otelgrpc

README

Go gRPC Interceptors for OTLP monitoring

Port of metric library - https://github.com/grpc-ecosystem/go-grpc-prometheus onto OTLP protocol

Open Telemetry monitoring for your gRPC Go servers and clients.

Interceptors

gRPC Go recently acquired support for Interceptors, i.e. middleware that is executed by a gRPC Server before the request is passed onto the user's application logic. It is a perfect way to implement common patterns: auth, logging and... monitoring.

To use Interceptors in chains, please see go-grpc-middleware.

This library requires Go 1.9 or later.

Server-side

import "github.com/tel-io/instrumentation/module/otelgrpc"
...
    // Explicitly initiate metrics after otel meter register
    myMetrics := otelgrpc.NewServerMetrics()
	
    // Initialize your gRPC server's interceptor.
    myServer := grpc.NewServer(
        grpc.StreamInterceptor(myMetrics.StreamServerInterceptor),
        grpc.UnaryInterceptor(myMetrics.UnaryServerInterceptor),
    )
    // Register your gRPC service implementations.
    myservice.RegisterMyServiceServer(s.server, &myServiceImpl{})
    // After all your registrations, make sure all of the Prometheus metrics are initialized.
    grpc_prometheus.Register(myServer)
    // Register Prometheus metrics handler.    
    http.Handle("/metrics", promhttp.Handler())
...

Client-side

import "github.com/tel-io/instrumentation/module/otelgrpc"
...
    // Explicitly initiate metrics after otel meter register
    myMetrics := otelgrpc.NewClientMetrics()

   clientConn, err = grpc.Dial(
       address,
		   grpc.WithUnaryInterceptor(myMetrics.UnaryClientInterceptor),
		   grpc.WithStreamInterceptor(myMetrics.StreamClientInterceptor)
   )
   client = pb_testproto.NewTestServiceClient(clientConn)
   resp, err := client.PingEmpty(s.ctx, &myservice.Request{Msg: "hello"})
...

Metrics

Labels

All server-side metrics start with grpc_server as Prometheus subsystem name. All client-side metrics start with grpc_client. Both of them have mirror-concepts. Similarly all methods contain the same rich labels:

• grpc_service - the gRPC service name, which is the combination of protobuf package and the grpc_service section name. E.g. for package = mwitkow.testproto and service TestService the label will be grpc_service="mwitkow.testproto.TestService"

• grpc_method - the name of the method called on the gRPC service. E.g.
  grpc_method="Ping"

• grpc_type - the gRPC type of request. Differentiating between the two is important especially for latency measurements.

  • unary is single request, single response RPC
  • client_stream is a multi-request, single response RPC
  • server_stream is a single request, multi-response RPC
  • bidi_stream is a multi-request, multi-response RPC

Additionally for completed RPCs, the following labels are used:

• grpc_code - the human-readable gRPC status code. The list of all statuses is to long, but here are some common ones:

  • OK - means the RPC was successful
  • IllegalArgument - RPC contained bad values
  • Internal - server-side error not disclosed to the clients

Counters

The counters and their up to date documentation is in server_reporter.go and client_reporter.go the respective Prometheus handler (usually /metrics).

For the purpose of this documentation we will only discuss grpc_server metrics. The grpc_client ones contain mirror concepts.

For simplicity, let's assume we're tracking a single server-side RPC call of mwitkow.testproto.TestService, calling the method PingList. The call succeeds and returns 20 messages in the stream.

First, immediately after the server receives the call it will increment the grpc_server_started_total and start the handling time clock (if histograms are enabled).

grpc_server_started_total{grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 1

Then the user logic gets invoked. It receives one message from the client containing the request (it's a server_stream):

grpc_server_msg_received_total{grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 1

The user logic may return an error, or send multiple messages back to the client. In this case, on each of the 20 messages sent back, a counter will be incremented:

grpc_server_msg_sent_total{grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 20

After the call completes, its status (OK or other gRPC status code) and the relevant call labels increment the grpc_server_handled_total counter.

grpc_server_handled_total{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 1

Histograms

Prometheus histograms are a great way to measure latency distributions of your RPCs. However, since it is bad practice to have metrics of high cardinality the latency monitoring metrics are disabled by default. To enable them please call the following in your server initialization code:

grpc_prometheus.EnableHandlingTimeHistogram()

After the call completes, its handling time will be recorded in a Prometheus histogram variable grpc_server_handling_seconds. The histogram variable contains three sub-metrics:

• grpc_server_handling_seconds_count - the count of all completed RPCs by status and method
• grpc_server_handling_seconds_sum - cumulative time of RPCs by status and method, useful for calculating average handling times
• grpc_server_handling_seconds_bucket - contains the counts of RPCs by status and method in respective handling-time buckets. These buckets can be used by Prometheus to estimate SLAs (see here)

The counter values will look as follows:

grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.005"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.01"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.025"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.05"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.1"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.25"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="0.5"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="1"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="2.5"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="5"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="10"} 1
grpc_server_handling_seconds_bucket{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream",le="+Inf"} 1
grpc_server_handling_seconds_sum{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 0.0003866430000000001
grpc_server_handling_seconds_count{grpc_code="OK",grpc_method="PingList",grpc_service="mwitkow.testproto.TestService",grpc_type="server_stream"} 1

Useful query examples

Prometheus philosophy is to provide raw metrics to the monitoring system, and let the aggregations be handled there. The verbosity of above metrics make it possible to have that flexibility. Here's a couple of useful monitoring queries:

request inbound rate

sum(rate(grpc_server_started_total{job="foo"}[1m])) by (grpc_service)

For job="foo" (common label to differentiate between Prometheus monitoring targets), calculate the rate of requests per second (1 minute window) for each gRPC grpc_service that the job has. Please note how the grpc_method is being omitted here: all methods of a given gRPC service will be summed together.

unary request error rate

sum(rate(grpc_server_handled_total{job="foo",grpc_type="unary",grpc_code!="OK"}[1m])) by (grpc_service)

For job="foo", calculate the per-grpc_service rate of unary (1:1) RPCs that failed, i.e. the ones that didn't finish with OK code.

unary request error percentage

sum(rate(grpc_server_handled_total{job="foo",grpc_type="unary",grpc_code!="OK"}[1m])) by (grpc_service)
 / 
sum(rate(grpc_server_started_total{job="foo",grpc_type="unary"}[1m])) by (grpc_service)
 * 100.0

For job="foo", calculate the percentage of failed requests by service. It's easy to notice that this is a combination of the two above examples. This is an example of a query you would like to alert on in your system for SLA violations, e.g. "no more than 1% requests should fail".

average response stream size

sum(rate(grpc_server_msg_sent_total{job="foo",grpc_type="server_stream"}[10m])) by (grpc_service)
 /
sum(rate(grpc_server_started_total{job="foo",grpc_type="server_stream"}[10m])) by (grpc_service)

For job="foo" what is the grpc_service-wide 10m average of messages returned for all server_stream RPCs. This allows you to track the stream sizes returned by your system, e.g. allows you to track when clients started to send "wide" queries that ret Note the divisor is the number of started RPCs, in order to account for in-flight requests.

99%-tile latency of unary requests

histogram_quantile(0.99, 
  sum(rate(grpc_server_handling_seconds_bucket{job="foo",grpc_type="unary"}[5m])) by (grpc_service,le)
)

For job="foo", returns an 99%-tile quantile estimation of the handling time of RPCs per service. Please note the 5m rate, this means that the quantile estimation will take samples in a rolling 5m window. When combined with other quantiles (e.g. 50%, 90%), this query gives you tremendous insight into the responsiveness of your system (e.g. impact of caching).

percentage of slow unary queries (>250ms)

100.0 - (
sum(rate(grpc_server_handling_seconds_bucket{job="foo",grpc_type="unary",le="0.25"}[5m])) by (grpc_service)
 / 
sum(rate(grpc_server_handling_seconds_count{job="foo",grpc_type="unary"}[5m])) by (grpc_service)
) * 100.0

For job="foo" calculate the by-grpc_service fraction of slow requests that took longer than 0.25 seconds. This query is relatively complex, since the Prometheus aggregations use le (less or equal) buckets, meaning that counting "fast" requests fractions is easier. However, simple maths helps. This is an example of a query you would like to alert on in your system for SLA violations, e.g. "less than 1% of requests are slower than 250ms".

Documentation

Index

• Constants
• func MustCounter(v syncint64.Counter, err error) syncint64.Counter
• func MustHistogram(v syncfloat64.Histogram, err error) syncfloat64.Histogram
• func SemVersion() string
• func Version() string
• type ClientMetrics
  • func NewClientMetrics(counterOpts ...Option) *ClientMetrics
  • func (m *ClientMetrics) StreamClientInterceptor() ...
  • func (m *ClientMetrics) UnaryClientInterceptor() func(ctx context.Context, method string, req, reply interface{}, ...) error
• type Option
  • func WithBucket(bucket []float64) Option
  • func WithConstLabels(l ...attribute.KeyValue) Option
  • func WithMeterProvider(provider metric.MeterProvider) Option
  • func WithServerHandledHistogram(v bool) Option
• type ServerMetrics
  • func NewServerMetrics(counterOpts ...Option) *ServerMetrics
  • func (m *ServerMetrics) StreamServerInterceptor() ...
  • func (m *ServerMetrics) UnaryServerInterceptor() ...

Constants

const (
	AttrType    = "grpc_type"
	AttrService = "grpc_service"
	AttrMethod  = "grpc_method"
	AttrCode    = "grpc_code"
)

const (
	Unary        grpcType = "unary"
	ClientStream grpcType = "client_stream"
	ServerStream grpcType = "server_stream"
	BidiStream   grpcType = "bidi_stream"
)

Functions

func MustCounter

func MustCounter(v syncint64.Counter, err error) syncint64.Counter

func MustHistogram

func MustHistogram(v syncfloat64.Histogram, err error) syncfloat64.Histogram

func SemVersion

func SemVersion() string

SemVersion is the semantic version to be supplied to tracer/meter creation.

func Version

func Version() string

Version is the current release version of the otelhttp instrumentation.

Types

type ClientMetrics

type ClientMetrics struct {
	// contains filtered or unexported fields
}

ClientMetrics represents a collection of metrics to be registered on a Prometheus metrics registry for a gRPC client.

func NewClientMetrics

func NewClientMetrics(counterOpts ...Option) *ClientMetrics

NewClientMetrics returns a ClientMetrics object. Use a new instance of ClientMetrics when not using the default Prometheus metrics registry, for example when wanting to control which metrics are added to a registry as opposed to automatically adding metrics via init functions.

func (*ClientMetrics) StreamClientInterceptor

func (m *ClientMetrics) StreamClientInterceptor() func(ctx context.Context, desc *grpc.StreamDesc, cc *grpc.ClientConn, method string, streamer grpc.Streamer, opts ...grpc.CallOption) (grpc.ClientStream, error)

StreamClientInterceptor is a gRPC client-side interceptor that provides Prometheus monitoring for Streaming RPCs.

func (*ClientMetrics) UnaryClientInterceptor

func (m *ClientMetrics) UnaryClientInterceptor() func(ctx context.Context, method string, req, reply interface{}, cc *grpc.ClientConn, invoker grpc.UnaryInvoker, opts ...grpc.CallOption) error

UnaryClientInterceptor is a gRPC client-side interceptor that provides Prometheus monitoring for Unary RPCs.

type Option

type Option interface {
	// contains filtered or unexported methods
}

Option interface used for setting optional config properties.

func WithBucket

func WithBucket(bucket []float64) Option

WithBucket for grpc_server_handling_seconds metric

func WithConstLabels

func WithConstLabels(l ...attribute.KeyValue) Option

func WithMeterProvider

func WithMeterProvider(provider metric.MeterProvider) Option

WithMeterProvider specifies a meter provider to use for creating a meter. If none is specified, the global provider is used.

func WithServerHandledHistogram

func WithServerHandledHistogram(v bool) Option

type ServerMetrics

type ServerMetrics struct {
	// contains filtered or unexported fields
}

ServerMetrics represents a collection of metrics to be registered on a Prometheus metrics registry for a gRPC server.

func NewServerMetrics

func NewServerMetrics(counterOpts ...Option) *ServerMetrics

NewServerMetrics returns a ServerMetrics object. Use a new instance of ServerMetrics when not using the default Prometheus metrics registry, for example when wanting to control which metrics are added to a registry as opposed to automatically adding metrics via init functions.

func (*ServerMetrics) StreamServerInterceptor

func (m *ServerMetrics) StreamServerInterceptor() func(srv interface{}, ss grpc.ServerStream, info *grpc.StreamServerInfo, handler grpc.StreamHandler) error

StreamServerInterceptor is a gRPC server-side interceptor that provides Prometheus monitoring for Streaming RPCs.

func (*ServerMetrics) UnaryServerInterceptor

func (m *ServerMetrics) UnaryServerInterceptor() func(ctx context.Context, req interface{}, info *grpc.UnaryServerInfo, handler grpc.UnaryHandler) (interface{}, error)

UnaryServerInterceptor is a gRPC server-side interceptor that provides Prometheus monitoring for Unary RPCs.