Use fewer connections in the multi-version client (release-6.3)

bindings/c/fdb_c.cpp

@@ -22,6 +22,7 @@
 #define FDB_INCLUDE_LEGACY_TYPES
 
 #include "fdbclient/MultiVersionTransaction.h"
+#include "fdbclient/MultiVersionAssignmentVars.h"
 #include "foundationdb/fdb_c.h"
 
 int g_api_version = 0;
@@ -322,6 +323,22 @@ extern "C" DLLEXPORT fdb_error_t fdb_database_create_transaction(FDBDatabase* d,
 	                 *out_transaction = (FDBTransaction*)tr.extractPtr(););
 }
 
+// Returns the protocol version reported by the coordinator this client is connected to
+// If an expected version is non-zero, the future won't return until the protocol version is different than expected
+// Note: this will never return if the server is running a protocol from FDB 5.0 or older
+extern "C" DLLEXPORT FDBFuture* fdb_database_get_server_protocol(FDBDatabase* db, uint64_t expected_version) {
+	Optional<ProtocolVersion> expected;
+	if (expected_version > 0) {
+		expected = ProtocolVersion(expected_version);
+	}
+
+	return (
+	    FDBFuture*)(mapThreadFuture<ProtocolVersion,
+	                                uint64_t>(DB(db)->getServerProtocol(expected), [](ErrorOr<ProtocolVersion> result) {
+		                return result.map<uint64_t>([](ProtocolVersion pv) { return pv.versionWithFlags(); });
+	                }).extractPtr());
+}
+
 extern "C" DLLEXPORT void fdb_transaction_destroy(FDBTransaction* tr) {
 	try {
 		TXN(tr)->delref();

documentation/sphinx/source/release-notes/release-notes-630.rst

@@ -4,6 +4,10 @@
 Release Notes
 #############
 
+6.3.13
+======
+* The multi-version client now requires at most two client connections with version 6.2 or larger, regardless of how many external clients are configured. Clients older than 6.2 will continue to create an additional connection each. `(PR #4667) <https://github.com/apple/foundationdb/pull/4667>`_
+
 6.3.12
 ======
 * Change the default for --knob_tls_server_handshake_threads to 64. The previous was 1000. This avoids starting 1000 threads by default, but may adversely affect recovery time for large clusters using tls. Users with large tls clusters should consider explicitly setting this knob in their foundationdb.conf file. `(PR #4421) <https://github.com/apple/foundationdb/pull/4421>`_

fdbclient/CoordinationInterface.h

@@ -30,13 +30,18 @@
 
 const int MAX_CLUSTER_FILE_BYTES = 60000;
 
+// The coordinator interface as exposed to clients
 struct ClientLeaderRegInterface {
 	RequestStream<struct GetLeaderRequest> getLeader;
 	RequestStream<struct OpenDatabaseCoordRequest> openDatabase;
 
 	ClientLeaderRegInterface() {}
 	ClientLeaderRegInterface(NetworkAddress remote);
 	ClientLeaderRegInterface(INetwork* local);
+
+	bool operator==(const ClientLeaderRegInterface& rhs) const {
+		return getLeader == rhs.getLeader && openDatabase == rhs.openDatabase;
+	}
 };
 
 class ClusterConnectionString {

fdbclient/DatabaseContext.h

@@ -117,6 +117,7 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 		return (DatabaseContext*)DatabaseContext::operator new(sizeof(DatabaseContext));
 	}
 
+	// Static constructor used by server processes to create a DatabaseContext
 	// For internal (fdbserver) use only
 	static Database create(Reference<AsyncVar<ClientDBInfo>> clientInfo,
 	                       Future<Void> clientInfoMonitor,
@@ -129,9 +130,11 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 
 	~DatabaseContext();
 
+	// Constructs a new copy of this DatabaseContext from the parameters of this DatabaseContext
 	Database clone() const {
 		return Database(new DatabaseContext(connectionFile,
 		                                    clientInfo,
+		                                    coordinator,
 		                                    clientInfoMonitor,
 		                                    taskID,
 		                                    clientLocality,
@@ -158,6 +161,11 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 	Future<Void> onMasterProxiesChanged();
 	Future<HealthMetrics> getHealthMetrics(bool detailed);
 
+	// Returns the protocol version reported by the coordinator this client is connected to
+	// If an expected version is given, the future won't return until the protocol version is different than expected
+	// Note: this will never return if the server is running a protocol from FDB 5.0 or older
+	Future<ProtocolVersion> getClusterProtocol(Optional<ProtocolVersion> expectedVersion = Optional<ProtocolVersion>());
+
 	// Update the watch counter for the database
 	void addWatch();
 	void removeWatch();
@@ -195,6 +203,7 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 	// private:
 	explicit DatabaseContext(Reference<AsyncVar<Reference<ClusterConnectionFile>>> connectionFile,
 	                         Reference<AsyncVar<ClientDBInfo>> clientDBInfo,
+	                         Reference<AsyncVar<Optional<ClientLeaderRegInterface>>> coordinator,
 	                         Future<Void> clientInfoMonitor,
 	                         TaskPriority taskID,
 	                         LocalityData const& clientLocality,
@@ -324,6 +333,9 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 	Future<Void> clientInfoMonitor;
 	Future<Void> connected;
 
+	// An AsyncVar that reports the coordinator this DatabaseContext is interacting with
+	Reference<AsyncVar<Optional<ClientLeaderRegInterface>>> coordinator;
+
 	Reference<AsyncVar<Optional<ClusterInterface>>> statusClusterInterface;
 	Future<Void> statusLeaderMon;
 	double lastStatusFetch;

fdbclient/IClientApi.h

@@ -94,6 +94,12 @@ class IDatabase {
 	virtual Reference<ITransaction> createTransaction() = 0;
 	virtual void setOption(FDBDatabaseOptions::Option option, Optional<StringRef> value = Optional<StringRef>()) = 0;
 
+	// Returns the protocol version reported by the coordinator this client is connected to
+	// If an expected version is given, the future won't return until the protocol version is different than expected
+	// Note: this will never return if the server is running a protocol from FDB 5.0 or older
+	virtual ThreadFuture<ProtocolVersion> getServerProtocol(
+	    Optional<ProtocolVersion> expectedVersion = Optional<ProtocolVersion>()) = 0;
+
 	virtual void addref() = 0;
 	virtual void delref() = 0;
 };

fdbclient/MonitorLeader.actor.cpp

@@ -740,6 +740,7 @@ void shrinkProxyList(ClientDBInfo& ni,
 ACTOR Future<MonitorLeaderInfo> monitorProxiesOneGeneration(
     Reference<ClusterConnectionFile> connFile,
     Reference<AsyncVar<ClientDBInfo>> clientInfo,
+    Reference<AsyncVar<Optional<ClientLeaderRegInterface>>> coordinator,
     MonitorLeaderInfo info,
     Reference<ReferencedObject<Standalone<VectorRef<ClientVersionRef>>>> supportedVersions,
     Key traceLogGroup) {
@@ -755,6 +756,9 @@ ACTOR Future<MonitorLeaderInfo> monitorProxiesOneGeneration(
 	loop {
 		state ClientLeaderRegInterface clientLeaderServer(addrs[idx]);
 		state OpenDatabaseCoordRequest req;
+
+		coordinator->set(clientLeaderServer);
+
 		req.clusterKey = cs.clusterKey();
 		req.coordinators = cs.coordinators();
 		req.knownClientInfoID = clientInfo->get().id;
@@ -821,13 +825,14 @@ ACTOR Future<MonitorLeaderInfo> monitorProxiesOneGeneration(
 ACTOR Future<Void> monitorProxies(
     Reference<AsyncVar<Reference<ClusterConnectionFile>>> connFile,
     Reference<AsyncVar<ClientDBInfo>> clientInfo,
+    Reference<AsyncVar<Optional<ClientLeaderRegInterface>>> coordinator,
     Reference<ReferencedObject<Standalone<VectorRef<ClientVersionRef>>>> supportedVersions,
     Key traceLogGroup) {
 	state MonitorLeaderInfo info(connFile->get());
 	loop {
 		choose {
 			when(MonitorLeaderInfo _info = wait(monitorProxiesOneGeneration(
-			         connFile->get(), clientInfo, info, supportedVersions, traceLogGroup))) {
+			         connFile->get(), clientInfo, coordinator, info, supportedVersions, traceLogGroup))) {
 				info = _info;
 			}
 			when(wait(connFile->onChange())) {

fdbclient/MonitorLeader.h

@@ -76,6 +76,7 @@ Future<Void> monitorLeaderForProxies(Value const& key,
 Future<Void> monitorProxies(
     Reference<AsyncVar<Reference<ClusterConnectionFile>>> const& connFile,
     Reference<AsyncVar<ClientDBInfo>> const& clientInfo,
+    Reference<AsyncVar<Optional<ClientLeaderRegInterface>>> const& coordinator,
     Reference<ReferencedObject<Standalone<VectorRef<ClientVersionRef>>>> const& supportedVersions,
     Key const& traceLogGroup);