std::make_pair(upTo, durableKnownCommittedVersion);
					if (prev == 0) {
						popActors.add(popFromLog(
						    this, log, tag, 0.0)); // Fast pop time because log routers can only hold 5 seconds of data.
					}
				}
			}
		}

		for (auto& old : oldLogData) {
			for (auto& t : old.tLogs) {
				if (t->locality == popLocality) {
					for (auto& log : t->logRouters) {
						Version prev = outstandingPops[std::make_pair(log->get().id(), tag)].first;
						if (prev < upTo)
							outstandingPops[std::make_pair(log->get().id(), tag)] =
							    std::make_pair(upTo, durableKnownCommittedVersion);
						if (prev == 0)
							popActors.add(popFromLog(this, log, tag, 0.0));
					}
				}
			}
		}
	}

	void popTxs(Version upTo, int8_t popLocality) final {
		if (getTLogVersion() < TLogVersion::V4) {
			pop(upTo, txsTag, 0, popLocality);
		} else {
			for (int i = 0; i < txsTags; i++) {
				pop(upTo, Tag(tagLocalityTxs, i), 0, popLocality);
			}
		}
	}

	// pop 'tag.locality' type data up to the 'upTo' version
	void pop(Version upTo, Tag tag, Version durableKnownCommittedVersion, int8_t popLocality) final {
		if (upTo <= 0)
			return;
		if (tag.locality == tagLocalityRemoteLog) {
			popLogRouter(upTo, tag, durableKnownCommittedVersion, popLocality);
			return;
		}
		for (auto& t : tLogs) {
			if (t->locality == tagLocalitySpecial || t->locality == tag.locality ||
			    tag.locality == tagLocalityUpgraded ||
			    (tag.locality < 0 && ((popLocality == tagLocalityInvalid) == t->isLocal))) {
				for (auto& log : t->logServers) {
					Version prev = outstandingPops[std::make_pair(log->get().id(), tag)].first;
					if (prev < upTo) {
						// update pop version for popFromLog actor
						outstandingPops[std::make_pair(log->get().id(), tag)] =
						    std::make_pair(upTo, durableKnownCommittedVersion);
					}
					if (prev == 0) {
						// pop tag from log upto version defined in outstandingPops[].first
						popActors.add(popFromLog(this, log, tag, 1.0)); //< FIXME: knob
					}
				}
			}
		}
	}

	// pop tag from log up to the version defined in self->outstandingPops[].first
	ACTOR static Future<Void> popFromLog(TagPartitionedLogSystem* self,
	                                     Reference<AsyncVar<OptionalInterface<TLogInterface>>> log,
	                                     Tag tag,
	                                     double time) {
		state Version last = 0;
		loop {
			wait(delay(time, TaskPriority::TLogPop));

			// to: first is upto version, second is durableKnownComittedVersion
			state std::pair<Version, Version> to = self->outstandingPops[std::make_pair(log->get().id(), tag)];

			if (to.first <= last) {
				self->outstandingPops.erase(std::make_pair(log->get().id(), tag));
				return Void();
			}

			try {
				if (!log->get().present())
					return Void();
				wait(log->get().interf().popMessages.getReply(TLogPopRequest(to.first, to.second, tag),
				                                              TaskPriority::TLogPop));

				last = to.first;
			} catch (Error& e) {
				if (e.code() == error_code_actor_cancelled)
					throw;
				TraceEvent((e.code() == error_code_broken_promise) ? SevInfo : SevError, "LogPopError", self->dbgid)
				    .error(e)
				    .detail("Log", log->get().id());
				return Void(); // Leaving outstandingPops filled in means no further pop requests to this tlog from this
				               // logSystem
			}
		}
	}

	ACTOR static Future<Version> getPoppedFromTLog(Reference<AsyncVar<OptionalInterface<TLogInterface>>> log, Tag tag) {
		loop {
			choose {
				when(TLogPeekReply rep =
				         wait(log->get().present() ? brokenPromiseToNever(log->get().interf().peekMessages.getReply(
				                                         TLogPeekRequest(-1, tag, false, false)))
				                                   : Never())) {
					ASSERT(rep.popped.present());
					return rep.popped.get();
				}
				when(wait(log->onChange())) {}
			}
		}
	}

	ACTOR static Future<Version> getPoppedTxs(TagPartitionedLogSystem* self) {
		state std::vector<std::vector<Future<Version>>> poppedFutures;
		state std::vector<Future<Void>> poppedReady;
		if (self->tLogs.size()) {
			poppedFutures.push_back(std::vector<Future<Version>>());
			for (auto& it : self->tLogs) {
				for (auto& log : it->logServers) {
					poppedFutures.back().push_back(getPoppedFromTLog(
					    log, self->tLogs[0]->tLogVersion < TLogVersion::V4 ? txsTag : Tag(tagLocalityTxs, 0)));
				}
			}
			poppedReady.push_back(waitForAny(poppedFutures.back()));
		}

		for (auto& old : self->oldLogData) {
			if (old.tLogs.size()) {
				poppedFutures.push_back(std::vector<Future<Version>>());
				for (auto& it : old.tLogs) {
					for (auto& log : it->logServers) {
						poppedFutures.back().push_back(getPoppedFromTLog(
						    log, old.tLogs[0]->tLogVersion < TLogVersion::V4 ? txsTag : Tag(tagLocalityTxs, 0)));
					}
				}
				poppedReady.push_back(waitForAny(poppedFutures.back()));
			}
		}

		state Future<Void> maxGetPoppedDuration = delay(SERVER_KNOBS->TXS_POPPED_MAX_DELAY);
		wait(waitForAll(poppedReady) || maxGetPoppedDuration);

		if (maxGetPoppedDuration.isReady()) {
			TraceEvent(SevWarnAlways, "PoppedTxsNotReady", self->dbgid);
		}

		Version maxPopped = 1;
		for (auto& it : poppedFutures) {
			for (auto& v : it) {
				if (v.isReady()) {
					maxPopped = std::max(maxPopped, v.get());
				}
			}
		}
		return maxPopped;
	}

	Future<Version> getTxsPoppedVersion() final { return getPoppedTxs(this); }

	ACTOR static Future<Void> confirmEpochLive_internal(Reference<LogSet> logSet, Optional<UID> debugID) {
		state vector<Future<Void>> alive;
		int numPresent = 0;
		for (auto& t : logSet->logServers) {
			if (t->get().present()) {
				alive.push_back(brokenPromiseToNever(t->get().interf().confirmRunning.getReply(
				    TLogConfirmRunningRequest(debugID), TaskPriority::TLogConfirmRunningReply)));
				numPresent++;
			} else {
				alive.push_back(Never());
			}
		}

		wait(quorum(alive, std::min(logSet->tLogReplicationFactor, numPresent - logSet->tLogWriteAntiQuorum)));

		state std::vector<LocalityEntry> aliveEntries;
		state std::vector<bool> responded(alive.size(), false);
		loop {
			for (int i = 0; i < alive.size(); i++) {
				if (!responded[i] && alive[i].isReady() && !alive[i].isError()) {
					aliveEntries.push_back(logSet->logEntryArray[i]);
					responded[i] = true;
				}
			}

			if (logSet->satisfiesPolicy(aliveEntries)) {
				return Void();
			}

			// The current set of responders that we have weren't enough to form a quorum, so we must
			// wait for more responses and try again.
			std::vector<Future<Void>> changes;
			for (int i = 0; i < alive.size(); i++) {
				if (!alive[i].isReady()) {
					changes.push_back(ready(alive[i]));
				} else if (alive[i].isReady() && alive[i].isError() &&
				           alive[i].getError().code() == error_code_tlog_stopped) {
					// All commits must go to all TLogs.  If any TLog is stopped, then our epoch has ended.
					return Never();
				}
			}
			ASSERT(changes.size() != 0);
			wait(waitForAny(changes));
		}
	}

	// Returns success after confirming that pushes in the current epoch are still possible
	Future<Void> confirmEpochLive(Optional<UID> debugID) final {
		vector<Future<Void>> quorumResults;
		for (auto& it : tLogs) {
			if (it->isLocal && it->logServers.size()) {
				quorumResults.push_back(confirmEpochLive_internal(it, debugID));
			}
		}

		return waitForAll(quorumResults);
	}

	Future<Void> endEpoch() final {
		std::vector<Future<Void>> lockResults;
		for (auto& logSet : tLogs) {
			for (auto& log : logSet->logServers) {
				lockResults.push_back(success(lockTLog(dbgid, log)));
			}
		}
		return waitForAll(lockResults);
	}

	// Call only after end_epoch() has successfully completed.  Returns a new epoch immediately following this one.
	// The new epoch is only provisional until the caller updates the coordinated DBCoreState.
	Future<Reference<ILogSystem>> newEpoch(RecruitFromConfigurationReply const& recr,
	                                       Future<RecruitRemoteFromConfigurationReply> const& fRemoteWorkers,
	                                       DatabaseConfiguration const& config,
	                                       LogEpoch recoveryCount,
	                                       int8_t primaryLocality,
	                                       int8_t remoteLocality,
	                                       std::vector<Tag> const& allTags,
	                                       Reference<AsyncVar<bool>> const& recruitmentStalled) final {
		return newEpoch(Reference<TagPartitionedLogSystem>::addRef(this),
		                recr,
		                fRemoteWorkers,
		                config,
		                recoveryCount,
		                primaryLocality,
		                remoteLocality,
		                allTags,
		                recruitmentStalled);
	}

	LogSystemConfig getLogSystemConfig() const final {
		LogSystemConfig logSystemConfig(epoch);
		logSystemConfig.logSystemType = logSystemType;
		logSystemConfig.expectedLogSets = expectedLogSets;
		logSystemConfig.logRouterTags = logRouterTags;
		logSystemConfig.txsTags = txsTags;
		logSystemConfig.recruitmentID = recruitmentID;
		logSystemConfig.stopped = stopped;
		logSystemConfig.recoveredAt = recoveredAt;
		logSystemConfig.pseudoLocalities = pseudoLocalities;
		logSystemConfig.oldestBackupEpoch = oldestBackupEpoch;
		for (const Reference<LogSet>& logSet : tLogs) {
			if (logSet->isLocal || remoteLogsWrittenToCoreState) {
				logSystemConfig.tLogs.emplace_back(*logSet);
			}
		}

		if (!recoveryCompleteWrittenToCoreState.get()) {
			for (const auto& oldData : oldLogData) {
				logSystemConfig.oldTLogs.emplace_back(oldData);
			}
		}
		return logSystemConfig;
	}

	Standalone<StringRef> getLogsValue() const final {
		vector<std::pair<UID, NetworkAddress>> logs;
		vector<std::pair<UID, NetworkAddress>> oldLogs;
		for (auto& t : tLogs) {
			if (t->isLocal || remoteLogsWrittenToCoreState) {
				for (int i = 0; i < t->logServers.size(); i++) {
					logs.emplace_back(t->logServers[i]->get().id(),
					                  t->logServers[i]->get().present() ? t->logServers[i]->get().interf().address()
					                                                    : NetworkAddress());
				}
			}
		}
		if (!recoveryCompleteWrittenToCoreState.get()) {
			for (int i = 0; i < oldLogData.size(); i++) {
				for (auto& t : oldLogData[i].tLogs) {
					for (int j = 0; j < t->logServers.size(); j++) {
						oldLogs.emplace_back(t->logServers[j]->get().id(),
						                     t->logServers[j]->get().present()
						                         ? t->logServers[j]->get().interf().address()
						                         : NetworkAddress());
					}
				}
			}
		}
		return logsValue(logs, oldLogs);
	}

	Future<Void> onLogSystemConfigChange() final {
		std::vector<Future<Void>> changes;
		changes.push_back(logSystemConfigChanged.onTrigger());
		for (auto& t : tLogs) {
			for (int i = 0; i < t->logServers.size(); i++) {
				changes.push_back(t->logServers[i]->onChange());
			}
		}
		for (int i = 0; i < oldLogData.size(); i++) {
			for (auto& t : oldLogData[i].tLogs) {
				for (int j = 0; j < t->logServers.size(); j++) {
					changes.push_back(t->logServers[j]->onChange());
				}
			}
		}

		if (hasRemoteServers && !remoteRecovery.isReady()) {
			changes.push_back(remoteRecovery);
		}

		return waitForAny(changes);
	}

	Version getEnd() const final {
		ASSERT(recoverAt.present());
		return recoverAt.get() + 1;
	}

	Version getPeekEnd() const {
		if (recoverAt.present())
			return getEnd();
		else
			return std::numeric_limits<Version>::max();
	}

	void getPushLocations(VectorRef<Tag> tags, std::vector<int>& locations, bool allLocations) const final {
		int locationOffset = 0;
		for (auto& log : tLogs) {
			if (log->isLocal && log->logServers.size()) {
				log->getPushLocations(tags, locations, locationOffset, allLocations);
				locationOffset += log->logServers.size();
			}
		}
	}

	bool hasRemoteLogs() const final { return logRouterTags > 0 || pseudoLocalities.size() > 0; }

	Tag getRandomRouterTag() const final {
		return Tag(tagLocalityLogRouter, deterministicRandom()->randomInt(0, logRouterTags));
	}

	Tag getRandomTxsTag() const final { return Tag(tagLocalityTxs, deterministicRandom()->randomInt(0, txsTags)); }

	TLogVersion getTLogVersion() const final { return tLogs[0]->tLogVersion; }

	int getLogRouterTags() const final { return logRouterTags; }

	Version getBackupStartVersion() const final {
		ASSERT(tLogs.size() > 0);
		return backupStartVersion;
	}

	std::map<LogEpoch, ILogSystem::EpochTagsVersionsInfo> getOldEpochTagsVersionsInfo() const final {
		std::map<LogEpoch, EpochTagsVersionsInfo> epochInfos;
		for (const auto& old : oldLogData) {
			epochInfos.insert(
			    { old.epoch, ILogSystem::EpochTagsVersionsInfo(old.logRouterTags, old.epochBegin, old.epochEnd) });
			TraceEvent("OldEpochTagsVersions", dbgid)
			    .detail("Epoch", old.epoch)
			    .detail("Tags", old.logRouterTags)
			    .detail("BeginVersion", old.epochBegin)
			    .detail("EndVersion", old.epochEnd);
		}
		return epochInfos;
	}

	inline Reference<LogSet> getEpochLogSet(LogEpoch epoch) const {
		for (const auto& old : oldLogData) {
			if (epoch == old.epoch)
				return old.tLogs[0];
		}
		return Reference<LogSet>(nullptr);
	}

	void setBackupWorkers(const std::vector<InitializeBackupReply>& replies) final {
		ASSERT(tLogs.size() > 0);

		Reference<LogSet> logset = tLogs[0]; // Master recruits this epoch's worker first.
		LogEpoch logsetEpoch = this->epoch;
		oldestBackupEpoch = this->epoch;
		for (const auto& reply : replies) {
			if (removedBackupWorkers.count(reply.interf.id()) > 0) {
				removedBackupWorkers.erase(reply.interf.id());
				continue;
			}
			auto worker = makeReference<AsyncVar<OptionalInterface<BackupInterface>>>(
			    OptionalInterface<BackupInterface>(reply.interf));
			if (reply.backupEpoch != logsetEpoch) {
				// find the logset from oldLogData
				logsetEpoch = reply.backupEpoch;
				oldestBackupEpoch = std::min(oldestBackupEpoch, logsetEpoch);
				logset = getEpochLogSet(logsetEpoch);
				ASSERT(logset.isValid());
			}
			logset->backupWorkers.push_back(worker);
			TraceEvent("AddBackupWorker", dbgid)
			    .detail("Epoch", logsetEpoch)
			    .detail("BackupWorkerID", reply.interf.id());
		}
		TraceEvent("SetOldestBackupEpoch", dbgid).detail("Epoch", oldestBackupEpoch);
		backupWorkerChanged.trigger();
	}

	bool removeBackupWorker(const BackupWorkerDoneRequest& req) final {
		bool removed = false;
		Reference<LogSet> logset = getEpochLogSet(req.backupEpoch);
		if (logset.isValid()) {
			for (auto it = logset->backupWorkers.begin(); it != logset->backupWorkers.end(); it++) {
				if (it->getPtr()->get().interf().id() == req.workerUID) {
					logset->backupWorkers.erase(it);
					removed = true;
					break;
				}
			}
		}

		if (removed) {
			oldestBackupEpoch = epoch;
			for (const auto& old : oldLogData) {
				if (old.epoch < oldestBackupEpoch && old.tLogs[0]->backupWorkers.size() > 0) {
					oldestBackupEpoch = old.epoch;
				}
			}
			backupWorkerChanged.trigger();
		} else {
			removedBackupWorkers.insert(req.workerUID);
		}

		TraceEvent("RemoveBackupWorker", dbgid)
		    .detail("Removed", removed)
		    .detail("BackupEpoch", req.backupEpoch)
		    .detail("WorkerID", req.workerUID)
		    .detail("OldestBackupEpoch", oldestBackupEpoch);
		return removed;
	}

	LogEpoch getOldestBackupEpoch() const final { return oldestBackupEpoch; }

	void setOldestBackupEpoch(LogEpoch epoch) final {
		oldestBackupEpoch = epoch;
		backupWorkerChanged.trigger();
	}

	ACTOR static Future<Void> monitorLog(Reference<AsyncVar<OptionalInterface<TLogInterface>>> logServer,
	                                     Reference<AsyncVar<bool>> failed) {
		state Future<Void> waitFailure;
		loop {
			if (logServer->get().present())
				waitFailure = waitFailureTracker(logServer->get().interf().waitFailure, failed);
			else
				failed->set(true);
			wait(logServer->onChange());
		}
	}

	Optional<std::pair<Version, Version>> static getDurableVersion(
	    UID dbgid,
	    LogLockInfo lockInfo,
	    std::vector<Reference<AsyncVar<bool>>> failed = std::vector<Reference<AsyncVar<bool>>>(),
	    Optional<Version> lastEnd = Optional<Version>()) {
		Reference<LogSet> logSet = lockInfo.logSet;
		// To ensure consistent recovery, the number of servers NOT in the write quorum plus the number of servers NOT
		// in the read quorum have to be strictly less than the replication factor.  Otherwise there could be a replica
		// set consistent entirely of servers that are out of date due to not being in the write quorum or unavailable
		// due to not being in the read quorum. So with N = # of tlogs, W = antiquorum, R = required count, F =
		// replication factor, W + (N - R) < F, and optimally (N-W)+(N-R)=F-1.  Thus R=N+1-F+W.
		int requiredCount =
		    (int)logSet->logServers.size() + 1 - logSet->tLogReplicationFactor + logSet->tLogWriteAntiQuorum;
		ASSERT(requiredCount > 0 && requiredCount <= logSet->logServers.size());
		ASSERT(logSet->tLogReplicationFactor >= 1 && logSet->tLogReplicationFactor <= logSet->logServers.size());
		ASSERT(logSet->tLogWriteAntiQuorum >= 0 && logSet->tLogWriteAntiQuorum < logSet->logServers.size());

		std::vector<LocalityData> availableItems, badCombo;
		std::vector<TLogLockResult> results;
		std::string sServerState;
		LocalityGroup unResponsiveSet;

		for (int t = 0; t < logSet->logServers.size(); t++) {
			if (lockInfo.replies[t].isReady() && !lockInfo.replies[t].isError() &&
			    (!failed.size() || !failed[t]->get())) {
				results.push_back(lockInfo.replies[t].get());
				availableItems.push_back(logSet->tLogLocalities[t]);
				sServerState += 'a';
			} else {
				unResponsiveSet.add(logSet->tLogLocalities[t]);
				sServerState += 'f';
			}
		}

		// Check if the list of results is not larger than the anti quorum
		bool bTooManyFailures = (results.size() <= logSet->tLogWriteAntiQuorum);

		// Check if failed logs complete the policy
		bTooManyFailures = bTooManyFailures || ((unResponsiveSet.size() >= logSet->tLogReplicationFactor) &&
		                                        (unResponsiveSet.validate(logSet->tLogPolicy)));

		// Check all combinations of the AntiQuorum within the failed
		if (!bTooManyFailures && (logSet->tLogWriteAntiQuorum) &&
		    (!validateAllCombinations(
		        badCombo, unResponsiveSet, logSet->tLogPolicy, availableItems, logSet->tLogWriteAntiQuorum, false))) {
			TraceEvent("EpochEndBadCombo", dbgid)
			    .detail("Required", requiredCount)
			    .detail("Present", results.size())
			    .detail("ServerState", sServerState);
			bTooManyFailures = true;
		}

		ASSERT(logSet->logServers.size() == lockInfo.replies.size());
		if (!bTooManyFailures) {
			std::sort(results.begin(), results.end(), sort_by_end());
			int absent = logSet->logServers.size() - results.size();
			int safe_range_begin = logSet->tLogWriteAntiQuorum;
			int new_safe_range_begin = std::min(logSet->tLogWriteAntiQuorum, (int)(results.size() - 1));
			int safe_range_end = logSet->tLogReplicationFactor - absent;

			if (!lastEnd.present() || ((safe_range_end > 0) && (safe_range_end - 1 < results.size()) &&
			                           results[safe_range_end - 1].end < lastEnd.get())) {
				Version knownCommittedVersion = 0;
				for (int i = 0; i < results.size(); i++) {
					knownCommittedVersion = std::max(knownCommittedVersion, results[i].knownCommittedVersion);
				}

				if (knownCommittedVersion > results[new_safe_range_begin].end) {
					knownCommittedVersion = results[new_safe_range_begin].end;
				}

				TraceEvent("GetDurableResult", dbgid)
				    .detail("Required", requiredCount)
				    .detail("Present", results.size())
				    .detail("ServerState", sServerState)
				    .detail("RecoveryVersion",
				            ((safe_range_end > 0) && (safe_range_end - 1 < results.size()))
				                ? results[safe_range_end - 1].end
				                : -1)
				    .detail("EndVersion", results[new_safe_range_begin].end)
				    .detail("SafeBegin", safe_range_begin)
				    .detail("SafeEnd", safe_range_end)
				    .detail("NewSafeBegin", new_safe_range_begin)
				    .detail("KnownCommittedVersion", knownCommittedVersion)
				    .detail("EpochEnd", lockInfo.epochEnd);

				return std::make_pair(knownCommittedVersion, results[new_safe_range_begin].end);
			}
		}
		TraceEvent("GetDurableResultWaiting", dbgid)
		    .detail("Required", requiredCount)
		    .detail("Present", results.size())
		    .detail("ServerState", sServerState);
		return Optional<std::pair<Version, Version>>();
	}

	ACTOR static Future<Void> getDurableVersionChanged(
	    LogLockInfo lockInfo,
	    std::vector<Reference<AsyncVar<bool>>> failed = std::vector<Reference<AsyncVar<bool>>>()) {
		// Wait for anything relevant to change
		std::vector<Future<Void>> changes;
		for (int j = 0; j < lockInfo.logSet->logServers.size(); j++) {
			if (!lockInfo.replies[j].isReady())
				changes.push_back(ready(lockInfo.replies[j]));
			else {
				changes.push_back(lockInfo.logSet->logServers[j]->onChange());
				if (failed.size()) {
					changes.push_back(failed[j]->onChange());
				}
			}
		}
		ASSERT(changes.size());
		wait(waitForAny(changes));
		return Void();
	}

	ACTOR static Future<Void> epochEnd(Reference<AsyncVar<Reference<ILogSystem>>> outLogSystem,
	                                   UID dbgid,
	                                   DBCoreState prevState,
	                                   FutureStream<TLogRejoinRequest> rejoinRequests,
	                                   LocalityData locality,
	                                   bool* forceRecovery) {
		// Stops a co-quorum of tlogs so that no further versions can be committed until the DBCoreState coordination
		// state is changed Creates a new logSystem representing the (now frozen) epoch No other important side effects.
		// The writeQuorum in the master info is from the previous configuration

		if (!prevState.tLogs.size()) {
			// This is a brand new database
			auto logSystem = makeReference<TagPartitionedLogSystem>(dbgid, locality, 0);
			logSystem->logSystemType = prevState.logSystemType;
			logSystem->recoverAt = 0;
			logSystem->knownCommittedVersion = 0;
			logSystem->stopped = true;
			outLogSystem->set(logSystem);
			wait(Future<Void>(Never()));
			throw internal_error();
		}

		if (*forceRecovery) {
			DBCoreState modifiedState = prevState;

			int8_t primaryLocality = -1;
			for (auto& coreSet : modifiedState.tLogs) {
				if (coreSet.isLocal && coreSet.locality >= 0 && coreSet.tLogLocalities[0].dcId() != locality.dcId()) {
					primaryLocality = coreSet.locality;
					break;
				}
			}

			bool foundRemote = false;
			int8_t remoteLocality = -1;
			int modifiedLogSets = 0;
			int removedLogSets = 0;
			if (primaryLocality >= 0) {
				auto copiedLogs = modifiedState.tLogs;
				for (auto& coreSet : copiedLogs) {
					if (coreSet.locality != primaryLocality && coreSet.locality >= 0) {
						foundRemote = true;
						remoteLocality = coreSet.locality;
						modifiedState.tLogs.clear();
						modifiedState.tLogs.push_back(coreSet);
						modifiedState.tLogs[0].isLocal = true;
						modifiedState.logRouterTags = 0;
						modifiedLogSets++;
						break;
					}
				}

				while (!foundRemote && modifiedState.oldTLogData.size()) {
					for (auto& coreSet : modifiedState.oldTLogData[0].tLogs) {
						if (coreSet.locality != primaryLocality && coreSet.locality >= tagLocalitySpecial) {
							foundRemote = true;
							remoteLocality = coreSet.locality;
							modifiedState.tLogs.clear();
							modifiedState.tLogs.push_back(coreSet);
							modifiedState.tLogs[0].isLocal = true;
							modifiedState.logRouterTags = 0;
							modifiedState.txsTags = modifiedState.oldTLogData[0].txsTags;
							modifiedLogSets++;
							break;
						}
					}
					modifiedState.oldTLogData.erase(modifiedState.oldTLogData.begin());
					removedLogSets++;
				}

				if (foundRemote) {
					for (int i = 0; i < modifiedState.oldTLogData.size(); i++) {
						bool found = false;
						auto copiedLogs = modifiedState.oldTLogData[i].tLogs;
						for (auto& coreSet : copiedLogs) {
							if (coreSet.locality == remoteLocality || coreSet.locality == tagLocalitySpecial) {
								found = true;
								if (!coreSet.isLocal || copiedLogs.size() > 1) {
									modifiedState.oldTLogData[i].tLogs.clear();
									modifiedState.oldTLogData[i].tLogs.push_back(coreSet);
									modifiedState.oldTLogData[i].tLogs[0].isLocal = true;
									modifiedState.oldTLogData[i].logRouterTags = 0;
									modifiedState.oldTLogData[i].epochBegin =
									    modifiedState.oldTLogData[i].tLogs[0].startVersion;
									modifiedState.oldTLogData[i].epochEnd =
									    (i == 0 ? modifiedState.tLogs[0].startVersion
									            : modifiedState.oldTLogData[i - 1].tLogs[0].startVersion);
									modifiedLogSets++;
								}
								break;
							}
						}
						if (!found) {
							modifiedState.oldTLogData.erase(modifiedState.oldTLogData.begin() + i);
							removedLogSets++;
							i--;
						}
					}
					prevState = modifiedState;
				} else {
					*forceRecovery = false;
				}
			} else {
				*forceRecovery = false;
			}
			TraceEvent(SevWarnAlways, "ForcedRecovery", dbgid)
			    .detail("PrimaryLocality", primaryLocality)
			    .detail("RemoteLocality", remoteLocality)
			    .detail("FoundRemote", foundRemote)
			    .detail("Modified", modifiedLogSets)
			    .detail("Removed", removedLogSets);
			for (int i = 0; i < prevState.tLogs.size(); i++) {
				TraceEvent("ForcedRecoveryTLogs", dbgid)
				    .detail("I", i)
				    .detail("Log", ::describe(prevState.tLogs[i].tLogs))
				    .detail("Loc", prevState.tLogs[i].locality)
				    .detail("Txs", prevState.txsTags);
			}
			for (int i = 0; i < prevState.oldTLogData.size(); i++) {
				for (int j = 0; j < prevState.oldTLogData[i].tLogs.size(); j++) {
					TraceEvent("ForcedRecoveryTLogs", dbgid)
					    .detail("I", i)
					    .detail("J", j)
					    .detail("Log", ::describe(prevState.oldTLogData[i].tLogs[j].tLogs))
					    .detail("Loc", prevState.oldTLogData[i].tLogs[j].locality)
					    .detail("Txs", prevState.oldTLogData[i].txsTags);
				}
			}
		}

		TEST(true); // Master recovery from pre-existing database

		// trackRejoins listens for rejoin requests from the tLogs that we are recovering from, to learn their
		// TLogInterfaces
		state std::vector<LogLockInfo> lockResults;
		state
		    std::vector<std::pair<Reference<AsyncVar<OptionalInterface<TLogInterface>>>, Reference<IReplicationPolicy>>>
		        allLogServers;
		state std::vector<Reference<LogSet>> logServers;
		state std::vector<OldLogData> oldLogData;
		state std::vector<std::vector<Reference<AsyncVar<bool>>>> logFailed;
		state std::vector<Future<Void>> failureTrackers;

		for (const CoreTLogSet& coreSet : prevState.tLogs) {
			logServers.push_back(makeReference<LogSet>(coreSet));
			std::vector<Reference<AsyncVar<bool>>> failed;

			for (const auto& logVar : logServers.back()->logServers) {
				allLogServers.emplace_back(logVar, coreSet.tLogPolicy);
				failed.push_back(makeReference<AsyncVar<bool>>());
				failureTrackers.push_back(monitorLog(logVar, failed.back()));
			}
			logFailed.push_back(failed);
		}

		for (const auto& oldTlogData : prevState.oldTLogData) {
			oldLogData.emplace_back(oldTlogData);

			for (const auto& logSet : oldLogData.back().tLogs) {
				for (const auto& logVar : logSet->logServers) {
					allLogServers.emplace_back(logVar, logSet->tLogPolicy);
				}
			}
		}
		state Future<Void> rejoins = trackRejoins(dbgid, allLogServers, rejoinRequests);

		lockResults.resize(logServers.size());
		std::set<int8_t> lockedLocalities;
		bool foundSpecial = false;
		for (int i = 0; i < logServers.size(); i++) {
			if (logServers[i]->locality == tagLocalitySpecial || logServers[i]->locality == tagLocalityUpgraded) {
				foundSpecial = true;
			}
			lockedLocalities.insert(logServers[i]->locality);
			lockResults[i].isCurrent = true;
			lockResults[i].logSet = logServers[i];
			for (int t = 0; t < logServers[i]->logServers.size(); t++) {
				lockResults[i].replies.push_back(lockTLog(dbgid, logServers[i]->logServers[t]));
			}
		}

		for (auto& old : oldLogData) {
			if (foundSpecial) {
				break;
			}
			for (auto& log : old.tLogs) {
				if (log->locality == tagLocalitySpecial || log->locality == tagLocalityUpgraded) {
					foundSpecial = true;
					break;
				}
				if (!lockedLocalities.count(log->locality)) {
					TraceEvent("EpochEndLockExtra").detail("Locality", log->locality);
					TEST(true); // locking old generations for version information
					lockedLocalities.insert(log->locality);
					LogLockInfo lockResult;
					lockResult.epochEnd = old.epochEnd;
					lockResult.logSet = log;
					for (int t = 0; t < log->logServers.size(); t++) {
						lockResult.replies.push_back(lockTLog(dbgid, log->logServers[t]));
					}
					lockResults.push_back(lockResult);
				}
			}
		}

		if (*forceRecovery) {
			state std::vector<LogLockInfo> allLockResults;
			ASSERT(lockResults.size() == 1);
			allLockResults.push_back(lockResults[0]);
			for (auto& old : oldLogData) {
				ASSERT(old.tLogs.size() == 1);
				LogLockInfo lockResult;
				lockResult.epochEnd = old.epochEnd;
				lockResult.logSet = old.tLogs[0];
				for (int t = 0; t < old.tLogs[0]->logServers.size(); t++) {
					lockResult.replies.push_back(lockTLog(dbgid, old.tLogs[0]->logServers[t]));
				}
				allLockResults.push_back(lockResult);
			}

			state int lockNum = 0;
			state Version maxRecoveryVersion = 0;
			state int maxRecoveryIndex = 0;
			while (lockNum < allLockResults.size()) {
				auto versions = TagPartitionedLogSystem::getDurableVersion(dbgid, allLockResults[lockNum]);
				if (versions.present()) {
					if (versions.get().second > maxRecoveryVersion) {
						TraceEvent("HigherRecoveryVersion", dbgid)
						    .detail("Idx", lockNum)
						    .detail("Ver", versions.get().second);
						maxRecoveryVersion = versions.get().second;
						maxRecoveryIndex = lockNum;
					}
					lockNum++;
				} else {
					wait(TagPartitionedLogSystem::getDurableVersionChanged(allLockResults[lockNum]));
				}
			}
			if (maxRecoveryIndex > 0) {
				logServers = oldLogData[maxRecoveryIndex - 1].tLogs;
				prevState.txsTags = oldLogData[maxRecoveryIndex - 1].txsTags;
				lockResults[0] = allLockResults[maxRecoveryIndex];
				lockResults[0].isCurrent = true;

				std::vector<Reference<AsyncVar<bool>>> failed;
				for (auto& log : logServers[0]->logServers) {
					failed.push_back(makeReference<AsyncVar<bool>>());
					failureTrackers.push_back(monitorLog(log, failed.back()));
				}
				ASSERT(logFailed.size() == 1);
				logFailed[0] = failed;
				oldLogData.erase(oldLogData.begin(), oldLogData.begin() + maxRecoveryIndex);
			}
		}

		state Optional<Version> lastEnd;
		state Version knownCommittedVersion = 0;
		loop {
			Version minEnd = std::numeric_limits<Version>::max();
			Version maxEnd = 0;
			std::vector<Future<Void>> changes;
			for (int log = 0; log < logServers.size(); log++) {
				if (!logServers[log]->isLocal) {
					continue;
				}
				auto versions =
				    TagPartitionedLogSystem::getDurableVersion(dbgid, lockResults[log], logFailed[log], lastEnd);
				if (versions.present()) {
					knownCommittedVersion = std::max(knownCommittedVersion, versions.get().first);
					maxEnd = std::max(maxEnd, versions.get().second);
					minEnd = std::min(minEnd, versions.get().second);
				}
				changes.push_back(TagPartitionedLogSystem::getDurableVersionChanged(lockResults[log], logFailed[log]));
			}

			if (maxEnd > 0 && (!lastEnd.present() || maxEnd < lastEnd.get())) {
				TEST(lastEnd.present()); // Restarting recovery at an earlier point

				auto logSystem = makeReference<TagPartitionedLogSystem>(dbgid, locality, prevState.recoveryCount);

				lastEnd = minEnd;
				logSystem->tLogs = logServers;
				logSystem->logRouterTags = prevState.logRouterTags;
				logSystem->txsTags = prevState.txsTags;
				logSystem->oldLogData = oldLogData;
				logSystem->logSystemType = prevState.logSystemType;
				logSystem->rejoins = rejoins;
				logSystem->lockResults = lockResults;
				if (knownCommittedVersion > minEnd) {
					knownCommittedVersion = minEnd;
				}
				logSystem->recoverAt = minEnd;
				logSystem->knownCommittedVersion = knownCommittedVersion;
				TraceEvent(SevDebug, "FinalRecoveryVersionInfo")
				    .detail("KCV", knownCommittedVersion)
				    .detail("MinEnd", minEnd);
				logSystem->remoteLogsWrittenToCoreState = true;
				logSystem->stopped = true;
				logSystem->pseudoLocalities = prevState.pseudoLocalities;

				outLogSystem->set(logSystem);
			}

			wait(waitForAny(changes));
		}
	}

	ACTOR static Future<Void> recruitOldLogRouters(TagPartitionedLogSystem* self,
	                                               vector<WorkerInterface> workers,
	                                               LogEpoch recoveryCount,
	                                               int8_t locality,
	                                               Version startVersion,
	                                               std::vector<LocalityData> tLogLocalities,
	                                               Reference<IReplicationPolicy> tLogPolicy,
	                                               bool forRemote) {
		state vector<vector<Future<TLogInterface>>> logRouterInitializationReplies;
		state vector<Future<TLogInterface>> allReplies;
		int nextRouter = 0;
		state Version lastStart = std::numeric_limits<Version>::max();

		if (!forRemote) {
			Version maxStart = getMaxLocalStartVersion(self->tLogs);

			lastStart = std::max(startVersion, maxStart);
			if (self->logRouterTags == 0) {
				ASSERT_WE_THINK(false);
				self->logSystemConfigChanged.trigger();
				return Void();
			}

			bool found = false;
			for (auto& tLogs : self->tLogs) {
				if (tLogs->locality == locality) {
					found = true;
				}

				tLogs->logRouters.clear();
			}

			if (!found) {
				TraceEvent("RecruitingOldLogRoutersAddingLocality")
				    .detail("Locality", locality)
				    .detail("LastStart", lastStart);
				auto newLogSet = makeReference<LogSet>();
				newLogSet->locality = locality;
				newLogSet->startVersion = lastStart;
				newLogSet->isLocal = false;
				self->tLogs.push_back(newLogSet);
			}

			for (auto& tLogs : self->tLogs) {
				// Recruit log routers for old generations of the primary locality
				if (tLogs->locality == locality) {
					logRouterInitializationReplies.emplace_back();
					for (int i = 0; i < self->logRouterTags; i++) {
						InitializeLogRouterRequest req;
						req.recoveryCount = recoveryCount;
						req.routerTag = Tag(tagLocalityLogRouter, i);
						req.startVersion = lastStart;
						req.tLogLocalities = tLogLocalities;
						req.tLogPolicy = tLogPolicy;
						req.locality = locality;
						auto reply = transformErrors(
						    throwErrorOr(workers[nextRouter].logRouter.getReplyUnlessFailedFor(
						        req, SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
						    master_recovery_failed());
						logRouterInitializationReplies.back().push_back(reply);
						allReplies.push_back(reply);
						nextRouter = (nextRouter + 1) % workers.size();
					}
				}
			}
		}

		for (auto& old : self->oldLogData) {
			Version maxStart = getMaxLocalStartVersion(old.tLogs);

			if (old.logRouterTags == 0 || maxStart >= lastStart) {
				break;
			}
			lastStart = std::max(startVersion, maxStart);
			bool found = false;
			for (auto& tLogs : old.tLogs) {
				if (tLogs->locality == locality) {
					found = true;
				}
				tLogs->logRouters.clear();
			}

			if (!found) {
				TraceEvent("RecruitingOldLogRoutersAddingLocality")
				    .detail("Locality", locality)
				    .detail("LastStart", lastStart);
				auto newLogSet = makeReference<LogSet>();
				newLogSet->locality = locality;
				newLogSet->startVersion = lastStart;
				old.tLogs.push_back(newLogSet);
			}

			for (auto& tLogs : old.tLogs) {
				// Recruit log routers for old generations of the primary locality
				if (tLogs->locality == locality) {
					logRouterInitializationReplies.emplace_back();
					for (int i = 0; i < old.logRouterTags; i++) {
						InitializeLogRouterRequest req;
						req.recoveryCount = recoveryCount;
						req.routerTag = Tag(tagLocalityLogRouter, i);
						req.startVersion = lastStart;
						req.tLogLocalities = tLogLocalities;
						req.tLogPolicy = tLogPolicy;
						req.locality = locality;
						auto reply = transformErrors(
						    throwErrorOr(workers[nextRouter].logRouter.getReplyUnlessFailedFor(
						        req, SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
						    master_recovery_failed());
						logRouterInitializationReplies.back().push_back(reply);
						allReplies.push_back(reply);
						nextRouter = (nextRouter + 1) % workers.size();
					}
				}
			}
		}

		wait(waitForAll(allReplies));

		int nextReplies = 0;
		lastStart = std::numeric_limits<Version>::max();
		vector<Future<Void>> failed;

		if (!forRemote) {
			Version maxStart = getMaxLocalStartVersion(self->tLogs);

			lastStart = std::max(startVersion, maxStart);
			for (auto& tLogs : self->tLogs) {
				if (tLogs->locality == locality) {
					for (int i = 0; i < logRouterInitializationReplies[nextReplies].size(); i++) {
						tLogs->logRouters.push_back(makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
						    OptionalInterface<TLogInterface>(logRouterInitializationReplies[nextReplies][i].get())));
						failed.push_back(waitFailureClient(
						    logRouterInitializationReplies[nextReplies][i].get().waitFailure,
						    SERVER_KNOBS->TLOG_TIMEOUT,
						    -SERVER_KNOBS->TLOG_TIMEOUT / SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
						    /*trace=*/true));
					}
					nextReplies++;
				}
			}
		}

		for (auto& old : self->oldLogData) {
			Version maxStart = getMaxLocalStartVersion(old.tLogs);
			if (old.logRouterTags == 0 || maxStart >= lastStart) {
				break;
			}
			lastStart = std::max(startVersion, maxStart);
			for (auto& tLogs : old.tLogs) {
				if (tLogs->locality == locality) {
					for (int i = 0; i < logRouterInitializationReplies[nextReplies].size(); i++) {
						tLogs->logRouters.push_back(makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
						    OptionalInterface<TLogInterface>(logRouterInitializationReplies[nextReplies][i].get())));
						if (!forRemote) {
							failed.push_back(waitFailureClient(
							    logRouterInitializationReplies[nextReplies][i].get().waitFailure,
							    SERVER_KNOBS->TLOG_TIMEOUT,
							    -SERVER_KNOBS->TLOG_TIMEOUT / SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY,
							    /*trace=*/true));
						}
					}
					nextReplies++;
				}
			}
		}

		if (!forRemote) {
			self->logSystemConfigChanged.trigger();
			wait(failed.size() ? tagError<Void>(quorum(failed, 1), master_tlog_failed()) : Future<Void>(Never()));
			throw internal_error();
		}
		return Void();
	}

	static Version getMaxLocalStartVersion(const std::vector<Reference<LogSet>>& tLogs) {
		Version maxStart = 0;
		for (const auto& logSet : tLogs) {
			if (logSet->isLocal) {
				maxStart = std::max(maxStart, logSet->startVersion);
			}
		}
		return maxStart;
	}

	static std::vector<Tag> getLocalTags(int8_t locality, const std::vector<Tag>& allTags) {
		std::vector<Tag> localTags;
		for (const auto& tag : allTags) {
			if (locality == tagLocalitySpecial || locality == tag.locality || tag.locality < 0) {
				localTags.push_back(tag);
			}
		}
		return localTags;
	}

	ACTOR static Future<Void> newRemoteEpoch(TagPartitionedLogSystem* self,
	                                         Reference<TagPartitionedLogSystem> oldLogSystem,
	                                         Future<RecruitRemoteFromConfigurationReply> fRemoteWorkers,
	                                         DatabaseConfiguration configuration,
	                                         LogEpoch recoveryCount,
	                                         int8_t remoteLocality,
	                                         std::vector<Tag> allTags) {
		TraceEvent("RemoteLogRecruitment_WaitingForWorkers");
		state RecruitRemoteFromConfigurationReply remoteWorkers = wait(fRemoteWorkers);

		state Reference<LogSet> logSet(new LogSet());
		logSet->tLogReplicationFactor = configuration.getRemoteTLogReplicationFactor();
		logSet->tLogVersion = configuration.tLogVersion;
		logSet->tLogPolicy = configuration.getRemoteTLogPolicy();
		logSet->isLocal = false;
		logSet->locality = remoteLocality;

		logSet->startVersion = oldLogSystem->knownCommittedVersion + 1;
		state int lockNum = 0;
		while (lockNum < oldLogSystem->lockResults.size()) {
			if (oldLogSystem->lockResults[lockNum].logSet->locality == remoteLocality) {
				loop {
					auto versions =
					    TagPartitionedLogSystem::getDurableVersion(self->dbgid, oldLogSystem->lockResults[lockNum]);
					if (versions.present()) {
						logSet->startVersion =
						    std::min(std::min(versions.get().first + 1, oldLogSystem->lockResults[lockNum].epochEnd),
						             logSet->startVersion);
						break;
					}
					wait(TagPartitionedLogSystem::getDurableVersionChanged(oldLogSystem->lockResults[lockNum]));
				}
				break;
			}
			lockNum++;
		}

		vector<LocalityData> localities;
		localities.resize(remoteWorkers.remoteTLogs.size());
		for (int i = 0; i < remoteWorkers.remoteTLogs.size(); i++) {
			localities[i] = remoteWorkers.remoteTLogs[i].locality;
		}

		state Future<Void> oldRouterRecruitment = Void();
		if (logSet->startVersion < oldLogSystem->knownCommittedVersion + 1) {
			ASSERT(oldLogSystem->logRouterTags > 0);
			oldRouterRecruitment = TagPartitionedLogSystem::recruitOldLogRouters(self,
			                                                                     remoteWorkers.logRouters,
			                                                                     recoveryCount,
			                                                                     remoteLocality,
			                                                                     logSet->startVersion,
			                                                                     localities,
			                                                                     logSet->tLogPolicy,
			                                                                     true);
		}

		state vector<Future<TLogInterface>> logRouterInitializationReplies;
		const Version startVersion = oldLogSystem->logRouterTags == 0
		                                 ? oldLogSystem->recoverAt.get() + 1
		                                 : std::max(self->tLogs[0]->startVersion, logSet->startVersion);
		for (int i = 0; i < self->logRouterTags; i++) {
			InitializeLogRouterRequest req;
			req.recoveryCount = recoveryCount;
			req.routerTag = Tag(tagLocalityLogRouter, i);
			req.startVersion = startVersion;
			req.tLogLocalities = localities;
			req.tLogPolicy = logSet->tLogPolicy;
			req.locality = remoteLocality;
			logRouterInitializationReplies.push_back(transformErrors(
			    throwErrorOr(
			        remoteWorkers.logRouters[i % remoteWorkers.logRouters.size()].logRouter.getReplyUnlessFailedFor(
			            req, SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
			    master_recovery_failed()));
		}

		std::vector<Tag> localTags = getLocalTags(remoteLocality, allTags);
		LogSystemConfig oldLogSystemConfig = oldLogSystem->getLogSystemConfig();

		logSet->tLogLocalities.resize(remoteWorkers.remoteTLogs.size());
		logSet->logServers.resize(
		    remoteWorkers.remoteTLogs
		        .size()); // Dummy interfaces, so that logSystem->getPushLocations() below uses the correct size
		logSet->updateLocalitySet(localities);

		state vector<Future<TLogInterface>> remoteTLogInitializationReplies;
		vector<InitializeTLogRequest> remoteTLogReqs(remoteWorkers.remoteTLogs.size());

		bool nonShardedTxs = self->getTLogVersion() < TLogVersion::V4;
		if (oldLogSystem->logRouterTags == 0) {
			std::vector<int> locations;
			for (Tag tag : localTags) {
				locations.clear();
				logSet->getPushLocations(VectorRef<Tag>(&tag, 1), locations, 0);
				for (int loc : locations)
					remoteTLogReqs[loc].recoverTags.push_back(tag);
			}

			if (oldLogSystem->tLogs.size()) {
				int maxTxsTags = oldLogSystem->txsTags;
				bool needsOldTxs = oldLogSystem->tLogs[0]->tLogVersion < TLogVersion::V4;
				for (auto& it : oldLogSystem->oldLogData) {
					maxTxsTags = std::max<int>(maxTxsTags, it.txsTags);
					needsOldTxs = needsOldTxs || it.tLogs[0]->tLogVersion < TLogVersion::V4;
				}
				for (int i = needsOldTxs ? -1 : 0; i < maxTxsTags; i++) {
					Tag tag = i == -1 ? txsTag : Tag(tagLocalityTxs, i);
					Tag pushTag = (i == -1 || nonShardedTxs) ? txsTag : Tag(tagLocalityTxs, i % self->txsTags);
					locations.clear();
					logSet->getPushLocations(VectorRef<Tag>(&pushTag, 1), locations, 0);
					for (int loc : locations)
						remoteTLogReqs[loc].recoverTags.push_back(tag);
				}
			}
		}

		if (oldLogSystem->tLogs.size()) {
			if (nonShardedTxs) {
				localTags.push_back(txsTag);
			} else {
				for (int i = 0; i < self->txsTags; i++) {
					localTags.push_back(Tag(tagLocalityTxs, i));
				}
			}
		}

		for (int i = 0; i < remoteWorkers.remoteTLogs.size(); i++) {
			InitializeTLogRequest& req = remoteTLogReqs[i];
			req.recruitmentID = self->recruitmentID;
			req.logVersion = configuration.tLogVersion;
			req.storeType = configuration.tLogDataStoreType;
			req.spillType = configuration.tLogSpillType;
			req.recoverFrom = oldLogSystemConfig;
			req.recoverAt = oldLogSystem->recoverAt.get();
			req.knownCommittedVersion = oldLogSystem->knownCommittedVersion;
			req.epoch = recoveryCount;
			req.remoteTag = Tag(tagLocalityRemoteLog, i);
			req.locality = remoteLocality;
			req.isPrimary = false;
			req.allTags = localTags;
			req.startVersion = logSet->startVersion;
			req.logRouterTags = 0;
			req.txsTags = self->txsTags;
		}

		remoteTLogInitializationReplies.reserve(remoteWorkers.remoteTLogs.size());
		for (int i = 0; i < remoteWorkers.remoteTLogs.size(); i++)
			remoteTLogInitializationReplies.push_back(transformErrors(
			    throwErrorOr(remoteWorkers.remoteTLogs[i].tLog.getReplyUnlessFailedFor(
			        remoteTLogReqs[i], SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
			    master_recovery_failed()));

		TraceEvent("RemoteLogRecruitment_InitializingRemoteLogs")
		    .detail("StartVersion", logSet->startVersion)
		    .detail("LocalStart", self->tLogs[0]->startVersion)
		    .detail("LogRouterTags", self->logRouterTags);
		wait(waitForAll(remoteTLogInitializationReplies) && waitForAll(logRouterInitializationReplies) &&
		     oldRouterRecruitment);

		for (int i = 0; i < logRouterInitializationReplies.size(); i++) {
			logSet->logRouters.push_back(makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
			    OptionalInterface<TLogInterface>(logRouterInitializationReplies[i].get())));
		}

		for (int i = 0; i < remoteTLogInitializationReplies.size(); i++) {
			logSet->logServers[i] = makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
			    OptionalInterface<TLogInterface>(remoteTLogInitializationReplies[i].get()));
			logSet->tLogLocalities[i] = remoteWorkers.remoteTLogs[i].locality;
		}
		filterLocalityDataForPolicy(logSet->tLogPolicy, &logSet->tLogLocalities);

		std::vector<Future<Void>> recoveryComplete;
		recoveryComplete.reserve(logSet->logServers.size());
		for (int i = 0; i < logSet->logServers.size(); i++)
			recoveryComplete.push_back(transformErrors(
			    throwErrorOr(logSet->logServers[i]->get().interf().recoveryFinished.getReplyUnlessFailedFor(
			        TLogRecoveryFinishedRequest(),
			        SERVER_KNOBS->TLOG_TIMEOUT,
			        SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
			    master_recovery_failed()));

		self->remoteRecoveryComplete = waitForAll(recoveryComplete);
		self->tLogs.push_back(logSet);
		TraceEvent("RemoteLogRecruitment_CompletingRecovery");
		return Void();
	}

	ACTOR static Future<Reference<ILogSystem>> newEpoch(Reference<TagPartitionedLogSystem> oldLogSystem,
	                                                    RecruitFromConfigurationReply recr,
	                                                    Future<RecruitRemoteFromConfigurationReply> fRemoteWorkers,
	                                                    DatabaseConfiguration configuration,
	                                                    LogEpoch recoveryCount,
	                                                    int8_t primaryLocality,
	                                                    int8_t remoteLocality,
	                                                    std::vector<Tag> allTags,
	                                                    Reference<AsyncVar<bool>> recruitmentStalled) {
		state double startTime = now();
		state Reference<TagPartitionedLogSystem> logSystem(
		    new TagPartitionedLogSystem(oldLogSystem->getDebugID(), oldLogSystem->locality, recoveryCount));
		logSystem->logSystemType = LogSystemType::tagPartitioned;
		logSystem->expectedLogSets = 1;
		logSystem->recoveredAt = oldLogSystem->recoverAt;
		logSystem->repopulateRegionAntiQuorum = configuration.repopulateRegionAntiQuorum;
		logSystem->recruitmentID = deterministicRandom()->randomUniqueID();
		logSystem->txsTags = configuration.tLogVersion >= TLogVersion::V4 ? recr.tLogs.size() : 0;
		oldLogSystem->recruitmentID = logSystem->recruitmentID;

		if (configuration.usableRegions > 1) {
			logSystem->logRouterTags =
			    recr.tLogs.size() *
			    std::max<int>(1, configuration.desiredLogRouterCount / std::max<int>(1, recr.tLogs.size()));
			logSystem->expectedLogSets++;
			logSystem->addPseudoLocality(tagLocalityLogRouterMapped);
			TraceEvent e("AddPseudoLocality", logSystem->getDebugID());
			e.detail("Locality1", "LogRouterMapped");
			if (configuration.backupWorkerEnabled) {
				logSystem->addPseudoLocality(tagLocalityBackup);
				e.detail("Locality2", "Backup");
			}
		} else if (configuration.backupWorkerEnabled) {
			// Single region uses log router tag for backup workers.
			logSystem->logRouterTags =
			    recr.tLogs.size() *
			    std::max<int>(1, configuration.desiredLogRouterCount / std::max<int>(1, recr.tLogs.size()));
			logSystem->addPseudoLocality(tagLocalityBackup);
			TraceEvent("AddPseudoLocality", logSystem->getDebugID()).detail("Locality", "Backup");
		}

		logSystem->tLogs.push_back(makeReference<LogSet>());
		logSystem->tLogs[0]->tLogVersion = configuration.tLogVersion;
		logSystem->tLogs[0]->tLogWriteAntiQuorum = configuration.tLogWriteAntiQuorum;
		logSystem->tLogs[0]->tLogReplicationFactor = configuration.tLogReplicationFactor;
		logSystem->tLogs[0]->tLogPolicy = configuration.tLogPolicy;
		logSystem->tLogs[0]->isLocal = true;
		logSystem->tLogs[0]->locality = primaryLocality;

		state RegionInfo region = configuration.getRegion(recr.dcId);

		state int maxTxsTags = oldLogSystem->txsTags;
		state bool needsOldTxs = oldLogSystem->tLogs.size() && oldLogSystem->getTLogVersion() < TLogVersion::V4;
		for (auto& it : oldLogSystem->oldLogData) {
			maxTxsTags = std::max<int>(maxTxsTags, it.txsTags);
			needsOldTxs = needsOldTxs || it.tLogs[0]->tLogVersion < TLogVersion::V4;
		}

		if (region.satelliteTLogReplicationFactor > 0 && configuration.usableRegions > 1) {
			logSystem->tLogs.push_back(makeReference<LogSet>());
			if (recr.satelliteFallback) {
				logSystem->tLogs[1]->tLogWriteAntiQuorum = region.satelliteTLogWriteAntiQuorumFallback;
				logSystem->tLogs[1]->tLogReplicationFactor = region.satelliteTLogReplicationFactorFallback;
				logSystem->tLogs[1]->tLogPolicy = region.satelliteTLogPolicyFallback;
			} else {
				logSystem->tLogs[1]->tLogWriteAntiQuorum = region.satelliteTLogWriteAntiQuorum;
				logSystem->tLogs[1]->tLogReplicationFactor = region.satelliteTLogReplicationFactor;
				logSystem->tLogs[1]->tLogPolicy = region.satelliteTLogPolicy;
			}
			logSystem->tLogs[1]->isLocal = true;
			logSystem->tLogs[1]->locality = tagLocalitySatellite;
			logSystem->tLogs[1]->tLogVersion = configuration.tLogVersion;
			logSystem->tLogs[1]->startVersion = oldLogSystem->knownCommittedVersion + 1;

			logSystem->tLogs[1]->tLogLocalities.resize(recr.satelliteTLogs.size());
			for (int i = 0; i < recr.satelliteTLogs.size(); i++) {
				logSystem->tLogs[1]->tLogLocalities[i] = recr.satelliteTLogs[i].locality;
			}
			filterLocalityDataForPolicy(logSystem->tLogs[1]->tLogPolicy, &logSystem->tLogs[1]->tLogLocalities);

			logSystem->tLogs[1]->logServers.resize(
			    recr.satelliteTLogs
			        .size()); // Dummy interfaces, so that logSystem->getPushLocations() below uses the correct size
			logSystem->tLogs[1]->updateLocalitySet(logSystem->tLogs[1]->tLogLocalities);
			logSystem->tLogs[1]->populateSatelliteTagLocations(
			    logSystem->logRouterTags, oldLogSystem->logRouterTags, logSystem->txsTags, maxTxsTags);
			logSystem->expectedLogSets++;
		}

		if (oldLogSystem->tLogs.size()) {
			logSystem->oldLogData.emplace_back();
			logSystem->oldLogData[0].tLogs = oldLogSystem->tLogs;
			logSystem->oldLogData[0].epochBegin = oldLogSystem->tLogs[0]->startVersion;
			logSystem->oldLogData[0].epochEnd = oldLogSystem->knownCommittedVersion + 1;
			logSystem->oldLogData[0].logRouterTags = oldLogSystem->logRouterTags;
			logSystem->oldLogData[0].txsTags = oldLogSystem->txsTags;
			logSystem->oldLogData[0].pseudoLocalities = oldLogSystem->pseudoLocalities;
			logSystem->oldLogData[0].epoch = oldLogSystem->epoch;
		}
		logSystem->oldLogData.insert(
		    logSystem->oldLogData.end(), oldLogSystem->oldLogData.begin(), oldLogSystem->oldLogData.end());

		logSystem->tLogs[0]->startVersion = oldLogSystem->knownCommittedVersion + 1;
		logSystem->backupStartVersion = oldLogSystem->knownCommittedVersion + 1;
		state int lockNum = 0;
		while (lockNum < oldLogSystem->lockResults.size()) {
			if (oldLogSystem->lockResults[lockNum].logSet->locality == primaryLocality) {
				if (oldLogSystem->lockResults[lockNum].isCurrent &&
				    oldLogSystem->lockResults[lockNum].logSet->isLocal) {
					break;
				}
				state Future<Void> stalledAfter = setAfter(recruitmentStalled, SERVER_KNOBS->MAX_RECOVERY_TIME, true);
				loop {
					auto versions = TagPartitionedLogSystem::getDurableVersion(logSystem->dbgid,
					                                                           oldLogSystem->lockResults[lockNum]);
					if (versions.present()) {
						logSystem->tLogs[0]->startVersion =
						    std::min(std::min(versions.get().first + 1, oldLogSystem->lockResults[lockNum].epochEnd),
						             logSystem->tLogs[0]->startVersion);
						break;
					}
					wait(TagPartitionedLogSystem::getDurableVersionChanged(oldLogSystem->lockResults[lockNum]));
				}
				stalledAfter.cancel();
				break;
			}
			lockNum++;
		}

		vector<LocalityData> localities;
		localities.resize(recr.tLogs.size());
		for (int i = 0; i < recr.tLogs.size(); i++) {
			localities[i] = recr.tLogs[i].locality;
		}

		state Future<Void> oldRouterRecruitment = Never();
		TraceEvent("NewEpochStartVersion", oldLogSystem->getDebugID())
		    .detail("StartVersion", logSystem->tLogs[0]->startVersion)
		    .detail("EpochEnd", oldLogSystem->knownCommittedVersion + 1)
		    .detail("Locality", primaryLocality)
		    .detail("OldLogRouterTags", oldLogSystem->logRouterTags);
		if (oldLogSystem->logRouterTags > 0 ||
		    logSystem->tLogs[0]->startVersion < oldLogSystem->knownCommittedVersion + 1) {
			oldRouterRecruitment = TagPartitionedLogSystem::recruitOldLogRouters(oldLogSystem.getPtr(),
			                                                                     recr.oldLogRouters,
			                                                                     recoveryCount,
			                                                                     primaryLocality,
			                                                                     logSystem->tLogs[0]->startVersion,
			                                                                     localities,
			                                                                     logSystem->tLogs[0]->tLogPolicy,
			                                                                     false);
			if (oldLogSystem->knownCommittedVersion - logSystem->tLogs[0]->startVersion >
			    SERVER_KNOBS->MAX_RECOVERY_VERSIONS) {
				// make sure we can recover in the other DC.
				for (auto& lockResult : oldLogSystem->lockResults) {
					if (lockResult.logSet->locality == remoteLocality) {
						if (TagPartitionedLogSystem::getDurableVersion(logSystem->dbgid, lockResult).present()) {
							recruitmentStalled->set(true);
						}
					}
				}
			}
		} else {
			oldLogSystem->logSystemConfigChanged.trigger();
		}

		std::vector<Tag> localTags = getLocalTags(primaryLocality, allTags);
		state LogSystemConfig oldLogSystemConfig = oldLogSystem->getLogSystemConfig();

		state vector<Future<TLogInterface>> initializationReplies;
		vector<InitializeTLogRequest> reqs(recr.tLogs.size());

		logSystem->tLogs[0]->tLogLocalities.resize(recr.tLogs.size());
		logSystem->tLogs[0]->logServers.resize(
		    recr.tLogs.size()); // Dummy interfaces, so that logSystem->getPushLocations() below uses the correct size
		logSystem->tLogs[0]->updateLocalitySet(localities);

		std::vector<int> locations;
		for (Tag tag : localTags) {
			locations.clear();
			logSystem->tLogs[0]->getPushLocations(VectorRef<Tag>(&tag, 1), locations, 0);
			for (int loc : locations)
				reqs[loc].recoverTags.push_back(tag);
		}
		for (int i = 0; i < oldLogSystem->logRouterTags; i++) {
			Tag tag = Tag(tagLocalityLogRouter, i);
			reqs[logSystem->tLogs[0]->bestLocationFor(tag)].recoverTags.push_back(tag);
		}
		bool nonShardedTxs = logSystem->getTLogVersion() < TLogVersion::V4;
		if (oldLogSystem->tLogs.size()) {
			for (int i = needsOldTxs ? -1 : 0; i < maxTxsTags; i++) {
				Tag tag = i == -1 ? txsTag : Tag(tagLocalityTxs, i);
				Tag pushTag = (i == -1 || nonShardedTxs) ? txsTag : Tag(tagLocalityTxs, i % logSystem->txsTags);
				locations.clear();
				logSystem->tLogs[0]->getPushLocations(VectorRef<Tag>(&pushTag, 1), locations, 0);
				for (int loc : locations)
					reqs[loc].recoverTags.push_back(tag);
			}
			if (nonShardedTxs) {
				localTags.push_back(txsTag);
			} else {
				for (int i = 0; i < logSystem->txsTags; i++) {
					localTags.push_back(Tag(tagLocalityTxs, i));
				}
			}
		}

		for (int i = 0; i < recr.tLogs.size(); i++) {
			InitializeTLogRequest& req = reqs[i];
			req.recruitmentID = logSystem->recruitmentID;
			req.logVersion = configuration.tLogVersion;
			req.storeType = configuration.tLogDataStoreType;
			req.spillType = configuration.tLogSpillType;
			req.recoverFrom = oldLogSystemConfig;
			req.recoverAt = oldLogSystem->recoverAt.get();
			req.knownCommittedVersion = oldLogSystem->knownCommittedVersion;
			req.epoch = recoveryCount;
			req.locality = primaryLocality;
			req.remoteTag = Tag(tagLocalityRemoteLog, i);
			req.isPrimary = true;
			req.allTags = localTags;
			req.startVersion = logSystem->tLogs[0]->startVersion;
			req.logRouterTags = logSystem->logRouterTags;
			req.txsTags = logSystem->txsTags;
		}

		initializationReplies.reserve(recr.tLogs.size());
		for (int i = 0; i < recr.tLogs.size(); i++)
			initializationReplies.push_back(transformErrors(
			    throwErrorOr(recr.tLogs[i].tLog.getReplyUnlessFailedFor(
			        reqs[i], SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
			    master_recovery_failed()));

		state std::vector<Future<Void>> recoveryComplete;

		if (region.satelliteTLogReplicationFactor > 0 && configuration.usableRegions > 1) {
			state vector<Future<TLogInterface>> satelliteInitializationReplies;
			vector<InitializeTLogRequest> sreqs(recr.satelliteTLogs.size());
			std::vector<Tag> satelliteTags;

			if (logSystem->logRouterTags) {
				for (int i = 0; i < oldLogSystem->logRouterTags; i++) {
					Tag tag = Tag(tagLocalityLogRouter, i);
					// Satellite logs will index a mutation with tagLocalityLogRouter with an id greater than
					// the number of log routers as having an id mod the number of log routers.  We thus need
					// to make sure that if we're going from more log routers in the previous generation to
					// less log routers in the newer one, that we map the log router tags onto satellites that
					// are the preferred location for id%logRouterTags.
					Tag pushLocation = Tag(tagLocalityLogRouter, i % logSystem->logRouterTags);
					locations.clear();
					logSystem->tLogs[1]->getPushLocations(VectorRef<Tag>(&pushLocation, 1), locations, 0);
					for (int loc : locations)
						sreqs[loc].recoverTags.push_back(tag);
				}
			}
			if (oldLogSystem->tLogs.size()) {
				for (int i = needsOldTxs ? -1 : 0; i < maxTxsTags; i++) {
					Tag tag = i == -1 ? txsTag : Tag(tagLocalityTxs, i);
					Tag pushTag = (i == -1 || nonShardedTxs) ? txsTag : Tag(tagLocalityTxs, i % logSystem->txsTags);
					locations.clear();
					logSystem->tLogs[1]->getPushLocations(VectorRef<Tag>(&pushTag, 1), locations, 0);
					for (int loc : locations)
						sreqs[loc].recoverTags.push_back(tag);
				}
				if (nonShardedTxs) {
					satelliteTags.push_back(txsTag);
				} else {
					for (int i = 0; i < logSystem->txsTags; i++) {
						satelliteTags.push_back(Tag(tagLocalityTxs, i));
					}
				}
			}

			for (int i = 0; i < recr.satelliteTLogs.size(); i++) {
				InitializeTLogRequest& req = sreqs[i];
				req.recruitmentID = logSystem->recruitmentID;
				req.logVersion = configuration.tLogVersion;
				req.storeType = configuration.tLogDataStoreType;
				req.spillType = configuration.tLogSpillType;
				req.recoverFrom = oldLogSystemConfig;
				req.recoverAt = oldLogSystem->recoverAt.get();
				req.knownCommittedVersion = oldLogSystem->knownCommittedVersion;
				req.epoch = recoveryCount;
				req.locality = tagLocalitySatellite;
				req.remoteTag = Tag();
				req.isPrimary = true;
				req.allTags = satelliteTags;
				req.startVersion = oldLogSystem->knownCommittedVersion + 1;
				req.logRouterTags = logSystem->logRouterTags;
				req.txsTags = logSystem->txsTags;
			}

			satelliteInitializationReplies.reserve(recr.satelliteTLogs.size());
			for (int i = 0; i < recr.satelliteTLogs.size(); i++)
				satelliteInitializationReplies.push_back(transformErrors(
				    throwErrorOr(recr.satelliteTLogs[i].tLog.getReplyUnlessFailedFor(
				        sreqs[i], SERVER_KNOBS->TLOG_TIMEOUT, SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
				    master_recovery_failed()));

			wait(waitForAll(satelliteInitializationReplies) || oldRouterRecruitment);

			for (int i = 0; i < satelliteInitializationReplies.size(); i++) {
				logSystem->tLogs[1]->logServers[i] = makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
				    OptionalInterface<TLogInterface>(satelliteInitializationReplies[i].get()));
			}

			for (int i = 0; i < logSystem->tLogs[1]->logServers.size(); i++)
				recoveryComplete.push_back(transformErrors(
				    throwErrorOr(
				        logSystem->tLogs[1]->logServers[i]->get().interf().recoveryFinished.getReplyUnlessFailedFor(
				            TLogRecoveryFinishedRequest(),
				            SERVER_KNOBS->TLOG_TIMEOUT,
				            SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
				    master_recovery_failed()));
		}

		wait(waitForAll(initializationReplies) || oldRouterRecruitment);

		for (int i = 0; i < initializationReplies.size(); i++) {
			logSystem->tLogs[0]->logServers[i] = makeReference<AsyncVar<OptionalInterface<TLogInterface>>>(
			    OptionalInterface<TLogInterface>(initializationReplies[i].get()));
			logSystem->tLogs[0]->tLogLocalities[i] = recr.tLogs[i].locality;
		}
		filterLocalityDataForPolicy(logSystem->tLogs[0]->tLogPolicy, &logSystem->tLogs[0]->tLogLocalities);

		// Don't force failure of recovery if it took us a long time to recover. This avoids multiple long running
		// recoveries causing tests to timeout
		if (BUGGIFY && now() - startTime < 300 && g_network->isSimulated() && g_simulator.speedUpSimulation)
			throw master_recovery_failed();

		for (int i = 0; i < logSystem->tLogs[0]->logServers.size(); i++)
			recoveryComplete.push_back(transformErrors(
			    throwErrorOr(
			        logSystem->tLogs[0]->logServers[i]->get().interf().recoveryFinished.getReplyUnlessFailedFor(
			            TLogRecoveryFinishedRequest(),
			            SERVER_KNOBS->TLOG_TIMEOUT,
			            SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY)),
			    master_recovery_failed()));
		logSystem->recoveryComplete = waitForAll(recoveryComplete);

		if (configuration.usableRegions > 1) {
			logSystem->hasRemoteServers = true;
			logSystem->remoteRecovery = TagPartitionedLogSystem::newRemoteEpoch(logSystem.getPtr(),
			                                                                    oldLogSystem,
			                                                                    fRemoteWorkers,
			                                                                    configuration,
			                                                                    recoveryCount,
			                                                                    remoteLocality,
			                                                                    allTags);
			if (oldLogSystem->tLogs.size() > 0 && oldLogSystem->tLogs[0]->locality == tagLocalitySpecial) {
				// The wait is required so that we know both primary logs and remote logs have copied the data between
				// the known committed version and the recovery version.
				// FIXME: we can remove this wait once we are able to have log routers which can ship data to the remote
				// logs without using log router tags.
				wait(logSystem->remoteRecovery);
			}
		} else {
			logSystem->hasRemoteServers = false;
			logSystem->remoteRecovery = logSystem->recoveryComplete;
			logSystem->remoteRecoveryComplete = logSystem->recoveryComplete;
		}

		return logSystem;
	}

	ACTOR static Future<Void> trackRejoins(
	    UID dbgid,
	    std::vector<std::pair<Reference<AsyncVar<OptionalInterface<TLogInterface>>>, Reference<IReplicationPolicy>>>
	        logServers,
	    FutureStream<struct TLogRejoinRequest> rejoinRequests) {
		state std::map<UID, ReplyPromise<TLogRejoinReply>> lastReply;
		state std::set<UID> logsWaiting;
		state double startTime = now();
		state Future<Void> warnTimeout = delay(SERVER_KNOBS->TLOG_SLOW_REJOIN_WARN_TIMEOUT_SECS);

		for (const auto& log : logServers) {
			logsWaiting.insert(log.first->get().id());
		}

		try {
			loop choose {
				when(TLogRejoinRequest req = waitNext(rejoinRequests)) {
					int pos = -1;
					for (int i = 0; i < logServers.size(); i++) {
						if (logServers[i].first->get().id() == req.myInterface.id()) {
							pos = i;
							logsWaiting.erase(logServers[i].first->get().id());
							break;
						}
					}
					if (pos != -1) {
						TraceEvent("TLogJoinedMe", dbgid)
						    .detail("TLog", req.myInterface.id())
						    .detail("Address", req.myInterface.commit.getEndpoint().getPrimaryAddress().toString());
						if (!logServers[pos].first->get().present() ||
						    req.myInterface.commit.getEndpoint() !=
						        logServers[pos].first->get().interf().commit.getEndpoint()) {
							TLogInterface interf = req.myInterface;
							filterLocalityDataForPolicyDcAndProcess(logServers[pos].second, &interf.filteredLocality);
							logServers[pos].first->setUnconditional(OptionalInterface<TLogInterface>(interf));
						}
						lastReply[req.myInterface.id()].send(TLogRejoinReply{ false });
						lastReply[req.myInterface.id()] = req.reply;
					} else {
						TraceEvent("TLogJoinedMeUnknown", dbgid)
						    .detail("TLog", req.myInterface.id())
						    .detail("Address", req.myInterface.commit.getEndpoint().getPrimaryAddress().toString());
						req.reply.send(true);
					}
				}
				when(wait(warnTimeout)) {
					for (const auto& logId : logsWaiting) {
						TraceEvent(SevWarnAlways, "TLogRejoinSlow", dbgid)
						    .detail("Elapsed", startTime - now())
						    .detail("LogId", logId);
					}
					warnTimeout = Never();
				}
			}
		} catch (...) {
			for (auto it = lastReply.begin(); it != lastReply.end(); ++it)
				it->second.send(TLogRejoinReply{ true });
			throw;
		}
	}

	ACTOR static Future<TLogLockResult> lockTLog(UID myID, Reference<AsyncVar<OptionalInterface<TLogInterface>>> tlog) {
		TraceEvent("TLogLockStarted", myID).detail("TLog", tlog->get().id());
		loop {
			choose {
				when(TLogLockResult data =
				         wait(tlog->get().present()
				                  ? brokenPromiseToNever(tlog->get().interf().lock.getReply<TLogLockResult>())
				                  : Never())) {
					TraceEvent("TLogLocked", myID).detail("TLog", tlog->get().id()).detail("End", data.end);
					return data;
				}
				when(wait(tlog->onChange())) {}
			}
		}
	}

	// FIXME: disabled during merge, update and use in epochEnd()
	/*
	static void lockMinimalTLogSet(const UID& dbgid, const DBCoreState& prevState,
	                               const std::vector<Reference<AsyncVar<OptionalInterface<TLogInterface>>>>& logServers,
	                               const std::vector<Reference<AsyncVar<bool>>>& logFailed,
	                               vector<Future<TLogLockResult>>* tLogReply ) {
	    // Invariant: tLogReply[i] must correspond to the tlog stored as logServers[i].
	    ASSERT(tLogReply->size() == prevState.tLogLocalities.size());
	    ASSERT(logFailed.size() == tLogReply->size());

	    // For any given index, only one of the following will be true.
	    auto locking_completed = [&logFailed, tLogReply](int index) {
	        const auto& entry = tLogReply->at(index);
	        return !logFailed[index]->get() && entry.isValid() && entry.isReady() && !entry.isError();
	    };
	    auto locking_failed = [&logFailed, tLogReply](int index) {
	        const auto& entry = tLogReply->at(index);
	        return logFailed[index]->get() || (entry.isValid() && entry.isReady() && entry.isError());
	    };
	    auto locking_pending = [&logFailed, tLogReply](int index) {
	        const auto& entry = tLogReply->at(index);
	        return !logFailed[index]->get() && (entry.isValid() && !entry.isReady());
	    };
	    auto locking_skipped = [&logFailed, tLogReply](int index) {
	        const auto& entry = tLogReply->at(index);
	        return !logFailed[index]->get() && !entry.isValid();
	    };

	    auto can_obtain_quorum = [&prevState](std::function<bool(int)> filter) {
	        LocalityGroup filter_true;
	        std::vector<LocalityData> filter_false, unused;
	        for (int i = 0; i < prevState.tLogLocalities.size() ; i++) {
	            if (filter(i)) {
	                filter_true.add(prevState.tLogLocalities[i]);
	            } else {
	                filter_false.push_back(prevState.tLogLocalities[i]);
	            }
	        }
	        bool valid = filter_true.validate(prevState.tLogPolicy);
	        if (!valid && prevState.tLogWriteAntiQuorum > 0 ) {
	            valid = !validateAllCombinations(unused, filter_true, prevState.tLogPolicy, filter_false,
	prevState.tLogWriteAntiQuorum, false);
	        }
	        return valid;
	    };

	    // Step 1: Verify that if all the failed TLogs come back, they can't form a quorum.
	    if (can_obtain_quorum(locking_failed)) {
	        TraceEvent(SevInfo, "MasterRecoveryTLogLockingImpossible", dbgid);
	        return;
	    }

	    // Step 2: It's possible for us to succeed, but we need to lock additional logs.
	    //
	    // First, we need an accurate picture of what TLogs we're capable of locking. We can't tell the
	    // difference between a temporarily failed TLog and a permanently failed TLog. Thus, we assume
	    // all failures are permanent, and manually re-issue lock requests if they rejoin.
	    for (int i = 0; i < logFailed.size(); i++) {
	        const auto& r = tLogReply->at(i);
	        TEST(locking_failed(i) && (r.isValid() && !r.isReady()));  // A TLog failed with a pending request.
	        // The reboot_a_tlog BUGGIFY below should cause the above case to be hit.
	        if (locking_failed(i)) {
	            tLogReply->at(i) = Future<TLogLockResult>();
	        }
	    }

	    // We're trying to paritition the set of old tlogs into two sets, L and R, such that:
	    // (1). R does not validate the policy
	    // (2). |R| is as large as possible
	    // (3). L contains all the already-locked TLogs
	    // and then we only issue lock requests to TLogs in L. This is safe, as R does not have quorum,
	    // so no commits may occur.  It does not matter if L forms a quorum or not.
	    //
	    // We form these sets by starting with L as all machines and R as the empty set, and moving a
	    // random machine from L to R until (1) or (2) no longer holds as true. Code-wise, L is
	    // [0..end-can_omit), and R is [end-can_omit..end), and we move a random machine via randomizing
	    // the order of the tlogs. Choosing a random machine was verified to generate a good-enough
	    // result to be interesting intests sufficiently frequently that we don't need to try to
	    // calculate the exact optimal solution.
	    std::vector<std::pair<LocalityData, int>> tlogs;
	    for (int i = 0; i < prevState.tLogLocalities.size(); i++) {
	        tlogs.emplace_back(prevState.tLogLocalities[i], i);
	    }
	    deterministicRandom()->randomShuffle(tlogs);
	    // Rearrange the array such that things that the left is logs closer to being locked, and
	    // the right is logs that can't be locked.  This makes us prefer locking already-locked TLogs,
	    // which is how we respect the decisions made in the previous execution.
	    auto idx_to_order = [&locking_completed, &locking_failed, &locking_pending, &locking_skipped](int index) {
	        bool complete = locking_completed(index);
	        bool pending = locking_pending(index);
	        bool skipped = locking_skipped(index);
	        bool failed = locking_failed(index);

	        ASSERT( complete + pending + skipped + failed == 1 );

	        if (complete) return 0;
	        if (pending) return 1;
	        if (skipped) return 2;
	        if (failed) return 3;

	        ASSERT(false);  // Programmer error.
	        return -1;
	    };
	    std::sort(tlogs.begin(), tlogs.end(),
	        // TODO: Change long type to `auto` once toolchain supports C++17.
	        [&idx_to_order](const std::pair<LocalityData, int>& lhs, const std::pair<LocalityData, int>& rhs) {
	            return idx_to_order(lhs.second) < idx_to_order(rhs.second);
	        });

	    // Indexes that aren't in the vector are the ones we're considering omitting. Remove indexes until
	    // the removed set forms a quorum.
	    int can_omit = 0;
	    std::vector<int> to_lock_indexes;
	    for (auto it = tlogs.cbegin() ; it != tlogs.cend() - 1 ; it++ ) {
	        to_lock_indexes.push_back(it->second);
	    }
	    auto filter = [&to_lock_indexes](int index) {
	        return std::find(to_lock_indexes.cbegin(), to_lock_indexes.cend(), index) == to_lock_indexes.cend();
	    };
	    while(true) {
	        if (can_obtain_quorum(filter)) {
	            break;
	        } else {
	            can_omit++;
	            ASSERT(can_omit < tlogs.size());
	            to_lock_indexes.pop_back();
	        }
	    }

	    if (prevState.tLogReplicationFactor - prevState.tLogWriteAntiQuorum == 1) {
	        ASSERT(can_omit == 0);
	    }
	    // Our previous check of making sure there aren't too many failed logs should have prevented this.
	    ASSERT(!locking_failed(tlogs[tlogs.size()-can_omit-1].second));

	    // If we've managed to leave more tlogs unlocked than (RF-AQ), it means we've hit the case
	    // where the policy engine has allowed us to have multiple logs in the same failure domain
	    // with independant sets of data. This case will validated that no code is relying on the old
	    // quorum=(RF-AQ) logic, and now goes through the policy engine instead.
	    TEST(can_omit >= prevState.tLogReplicationFactor - prevState.tLogWriteAntiQuorum);  // Locking a subset of the
	TLogs while ending an epoch. const bool reboot_a_tlog = g_network->now() - g_simulator.lastConnectionFailure >
	g_simulator.connectionFailuresDisableDuration && BUGGIFY && deterministicRandom()->random01() < 0.25;
	    TraceEvent(SevInfo, "MasterRecoveryTLogLocking", dbgid)
	        detail("Locks", tlogs.size() - can_omit)
	        detail("Skipped", can_omit)
	        detail("Replication", prevState.tLogReplicationFactor)
	        detail("Antiquorum", prevState.tLogWriteAntiQuorum)
	        detail("RebootBuggify", reboot_a_tlog);
	    for (int i = 0; i < tlogs.size() - can_omit; i++) {
	        const int index = tlogs[i].second;
	        Future<TLogLockResult>& entry = tLogReply->at(index);
	        if (!entry.isValid()) {
	            entry = lockTLog( dbgid, logServers[index] );
	        }
	    }
	    if (reboot_a_tlog) {
	        g_simulator.lastConnectionFailure = g_network->now();
	        for (int i = 0; i < tlogs.size() - can_omit; i++) {
	            const int index = tlogs[i].second;
	            if (logServers[index]->get().present()) {
	                g_simulator.rebootProcess(
	                    g_simulator.getProcessByAddress(
	                        logServers[index]->get().interf().address()),
	                    ISimulator::RebootProcess);
	                break;
	            }
	        }
	    }
	    // Intentionally leave `tlogs.size() - can_omit` .. `tlogs.size()` as !isValid() Futures.
	}*/

	template <class T>
	static vector<T> getReadyNonError(vector<Future<T>> const& futures) {
		// Return the values of those futures which have (non-error) values ready
		std::vector<T> result;
		for (auto& f : futures)
			if (f.isReady() && !f.isError())
				result.push_back(f.get());
		return result;
	}

	struct sort_by_end {
		bool operator()(TLogLockResult const& a, TLogLockResult const& b) const { return a.end < b.end; }
	};
};

Future<Void> ILogSystem::recoverAndEndEpoch(Reference<AsyncVar<Reference<ILogSystem>>> const& outLogSystem,
                                            UID const& dbgid,
                                            DBCoreState const& oldState,
                                            FutureStream<TLogRejoinRequest> const& rejoins,
                                            LocalityData const& locality,
                                            bool* forceRecovery) {
	return TagPartitionedLogSystem::recoverAndEndEpoch(outLogSystem, dbgid, oldState, rejoins, locality, forceRecovery);
}

Reference<ILogSystem> ILogSystem::fromLogSystemConfig(UID const& dbgid,
                                                      struct LocalityData const& locality,
                                                      struct LogSystemConfig const& conf,
                                                      bool excludeRemote,
                                                      bool useRecoveredAt,
                                                      Optional<PromiseStream<Future<Void>>> addActor) {
	if (conf.logSystemType == LogSystemType::empty)
		return Reference<ILogSystem>();
	else if (conf.logSystemType == LogSystemType::tagPartitioned)
		return TagPartitionedLogSystem::fromLogSystemConfig(
		    dbgid, locality, conf, excludeRemote, useRecoveredAt, addActor);
	else
		throw internal_error();
}

Reference<ILogSystem> ILogSystem::fromOldLogSystemConfig(UID const& dbgid,
                                                         struct LocalityData const& locality,
                                                         struct LogSystemConfig const& conf) {
	if (conf.logSystemType == LogSystemType::empty)
		return Reference<ILogSystem>();
	else if (conf.logSystemType == LogSystemType::tagPartitioned)
		return TagPartitionedLogSystem::fromOldLogSystemConfig(dbgid, locality, conf);
	else
		throw internal_error();
}

Reference<ILogSystem> ILogSystem::fromServerDBInfo(UID const& dbgid,
                                                   ServerDBInfo const& dbInfo,
                                                   bool useRecoveredAt,
                                                   Optional<PromiseStream<Future<Void>>> addActor) {
	return fromLogSystemConfig(dbgid, dbInfo.myLocality, dbInfo.logSystemConfig, false, useRecoveredAt, addActor);
}