From 10936f8e8b87f5681eae2187aeffed6fb5b80774 Mon Sep 17 00:00:00 2001
From: Nikolay Nikolaev <nikolay.nikolaev@konghq.com>
Date: Fri, 19 Feb 2021 10:43:17 +0200
Subject: [PATCH] chore(kuma-cp) set GRPC keepalives (#1580)

Signed-off-by: Nikolay Nikolaev <nikolay.nikolaev@konghq.com>

Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
---
 pkg/dp-server/server/server.go | 16 +++++++++++++++-
 pkg/kds/mux/server.go          | 16 +++++++++++++++-
 pkg/mads/server/grpc.go        | 16 +++++++++++++++-
 3 files changed, 45 insertions(+), 3 deletions(-)

diff --git a/pkg/dp-server/server/server.go b/pkg/dp-server/server/server.go
index 66bc2c6e6a83..b381f6a90bb2 100644
--- a/pkg/dp-server/server/server.go
+++ b/pkg/dp-server/server/server.go
@@ -5,6 +5,9 @@ import (
 	"fmt"
 	"net/http"
 	"strings"
+	"time"
+
+	"google.golang.org/grpc/keepalive"
 
 	http_prometheus "github.com/slok/go-http-metrics/metrics/prometheus"
 	"github.com/slok/go-http-metrics/middleware"
@@ -19,7 +22,10 @@ import (
 
 var log = core.Log.WithName("dp-server")
 
-const grpcMaxConcurrentStreams = 1000000
+const (
+	grpcMaxConcurrentStreams = 1000000
+	grpcKeepAliveTime        = 15 * time.Second
+)
 
 type DpServer struct {
 	config         dp_server.DpServerConfig
@@ -33,6 +39,14 @@ var _ component.Component = &DpServer{}
 func NewDpServer(config dp_server.DpServerConfig, metrics metrics.Metrics) *DpServer {
 	grpcOptions := []grpc.ServerOption{
 		grpc.MaxConcurrentStreams(grpcMaxConcurrentStreams),
+		grpc.KeepaliveParams(keepalive.ServerParameters{
+			Time:    grpcKeepAliveTime,
+			Timeout: grpcKeepAliveTime,
+		}),
+		grpc.KeepaliveEnforcementPolicy(keepalive.EnforcementPolicy{
+			MinTime:             grpcKeepAliveTime,
+			PermitWithoutStream: true,
+		}),
 	}
 	grpcOptions = append(grpcOptions, metrics.GRPCServerInterceptors()...)
 	grpcServer := grpc.NewServer(grpcOptions...)
diff --git a/pkg/kds/mux/server.go b/pkg/kds/mux/server.go
index acc331e6bb05..34d4b72f9b28 100644
--- a/pkg/kds/mux/server.go
+++ b/pkg/kds/mux/server.go
@@ -3,6 +3,9 @@ package mux
 import (
 	"fmt"
 	"net"
+	"time"
+
+	"google.golang.org/grpc/keepalive"
 
 	"github.com/pkg/errors"
 	"google.golang.org/grpc"
@@ -16,7 +19,10 @@ import (
 	core_metrics "github.com/kumahq/kuma/pkg/metrics"
 )
 
-const grpcMaxConcurrentStreams = 1000000
+const (
+	grpcMaxConcurrentStreams = 1000000
+	grpcKeepAliveTime        = 15 * time.Second
+)
 
 var (
 	muxServerLog = core.Log.WithName("mux-server")
@@ -52,6 +58,14 @@ func NewServer(callbacks []Callbacks, config multizone.KdsServerConfig, metrics
 func (s *server) Start(stop <-chan struct{}) error {
 	grpcOptions := []grpc.ServerOption{
 		grpc.MaxConcurrentStreams(grpcMaxConcurrentStreams),
+		grpc.KeepaliveParams(keepalive.ServerParameters{
+			Time:    grpcKeepAliveTime,
+			Timeout: grpcKeepAliveTime,
+		}),
+		grpc.KeepaliveEnforcementPolicy(keepalive.EnforcementPolicy{
+			MinTime:             grpcKeepAliveTime,
+			PermitWithoutStream: true,
+		}),
 	}
 	grpcOptions = append(grpcOptions, s.metrics.GRPCServerInterceptors()...)
 	useTLS := s.config.TlsCertFile != ""
diff --git a/pkg/mads/server/grpc.go b/pkg/mads/server/grpc.go
index c1b63ab5d5bc..34bd75aeab07 100644
--- a/pkg/mads/server/grpc.go
+++ b/pkg/mads/server/grpc.go
@@ -3,6 +3,9 @@ package server
 import (
 	"fmt"
 	"net"
+	"time"
+
+	"google.golang.org/grpc/keepalive"
 
 	observability_proto "github.com/kumahq/kuma/api/observability/v1alpha1"
 
@@ -14,7 +17,10 @@ import (
 	core_metrics "github.com/kumahq/kuma/pkg/metrics"
 )
 
-const grpcMaxConcurrentStreams = 1000000
+const (
+	grpcMaxConcurrentStreams = 1000000
+	grpcKeepAliveTime        = 15 * time.Second
+)
 
 var (
 	grpcServerLog = core.Log.WithName("mads-server").WithName("grpc")
@@ -33,6 +39,14 @@ var (
 func (s *grpcServer) Start(stop <-chan struct{}) error {
 	grpcOptions := []grpc.ServerOption{
 		grpc.MaxConcurrentStreams(grpcMaxConcurrentStreams),
+		grpc.KeepaliveParams(keepalive.ServerParameters{
+			Time:    grpcKeepAliveTime,
+			Timeout: grpcKeepAliveTime,
+		}),
+		grpc.KeepaliveEnforcementPolicy(keepalive.EnforcementPolicy{
+			MinTime:             grpcKeepAliveTime,
+			PermitWithoutStream: true,
+		}),
 	}
 	grpcOptions = append(grpcOptions, s.metrics.GRPCServerInterceptors()...)
 	grpcOptions = append(grpcOptions, grpc.MaxConcurrentStreams(grpcMaxConcurrentStreams))