foundationdb/fdbserver/CoordinatedState.actor.cpp

/*
 * CoordinatedState.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 "fdbserver/CoordinatedState.h"
#include "fdbserver/CoordinationInterface.h"
#include "fdbserver/Knobs.h"
#include "flow/ActorCollection.h"
#include "fdbserver/LeaderElection.h"
#include "flow/actorcompiler.h" // has to be last include

ACTOR Future<GenerationRegReadReply> waitAndSendRead( RequestStream<GenerationRegReadRequest> to, GenerationRegReadRequest req ) {
	if( SERVER_KNOBS->BUGGIFY_ALL_COORDINATION || BUGGIFY )
		wait( delay( SERVER_KNOBS->BUGGIFIED_EVENTUAL_CONSISTENCY*deterministicRandom()->random01() ) );
	state GenerationRegReadReply reply = wait( retryBrokenPromise( to, req ) );
	if( SERVER_KNOBS->BUGGIFY_ALL_COORDINATION || BUGGIFY )
		wait( delay( SERVER_KNOBS->BUGGIFIED_EVENTUAL_CONSISTENCY*deterministicRandom()->random01() ) );
	return reply;
}

ACTOR Future<UniqueGeneration> waitAndSendWrite(RequestStream<GenerationRegWriteRequest> to, GenerationRegWriteRequest req) {
	if( SERVER_KNOBS->BUGGIFY_ALL_COORDINATION || BUGGIFY )
		wait( delay( SERVER_KNOBS->BUGGIFIED_EVENTUAL_CONSISTENCY*deterministicRandom()->random01() ) );
	state UniqueGeneration reply = wait( retryBrokenPromise( to, req ) );
	if( SERVER_KNOBS->BUGGIFY_ALL_COORDINATION || BUGGIFY )
		wait( delay( SERVER_KNOBS->BUGGIFIED_EVENTUAL_CONSISTENCY*deterministicRandom()->random01() ) );
	return reply;
}

ACTOR Future<GenerationRegReadReply> emptyToNever( Future<GenerationRegReadReply> f ) {
	state GenerationRegReadReply r = wait(f);
	if (r.gen.generation == 0)
		wait( Future<Void>(Never()) );
	return r;
}

ACTOR Future<GenerationRegReadReply> nonemptyToNever( Future<GenerationRegReadReply> f ) {
	state GenerationRegReadReply r = wait(f);
	if (r.gen.generation != 0)
		wait( Future<Void>(Never()) );
	return r;
}

struct CoordinatedStateImpl {
	ServerCoordinators coordinators;
	int stage;
	UniqueGeneration gen;
	uint64_t conflictGen;
	bool doomed;
	ActorCollection ac; //Errors are not reported
	bool initial;

	CoordinatedStateImpl( ServerCoordinators const& c ) : coordinators(c), stage(0), conflictGen(0), doomed(false), ac(false), initial(false) {}
	uint64_t getConflict() { return conflictGen; }

	bool isDoomed( GenerationRegReadReply const& rep ) {
		return rep.gen > gen // setExclusive is doomed, because there was a write at least started at a higher generation, which means a read completed at that higher generation
			// || rep.rgen > gen // setExclusive isn't absolutely doomed, but it may/probably will fail
			;
	}

	ACTOR static Future<Value> read( CoordinatedStateImpl* self ) {
		ASSERT( self->stage == 0 );

		{
			self->stage = 1;
			GenerationRegReadReply rep = wait( self->replicatedRead( self, GenerationRegReadRequest( self->coordinators.clusterKey, UniqueGeneration() ) ) );
			self->conflictGen = std::max( self->conflictGen, std::max(rep.gen.generation, rep.rgen.generation) ) + 1;
			self->gen = UniqueGeneration( self->conflictGen, deterministicRandom()->randomUniqueID() );
		}

		{
			self->stage = 2;
			GenerationRegReadReply rep = wait( self->replicatedRead( self, GenerationRegReadRequest( self->coordinators.clusterKey, self->gen ) ) );
			self->stage = 3;
			self->conflictGen = std::max(self->conflictGen, std::max( rep.gen.generation, rep.rgen.generation ));
			if (self->isDoomed(rep))
				self->doomed = true;
			self->initial = rep.gen.generation == 0;

			self->stage = 4;
			return rep.value.present() ? rep.value.get() : Value();
		}
	}
	ACTOR static Future<Void> onConflict( CoordinatedStateImpl* self ) {
		ASSERT( self->stage == 4 );
		if (self->doomed) return Void();
		loop {
			wait( delay( SERVER_KNOBS->COORDINATED_STATE_ONCONFLICT_POLL_INTERVAL ) );
			GenerationRegReadReply rep = wait( self->replicatedRead( self, GenerationRegReadRequest( self->coordinators.clusterKey, UniqueGeneration() ) ) );
			if (self->stage > 4) break;
			self->conflictGen = std::max(self->conflictGen, std::max( rep.gen.generation, rep.rgen.generation ));
			if (self->isDoomed(rep))
				return Void();
		}
		wait( Future<Void>(Never()) );
		return Void();
	}
	ACTOR static Future<Void> setExclusive( CoordinatedStateImpl* self, Value v ) {
		ASSERT( self->stage == 4 );
		self->stage = 5;

		UniqueGeneration wgen = wait( self->replicatedWrite( self, GenerationRegWriteRequest( KeyValueRef(self->coordinators.clusterKey, v), self->gen ) ) );
		self->stage = 6;

		TraceEvent("CoordinatedStateSet").detail("Gen", self->gen.generation).detail("Wgen", wgen.generation)
			.detail("Genu", self->gen.uid).detail("Wgenu", wgen.uid)
			.detail("Cgen", self->conflictGen);

		if (wgen == self->gen)
			return Void();
		else {
			self->conflictGen = std::max(self->conflictGen, wgen.generation);
			throw coordinated_state_conflict();
		}
	}

	ACTOR static Future<GenerationRegReadReply> replicatedRead( CoordinatedStateImpl* self, GenerationRegReadRequest req ) {
		state std::vector<GenerationRegInterface> &replicas = self->coordinators.stateServers;
		state vector< Future<GenerationRegReadReply> > rep_empty_reply;
		state vector< Future<GenerationRegReadReply> > rep_reply;
		for(int i=0; i<replicas.size(); i++) {
			Future<GenerationRegReadReply> reply = waitAndSendRead( replicas[i].read, GenerationRegReadRequest(req.key, req.gen) );
			rep_empty_reply.push_back( nonemptyToNever( reply ) );
			rep_reply.push_back( emptyToNever( reply ) );
			self->ac.add( success( reply ) );
		}

		state Future<Void> majorityEmpty = quorum( rep_empty_reply, (replicas.size()+1)/2 ); //enough empty to ensure we cannot achieve a majority non-empty
		wait( quorum( rep_reply, replicas.size()/2 + 1 ) || majorityEmpty );

		if( majorityEmpty.isReady() ) {
			int best = -1;
			for(int i=0; i<rep_empty_reply.size(); i++) 
				if (rep_empty_reply[i].isReady() && !rep_empty_reply[i].isError()) {
					if (best < 0 || rep_empty_reply[i].get().rgen > rep_empty_reply[best].get().rgen )
						best = i;
				}
			ASSERT( best >= 0 );
			auto result = rep_empty_reply[best].get();
			return result;
		} else {
			int best = -1;
			for(int i=0; i<rep_reply.size(); i++) 
				if (rep_reply[i].isReady() && !rep_reply[i].isError()) {
					if (best < 0 || 
						rep_reply[i].get().gen > rep_reply[best].get().gen || 
						( rep_reply[i].get().gen == rep_reply[best].get().gen && rep_reply[i].get().rgen > rep_reply[best].get().rgen ) )
						best = i;
				}
			ASSERT( best >= 0 );
			auto result = rep_reply[best].get();
			return result;
		}
	}

	ACTOR static Future<UniqueGeneration> replicatedWrite( CoordinatedStateImpl* self, GenerationRegWriteRequest req ) {
		state std::vector<GenerationRegInterface> &replicas = self->coordinators.stateServers;
		state vector< Future<UniqueGeneration> > wrep_reply;
		for(int i=0; i<replicas.size(); i++) {
			Future<UniqueGeneration> reply = waitAndSendWrite( replicas[i].write, GenerationRegWriteRequest( req.kv, req.gen ) );
			wrep_reply.push_back( reply );
			self->ac.add( success( reply ) );
		}

		wait( quorum( wrep_reply, self->initial ? replicas.size() : replicas.size()/2 + 1 ) );

		UniqueGeneration maxGen;
		for(int i=0; i<wrep_reply.size(); i++)
			if (wrep_reply[i].isReady())
				maxGen = std::max(maxGen, wrep_reply[i].get());
		return maxGen;
	}
};

CoordinatedState::CoordinatedState( ServerCoordinators const& coord ) : impl( new CoordinatedStateImpl(coord) ) { }
CoordinatedState::~CoordinatedState() { delete impl; }
Future<Value> CoordinatedState::read() { return CoordinatedStateImpl::read(impl); }
Future<Void> CoordinatedState::onConflict() { return CoordinatedStateImpl::onConflict(impl); }
Future<Void> CoordinatedState::setExclusive(Value v) { return CoordinatedStateImpl::setExclusive(impl,v); }
uint64_t CoordinatedState::getConflict() { return impl->getConflict(); }

struct MovableValue {
	enum MoveState {
		MaybeTo = 1,
		Active = 2,
		MovingFrom = 3
	};
	
	Value value;
	int32_t mode;
	Optional<Value> other;   // a cluster connection string

	MovableValue() : mode( Active ) {}
	MovableValue( Value const& v, int mode, Optional<Value> other = Optional<Value>() ) : value( v ), mode( mode ), other( other ) {}

	template <class Ar>
	void serialize(Ar& ar) {
		ASSERT( ar.protocolVersion().hasMovableCoordinatedState() );
		serializer(ar, value, mode, other);
	}
};

struct MovableCoordinatedStateImpl {
	ServerCoordinators coordinators;
	CoordinatedState cs;
	Optional<Value> lastValue,  // The value passed to setExclusive()
		lastCSValue;       // The value passed to cs.setExclusive()

	MovableCoordinatedStateImpl( ServerCoordinators const& c ) : coordinators(c), cs(c) {}

	ACTOR static Future<Value> read( MovableCoordinatedStateImpl* self ) {
		state MovableValue moveState;
		Value rawValue = wait( self->cs.read() );
		if( rawValue.size() ) {
			BinaryReader r( rawValue, IncludeVersion() );
			if (!r.protocolVersion().hasMovableCoordinatedState()) {
				// Old coordinated state, not a MovableValue
				moveState.value = rawValue;
			} else
				r >> moveState;
		}
		// SOMEDAY: If moveState.mode == MovingFrom, read (without locking) old state and assert that it corresponds with our state and is ReallyTo(coordinators)
		if (moveState.mode == MovableValue::MaybeTo) {
			TEST(true);
			ASSERT( moveState.other.present() );
			wait( self->moveTo( self, &self->cs, ClusterConnectionString( moveState.other.get().toString() ), moveState.value ) );
		}
		return moveState.value;
	}

	Future<Void> onConflict() {
		return cs.onConflict();
	}

	Future<Void> setExclusive( Value v ) {
		lastValue=v;
		lastCSValue=BinaryWriter::toValue( MovableValue( v, MovableValue::Active ), IncludeVersion() );
		return cs.setExclusive( lastCSValue.get() );
	}

	ACTOR static Future<Void> move( MovableCoordinatedStateImpl* self, ClusterConnectionString nc ) {
	// Call only after setExclusive returns.  Attempts to move the coordinated state
	// permanently to the new ServerCoordinators, which must be uninitialized.  Returns when the process has
	// reached the point where a leader elected by the new coordinators should be doing the rest of the work
	// (and therefore the caller should die).
		state CoordinatedState cs( self->coordinators );
		state CoordinatedState nccs( ServerCoordinators( Reference<ClusterConnectionFile>( new ClusterConnectionFile(nc) ) ) );
		state Future<Void> creationTimeout = delay(30);
		ASSERT( self->lastValue.present() && self->lastCSValue.present() );
		TraceEvent("StartMove").detail("ConnectionString", nc.toString() );
		choose {
			when (wait(creationTimeout)) { throw new_coordinators_timed_out(); }
			when (Value ncInitialValue = wait( nccs.read() )) {
				ASSERT( !ncInitialValue.size() );  // The new coordinators must be uninitialized!
			}
		}
		TraceEvent("FinishedRead").detail("ConnectionString", nc.toString() );

		choose {
			when (wait(creationTimeout)) { throw new_coordinators_timed_out(); }
			when ( wait( nccs.setExclusive( BinaryWriter::toValue( MovableValue( self->lastValue.get(), MovableValue::MovingFrom, self->coordinators.ccf->getConnectionString().toString() ), IncludeVersion() ) ) ) ) {}
		}

		if (BUGGIFY) wait(delay(5));

		Value oldQuorumState = wait( cs.read() );
		if ( oldQuorumState != self->lastCSValue.get() ) {
			TEST(true);  // Quorum change aborted by concurrent write to old coordination state
			TraceEvent("QuorumChangeAbortedByConcurrency");
			throw coordinated_state_conflict();
		}

		wait( self->moveTo( self, &cs, nc, self->lastValue.get() ) );

		throw coordinators_changed();
	}

	ACTOR static Future<Void> moveTo( MovableCoordinatedStateImpl* self, CoordinatedState* coordinatedState, ClusterConnectionString nc, Value value ) {
		wait( coordinatedState->setExclusive( BinaryWriter::toValue( MovableValue( value, MovableValue::MaybeTo, nc.toString() ), IncludeVersion() ) ) );

		if (BUGGIFY) wait( delay(5) );

		// SOMEDAY: If we are worried about someone magically getting the new cluster ID and interfering, do a second cs.setExclusive( encode( ReallyTo, ... ) )
		TraceEvent("ChangingQuorum").detail("ConnectionString", nc.toString());
		wait( changeLeaderCoordinators( self->coordinators, StringRef(nc.toString()) ) );
		TraceEvent("ChangedQuorum").detail("ConnectionString", nc.toString());
		throw coordinators_changed();
	}
};

void MovableCoordinatedState::operator=(MovableCoordinatedState&& av) { 
	if(impl) {
		delete impl;
	}
	impl = av.impl; 
	av.impl = 0; 
}
MovableCoordinatedState::MovableCoordinatedState( class ServerCoordinators const& coord ) : impl( new MovableCoordinatedStateImpl(coord) ) {}
MovableCoordinatedState::~MovableCoordinatedState() { 
	if(impl) {
		delete impl; 
	}
}
Future<Value> MovableCoordinatedState::read() { return MovableCoordinatedStateImpl::read(impl); }
Future<Void> MovableCoordinatedState::onConflict() { return impl->onConflict(); }
Future<Void> MovableCoordinatedState::setExclusive(Value v) { return impl->setExclusive(v); }
Future<Void> MovableCoordinatedState::move( ClusterConnectionString const& nc ) { return MovableCoordinatedStateImpl::move(impl, nc); }