Allow specifying simple gRPC transport credentials

cmd/thanos/flags.go

@@ -16,14 +16,25 @@ import (
 	"gopkg.in/alecthomas/kingpin.v2"
 )
 
-func regCommonServerFlags(cmd *kingpin.CmdClause) (*string, *string, func(log.Logger, *prometheus.Registry, bool, string, bool) (*cluster.Peer, error)) {
-	grpcBindAddr := cmd.Flag("grpc-address", "Listen ip:port address for gRPC endpoints (StoreAPI). Make sure this address is routable from other components if you use gossip, 'grpc-advertise-address' is empty and you require cross-node connection.").
+func regCommonServerFlags(cmd *kingpin.CmdClause) (
+	grpcBindAddr *string,
+	httpBindAddr *string,
+	grpcTLSSrvCert *string,
+	grpcTLSSrvKey *string,
+	grpcTLSSrvClientCA *string,
+	peerFunc func(log.Logger, *prometheus.Registry, bool, string, bool) (*cluster.Peer, error)) {
+
+	grpcBindAddr = cmd.Flag("grpc-address", "Listen ip:port address for gRPC endpoints (StoreAPI). Make sure this address is routable from other components if you use gossip, 'grpc-advertise-address' is empty and you require cross-node connection.").
 		Default("0.0.0.0:10901").String()
 
 	grpcAdvertiseAddr := cmd.Flag("grpc-advertise-address", "Explicit (external) host:port address to advertise for gRPC StoreAPI in gossip cluster. If empty, 'grpc-address' will be used.").
 		String()
 
-	httpBindAddr := regHTTPAddrFlag(cmd)
+	grpcTLSSrvKey = cmd.Flag("grpc-server-tls-key", "TLS Key for the gRPC server, leave blank to disable TLS").Default("").String()
+	grpcTLSSrvCert = cmd.Flag("grpc-server-tls-cert", "TLS Certificate for gRPC server, leave blank to disable TLS").Default("").String()
+	grpcTLSSrvClientCA = cmd.Flag("grpc-server-tls-client-ca", "TLS CA to verify clients against").Default("").String()
+
+	httpBindAddr = regHTTPAddrFlag(cmd)
 
 	clusterBindAddr := cmd.Flag("cluster.address", "Listen ip:port address for gossip cluster.").
 		Default("0.0.0.0:10900").String()
@@ -53,6 +64,9 @@ func regCommonServerFlags(cmd *kingpin.CmdClause) (*string, *string, func(log.Lo
 
 	return grpcBindAddr,
 		httpBindAddr,
+		grpcTLSSrvCert,
+		grpcTLSSrvKey,
+		grpcTLSSrvClientCA,
 		func(logger log.Logger, reg *prometheus.Registry, waitIfEmpty bool, httpAdvertiseAddr string, queryAPIEnabled bool) (*cluster.Peer, error) {
 			host, port, err := cluster.CalculateAdvertiseAddress(*grpcBindAddr, *grpcAdvertiseAddr)
 			if err != nil {

cmd/thanos/main.go

@@ -2,7 +2,10 @@ package main
 
 import (
 	"context"
+	"crypto/tls"
+	"crypto/x509"
 	"fmt"
+	"io/ioutil"
 	"math"
 	"net"
 	"net/http"
@@ -32,6 +35,7 @@ import (
 	"github.com/prometheus/common/version"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/credentials"
 	"google.golang.org/grpc/status"
 	kingpin "gopkg.in/alecthomas/kingpin.v2"
 )
@@ -196,7 +200,7 @@ func registerMetrics(mux *http.ServeMux, g prometheus.Gatherer) {
 // - request histogram
 // - tracing
 // - panic recovery with panic counter
-func defaultGRPCServerOpts(logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer) []grpc.ServerOption {
+func defaultGRPCServerOpts(logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer, cert, key, clientCA string) ([]grpc.ServerOption, error) {
 	met := grpc_prometheus.NewServerMetrics()
 	met.EnableHandlingTimeHistogram(
 		grpc_prometheus.WithHistogramBuckets([]float64{
@@ -214,7 +218,7 @@ func defaultGRPCServerOpts(logger log.Logger, reg *prometheus.Registry, tracer o
 		return status.Errorf(codes.Internal, "%s", p)
 	}
 	reg.MustRegister(met, panicsTotal)
-	return []grpc.ServerOption{
+	opts := []grpc.ServerOption{
 		grpc.MaxSendMsgSize(math.MaxInt32),
 		grpc_middleware.WithUnaryServerChain(
 			met.UnaryServerInterceptor(),
@@ -227,6 +231,43 @@ func defaultGRPCServerOpts(logger log.Logger, reg *prometheus.Registry, tracer o
 			grpc_recovery.StreamServerInterceptor(grpc_recovery.WithRecoveryHandler(grpcPanicRecoveryHandler)),
 		),
 	}
+
+	if key == "" || cert == "" {
+		level.Info(logger).Log("msg", "disabled TLS, key and cert must be set to enable")
+		return opts, nil
+	}
+
+	tlsCfg := &tls.Config{}
+
+	tlsCert, err := tls.LoadX509KeyPair(cert, key)
+	if err != nil {
+		return nil, errors.Wrap(err, "server credentials")
+	}
+
+	level.Info(logger).Log("msg", "enabled gRPC server side TLS")
+
+	tlsCfg.Certificates = []tls.Certificate{tlsCert}
+
+	if clientCA != "" {
+		caPEM, err := ioutil.ReadFile(clientCA)
+		if err != nil {
+			return nil, errors.Wrap(err, "reading client CA")
+		}
+
+		certPool := x509.NewCertPool()
+		if !certPool.AppendCertsFromPEM(caPEM) {
+			return nil, errors.Wrap(err, "building client CA")
+		}
+		tlsCfg.ClientCAs = certPool
+		tlsCfg.ClientAuth = tls.RequireAndVerifyClientCert
+
+		level.Info(logger).Log("msg", "gRPC server TLS client verification enabled")
+	}
+
+	creds := credentials.NewTLS(tlsCfg)
+	opts = append(opts, grpc.Creds(creds))
+
+	return opts, nil
 }
 
 // metricHTTPListenGroup is a run.Group that servers HTTP endpoint with only Prometheus metrics.

cmd/thanos/query.go

@@ -2,7 +2,10 @@ package main
 
 import (
 	"context"
+	"crypto/tls"
+	"crypto/x509"
 	"fmt"
+	"io/ioutil"
 	"math"
 	"net"
 	"net/http"
@@ -28,18 +31,24 @@ import (
 	"github.com/prometheus/prometheus/promql"
 	"github.com/prometheus/tsdb/labels"
 	"google.golang.org/grpc"
+	"google.golang.org/grpc/credentials"
 	"gopkg.in/alecthomas/kingpin.v2"
 )
 
 // registerQuery registers a query command.
 func registerQuery(m map[string]setupFunc, app *kingpin.Application, name string) {
 	cmd := app.Command(name, "query node exposing PromQL enabled Query API with data retrieved from multiple store nodes")
 
-	grpcBindAddr, httpBindAddr, newPeerFn := regCommonServerFlags(cmd)
+	grpcBindAddr, httpBindAddr, srvCert, srvKey, srvClientCA, newPeerFn := regCommonServerFlags(cmd)
 
 	httpAdvertiseAddr := cmd.Flag("http-advertise-address", "Explicit (external) host:port address to advertise for HTTP QueryAPI in gossip cluster. If empty, 'http-address' will be used.").
 		String()
 
+	secure := cmd.Flag("grpc-client-tls-secure", "Use TLS when talking to the gRPC server").Default("false").Bool()
+	cert := cmd.Flag("grpc-client-tls-cert", "TLS Certificates to use to identify this client to the server").Default("").String()
+	key := cmd.Flag("grpc-client-tls-key", "TLS Key for the client's certificate").Default("").String()
+	caCert := cmd.Flag("grpc-client-tls-ca", "TLS CA Certificates to use to verify gRPC servers").Default("").String()
+
 	queryTimeout := modelDuration(cmd.Flag("query.timeout", "Maximum time to process query by query node.").
 		Default("2m"))
 
@@ -83,6 +92,13 @@ func registerQuery(m map[string]setupFunc, app *kingpin.Application, name string
 			reg,
 			tracer,
 			*grpcBindAddr,
+			*srvCert,
+			*srvKey,
+			*srvClientCA,
+			*secure,
+			*cert,
+			*key,
+			*caCert,
 			*httpBindAddr,
 			*maxConcurrentQueries,
 			time.Duration(*queryTimeout),
@@ -95,7 +111,7 @@ func registerQuery(m map[string]setupFunc, app *kingpin.Application, name string
 	}
 }
 
-func storeClientGRPCOpts(reg *prometheus.Registry, tracer opentracing.Tracer) []grpc.DialOption {
+func storeClientGRPCOpts(logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer, secure bool, cert, key, caCert string) ([]grpc.DialOption, error) {
 	grpcMets := grpc_prometheus.NewClientMetrics()
 	grpcMets.EnableClientHandlingTimeHistogram(
 		grpc_prometheus.WithHistogramBuckets([]float64{
@@ -108,7 +124,6 @@ func storeClientGRPCOpts(reg *prometheus.Registry, tracer opentracing.Tracer) []
 		// Current limit is ~2GB.
 		// TODO(bplotka): Split sent chunks on store node per max 4MB chunks if needed.
 		grpc.WithDefaultCallOptions(grpc.MaxCallRecvMsgSize(math.MaxInt32)),
-		grpc.WithInsecure(),
 		grpc.WithUnaryInterceptor(
 			grpc_middleware.ChainUnaryClient(
 				grpcMets.UnaryClientInterceptor(),
@@ -127,7 +142,52 @@ func storeClientGRPCOpts(reg *prometheus.Registry, tracer opentracing.Tracer) []
 		reg.MustRegister(grpcMets)
 	}
 
-	return dialOpts
+	if !secure {
+		dialOpts = append(dialOpts, grpc.WithInsecure())
+		return dialOpts, nil
+	}
+
+	level.Info(logger).Log("msg", "Enabling client to server TLS")
+
+	var certPool *x509.CertPool
+
+	if caCert != "" {
+		caPEM, err := ioutil.ReadFile(caCert)
+		if err != nil {
+			return nil, errors.Wrap(err, "reading client CA")
+		}
+
+		certPool = x509.NewCertPool()
+		if !certPool.AppendCertsFromPEM(caPEM) {
+			return nil, errors.Wrap(err, "building client CA")
+		}
+		level.Info(logger).Log("msg", "TLS Client using provided certificate pool")
+	} else {
+		var err error
+		certPool, err = x509.SystemCertPool()
+		if err != nil {
+			return nil, errors.Wrap(err, "reading system certificate pool")
+		}
+		level.Info(logger).Log("msg", "TLS Client using system certificate pool")
+	}
+
+	tlsCfg := &tls.Config{
+		RootCAs: certPool,
+	}
+
+	if cert != "" {
+		cert, err := tls.LoadX509KeyPair(cert, key)
+		if err != nil {
+			return nil, errors.Wrap(err, "client credentials")
+		}
+		tlsCfg.Certificates = []tls.Certificate{cert}
+		level.Info(logger).Log("msg", "TLS Client authentication enabled")
+	}
+
+	creds := credentials.NewTLS(tlsCfg)
+	dialOpts = append(dialOpts, grpc.WithTransportCredentials(creds))
+
+	return dialOpts, nil
 }
 
 // runQuery starts a server that exposes PromQL Query API. It is responsible for querying configured
@@ -138,6 +198,13 @@ func runQuery(
 	reg *prometheus.Registry,
 	tracer opentracing.Tracer,
 	grpcBindAddr string,
+	srvCert string,
+	srvKey string,
+	srvClientCA string,
+	secure bool,
+	cert string,
+	key string,
+	caCert string,
 	httpBindAddr string,
 	maxConcurrentQueries int,
 	queryTimeout time.Duration,
@@ -155,6 +222,12 @@ func runQuery(
 
 		staticSpecs = append(staticSpecs, query.NewGRPCStoreSpec(addr))
 	}
+
+	dialOpts, err := storeClientGRPCOpts(logger, reg, tracer, secure, cert, key, caCert)
+	if err != nil {
+		return errors.Wrap(err, "building gRPC client")
+	}
+
 	var (
 		stores = query.NewStoreSet(
 			logger,
@@ -172,7 +245,7 @@ func runQuery(
 				}
 				return specs
 			},
-			storeClientGRPCOpts(reg, tracer),
+			dialOpts,
 		)
 		proxy = store.NewProxyStore(logger, func(context.Context) ([]store.Client, error) {
 			return stores.Get(), nil
@@ -241,7 +314,12 @@ func runQuery(
 		}
 		logger := log.With(logger, "component", "query")
 
-		s := grpc.NewServer(defaultGRPCServerOpts(logger, reg, tracer)...)
+		opts, err := defaultGRPCServerOpts(logger, reg, tracer, srvCert, srvKey, srvClientCA)
+		if err != nil {
+			return errors.Wrapf(err, "build gRPC server")
+		}
+
+		s := grpc.NewServer(opts...)
 		storepb.RegisterStoreServer(s, proxy)
 
 		g.Add(func() error {

cmd/thanos/rule.go

@@ -52,7 +52,7 @@ import (
 func registerRule(m map[string]setupFunc, app *kingpin.Application, name string) {
 	cmd := app.Command(name, "ruler evaluating Prometheus rules against given Query nodes, exposing Store API and storing old blocks in bucket")
 
-	grpcBindAddr, httpBindAddr, newPeerFn := regCommonServerFlags(cmd)
+	grpcBindAddr, httpBindAddr, cert, key, clientCA, newPeerFn := regCommonServerFlags(cmd)
 
 	labelStrs := cmd.Flag("label", "Labels to be applied to all generated metrics (repeated).").
 		PlaceHolder("<name>=\"<value>\"").Strings()
@@ -107,6 +107,9 @@ func registerRule(m map[string]setupFunc, app *kingpin.Application, name string)
 			lset,
 			*alertmgrs,
 			*grpcBindAddr,
+			*cert,
+			*key,
+			*clientCA,
 			*httpBindAddr,
 			time.Duration(*evalInterval),
 			*dataDir,
@@ -131,6 +134,9 @@ func runRule(
 	lset labels.Labels,
 	alertmgrURLs []string,
 	grpcBindAddr string,
+	cert string,
+	key string,
+	clientCA string,
 	httpBindAddr string,
 	evalInterval time.Duration,
 	dataDir string,
@@ -374,7 +380,11 @@ func runRule(
 
 		store := store.NewTSDBStore(logger, reg, db, lset)
 
-		s := grpc.NewServer(defaultGRPCServerOpts(logger, reg, tracer)...)
+		opts, err := defaultGRPCServerOpts(logger, reg, tracer, cert, key, clientCA)
+		if err != nil {
+			return errors.Wrap(err, "setup gRPC options")
+		}
+		s := grpc.NewServer(opts...)
 		storepb.RegisterStoreServer(s, store)
 
 		g.Add(func() error {

cmd/thanos/sidecar.go

@@ -35,7 +35,7 @@ import (
 func registerSidecar(m map[string]setupFunc, app *kingpin.Application, name string) {
 	cmd := app.Command(name, "sidecar for Prometheus server")
 
-	grpcBindAddr, httpBindAddr, newPeerFn := regCommonServerFlags(cmd)
+	grpcBindAddr, httpBindAddr, cert, key, clientCA, newPeerFn := regCommonServerFlags(cmd)
 
 	promURL := cmd.Flag("prometheus.url", "URL at which to reach Prometheus's API. For better performance use local network.").
 		Default("http://localhost:9090").URL()
@@ -74,6 +74,9 @@ func registerSidecar(m map[string]setupFunc, app *kingpin.Application, name stri
 			reg,
 			tracer,
 			*grpcBindAddr,
+			*cert,
+			*key,
+			*clientCA,
 			*httpBindAddr,
 			*promURL,
 			*dataDir,
@@ -92,6 +95,9 @@ func runSidecar(
 	reg *prometheus.Registry,
 	tracer opentracing.Tracer,
 	grpcBindAddr string,
+	cert string,
+	key string,
+	clientCA string,
 	httpBindAddr string,
 	promURL *url.URL,
 	dataDir string,
@@ -208,7 +214,11 @@ func runSidecar(
 			return errors.Wrap(err, "create Prometheus store")
 		}
 
-		s := grpc.NewServer(defaultGRPCServerOpts(logger, reg, tracer)...)
+		opts, err := defaultGRPCServerOpts(logger, reg, tracer, cert, key, clientCA)
+		if err != nil {
+			return errors.Wrap(err, "setup gRPC server")
+		}
+		s := grpc.NewServer(opts...)
 		storepb.RegisterStoreServer(s, promStore)
 
 		g.Add(func() error {