// _ _ // __ _____ __ ___ ___ __ _| |_ ___ // \ \ /\ / / _ \/ _` \ \ / / |/ _` | __/ _ \ // \ V V / __/ (_| |\ V /| | (_| | || __/ // \_/\_/ \___|\__,_| \_/ |_|\__,_|\__\___| // // Copyright © 2016 - 2025 Weaviate B.V. All rights reserved. // // CONTACT: hello@weaviate.io // package rpc import ( "context" "errors" "fmt" "net" "strings" enterrors "github.com/weaviate/weaviate/entities/errors" "github.com/weaviate/weaviate/usecases/monitoring" grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware" grpc_sentry "github.com/johnbellone/grpc-middleware-sentry" "github.com/sirupsen/logrus" cmd "github.com/weaviate/weaviate/cluster/proto/api" "github.com/weaviate/weaviate/cluster/schema" "github.com/weaviate/weaviate/cluster/types" "google.golang.org/grpc" "google.golang.org/grpc/codes" "google.golang.org/grpc/status" ) type raftPeers interface { Join(id string, addr string, voter bool) error Notify(id string, addr string) error Remove(id string) error Leader() string } type raftFSM interface { Execute(ctx context.Context, cmd *cmd.ApplyRequest) (uint64, error) Query(ctx context.Context, req *cmd.QueryRequest) (*cmd.QueryResponse, error) } type Server struct { raftPeers raftPeers raftFSM raftFSM listenAddress string grpcMessageMaxSize int log *logrus.Logger sentryEnabled bool grpcServer *grpc.Server metrics *monitoring.GRPCServerMetrics } // NewServer returns the Server implementing the RPC interface for RAFT peers management and execute/query commands. // The server must subsequently be started with Open(). // The server will be configure the gRPC service with grpcMessageMaxSize. func NewServer( raftPeers raftPeers, raftFSM raftFSM, listenAddress string, grpcMessageMaxSize int, sentryEnabled bool, metrics *monitoring.GRPCServerMetrics, log *logrus.Logger, ) *Server { return &Server{ raftPeers: raftPeers, raftFSM: raftFSM, listenAddress: listenAddress, log: log, grpcMessageMaxSize: grpcMessageMaxSize, sentryEnabled: sentryEnabled, metrics: metrics, } } // JoinPeer will notify the RAFT cluster that a new peer is joining the cluster. // Returns an error and the current raft leader if joining fails. func (s *Server) JoinPeer(_ context.Context, req *cmd.JoinPeerRequest) (*cmd.JoinPeerResponse, error) { err := s.raftPeers.Join(req.Id, req.Address, req.Voter) if err != nil { return &cmd.JoinPeerResponse{Leader: s.raftPeers.Leader()}, toRPCError(err) } return &cmd.JoinPeerResponse{}, nil } // RemovePeer will notify the RAFT cluster that a peer is removed from the cluster. // Returns an error and the current raft leader if removal fails. func (s *Server) RemovePeer(_ context.Context, req *cmd.RemovePeerRequest) (*cmd.RemovePeerResponse, error) { err := s.raftPeers.Remove(req.Id) if err != nil { return &cmd.RemovePeerResponse{Leader: s.raftPeers.Leader()}, toRPCError(err) } return &cmd.RemovePeerResponse{}, nil } // NotifyPeer will notify the RAFT cluster that a peer has notified that it is ready to be joined. // Returns an error if notifying fails. func (s *Server) NotifyPeer(_ context.Context, req *cmd.NotifyPeerRequest) (*cmd.NotifyPeerResponse, error) { return &cmd.NotifyPeerResponse{}, toRPCError(s.raftPeers.Notify(req.Id, req.Address)) } // Apply will update the RAFT FSM representation to apply req. // Returns the FSM version of that change. // Returns an error and the current raft leader if applying fails. func (s *Server) Apply(ctx context.Context, req *cmd.ApplyRequest) (*cmd.ApplyResponse, error) { v, err := s.raftFSM.Execute(ctx, req) if err != nil { return &cmd.ApplyResponse{Leader: s.raftPeers.Leader()}, toRPCError(err) } return &cmd.ApplyResponse{Version: v}, nil } // Query will read the RAFT FSM schema representation using req. // Returns the result of the query. // Returns an error if querying fails. func (s *Server) Query(ctx context.Context, req *cmd.QueryRequest) (*cmd.QueryResponse, error) { resp, err := s.raftFSM.Query(ctx, req) if err != nil { return &cmd.QueryResponse{}, toRPCError(err) } return resp, nil } // Leader returns the current leader of the RAFT cluster. func (s *Server) Leader() string { return s.raftPeers.Leader() } // Open starts the server and registers it as the cluster service server. // Returns asynchronously once the server has started. // Returns an error if the configured listenAddress is invalid. // Returns an error if the configured listenAddress is un-usable to listen on. func (s *Server) Open() error { s.log.WithField("address", s.listenAddress).Info("starting cloud rpc server ...") if s.listenAddress == "" { return fmt.Errorf("address of rpc server cannot be empty") } listener, err := net.Listen("tcp", s.listenAddress) if err != nil { return fmt.Errorf("server tcp net.listen: %w", err) } var options []grpc.ServerOption options = append(options, grpc.MaxRecvMsgSize(s.grpcMessageMaxSize)) if s.sentryEnabled { options = append(options, grpc.UnaryInterceptor(grpc_middleware.ChainUnaryServer( grpc_sentry.UnaryServerInterceptor(), ))) } if s.metrics != nil { options = append(options, monitoring.InstrumentGrpc(s.metrics)...) } s.grpcServer = grpc.NewServer(options...) cmd.RegisterClusterServiceServer(s.grpcServer, s) enterrors.GoWrapper(func() { if err := s.grpcServer.Serve(listener); err != nil { s.log.WithError(err).Error("serving incoming requests") panic("error accepting incoming requests") } }, s.log) return nil } // Close closes the server and free any used ressources. func (s *Server) Close() { if s.grpcServer != nil { s.grpcServer.Stop() } } // toRPCError returns a gRPC error with the right error code based on the error. func toRPCError(err error) error { if err == nil { return nil } var ec codes.Code switch { case errors.Is(err, types.ErrNotLeader), errors.Is(err, types.ErrLeaderNotFound): ec = codes.ResourceExhausted case errors.Is(err, types.ErrNotOpen): ec = codes.Unavailable case errors.Is(err, schema.ErrMTDisabled): ec = codes.FailedPrecondition case strings.Contains(err.Error(), types.ErrNotFound.Error()): ec = codes.NotFound default: ec = codes.Internal } return status.Error(ec, err.Error()) }