feat(inputs.gnmi): Add keepalive settings (#15171)

Co-authored-by: Hunter Kelly <retnuh@gmail.com>

plugins/inputs/gnmi/README.md

@@ -52,6 +52,18 @@ See the [CONFIGURATION.md][CONFIGURATION.md] for more details.
   ## redial in case of failures after
   # redial = "10s"
 
+  ## gRPC Keepalive settings
+  ## See https://pkg.go.dev/google.golang.org/grpc/keepalive
+  ## The client will ping the server to see if the transport is still alive if it has
+  ## not see any activity for the given time.
+  ## If not set, none of the keep-alive setting (including those below) will be applied.
+  ## If set and set below 10 seconds, the gRPC library will apply a minimum value of 10s will be used instead.
+  # keepalive_time = ""
+
+  ## Timeout for seeing any activity after the keep-alive probe was
+  ## sent. If no activity is seen the connection is closed.
+  # keepalive_timeout = ""
+
   ## gRPC Maximum Message Size
   # max_msg_size = "4MB"
 

plugins/inputs/gnmi/gnmi.go

@@ -12,6 +12,7 @@ import (
 
 	"github.com/google/gnxi/utils/xpath"
 	gnmiLib "github.com/openconfig/gnmi/proto/gnmi"
+	"google.golang.org/grpc/keepalive"
 	"google.golang.org/grpc/metadata"
 
 	"github.com/influxdata/telegraf"
@@ -57,6 +58,8 @@ type GNMI struct {
 	GuessPathTag        bool              `toml:"guess_path_tag" deprecated:"1.30.0;use 'path_guessing_strategy' instead"`
 	GuessPathStrategy   string            `toml:"path_guessing_strategy"`
 	EnableTLS           bool              `toml:"enable_tls" deprecated:"1.27.0;use 'tls_enable' instead"`
+	KeepaliveTime       config.Duration   `toml:"keepalive_time"`
+	KeepaliveTimeout    config.Duration   `toml:"keepalive_timeout"`
 	Log                 telegraf.Logger   `toml:"-"`
 	internaltls.ClientConfig
 
@@ -233,6 +236,11 @@ func (c *GNMI) Start(acc telegraf.Accumulator) error {
 				trimSlash:           c.TrimFieldNames,
 				guessPathStrategy:   c.GuessPathStrategy,
 				log:                 c.Log,
+				ClientParameters: keepalive.ClientParameters{
+					Time:                time.Duration(c.KeepaliveTime),
+					Timeout:             time.Duration(c.KeepaliveTimeout),
+					PermitWithoutStream: false,
+				},
 			}
 			for ctx.Err() == nil {
 				if err := h.subscribeGNMI(ctx, acc, tlscfg, request); err != nil && ctx.Err() == nil {

plugins/inputs/gnmi/handler.go

@@ -17,6 +17,7 @@ import (
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials"
 	"google.golang.org/grpc/credentials/insecure"
+	"google.golang.org/grpc/keepalive"
 	"google.golang.org/protobuf/encoding/protojson"
 	"google.golang.org/protobuf/proto"
 
@@ -42,6 +43,7 @@ type handler struct {
 	trimSlash           bool
 	guessPathStrategy   string
 	log                 telegraf.Logger
+	keepalive.ClientParameters
 }
 
 // SubscribeGNMI and extract telemetry data
@@ -62,6 +64,10 @@ func (h *handler) subscribeGNMI(ctx context.Context, acc telegraf.Accumulator, t
 		))
 	}
 
+	if h.ClientParameters.Time > 0 {
+		opts = append(opts, grpc.WithKeepaliveParams(h.ClientParameters))
+	}
+
 	client, err := grpc.DialContext(ctx, h.address, opts...)
 	if err != nil {
 		return fmt.Errorf("failed to dial: %w", err)

plugins/inputs/gnmi/sample.conf

@@ -13,6 +13,18 @@
   ## redial in case of failures after
   # redial = "10s"
 
+  ## gRPC Keepalive settings
+  ## See https://pkg.go.dev/google.golang.org/grpc/keepalive
+  ## The client will ping the server to see if the transport is still alive if it has
+  ## not see any activity for the given time.
+  ## If not set, none of the keep-alive setting (including those below) will be applied.
+  ## If set and set below 10 seconds, the gRPC library will apply a minimum value of 10s will be used instead.
+  # keepalive_time = ""
+
+  ## Timeout for seeing any activity after the keep-alive probe was
+  ## sent. If no activity is seen the connection is closed.
+  # keepalive_timeout = ""
+
   ## gRPC Maximum Message Size
   # max_msg_size = "4MB"