Clue provides a set of Go packages for instrumenting microservices. The emphasis is on simplicity and ease of use. Although not a requirement, Clue works best when used in microservices written using Goa.
Clue covers the following topics:
- Instrumentation: the clue package configures OpenTelemetry for service instrumentation.
- Logging: the log package provides a context-based logging API that intelligently selects what to log.
- Health checks: the health package provides a simple way for services to expose a health check endpoint.
- Dependency mocks: the mock package provides a way to mock downstream dependencies for testing.
- Debugging: the debug package makes it possible to troubleshoot and profile services at runtime.
Clue's goal is to provide all the anciallary functionality required to efficiently operate a microservice style architecture including instrumentation, logging, debugging and health checks. Clue is not a framework and does not provide functionality that is already available in the standard library or in other packages. For example, Clue does not provide a HTTP router or a HTTP server implementation. Instead, Clue provides a way to instrument existing HTTP or gRPC servers and clients using the standard library and the OpenTelemetry API.
-
The
cluepackage provides a simple API for configuring OpenTelemetry instrumentation. The package also provides a way to configure the log package to automatically annotate log messages with trace and span IDs. The package also implements a dynamic trace sampler that can be used to sample traces based on a target maximum number of traces per second. -
The
logpackage offers a streamlined, context-based structured logger that efficiently buffers log messages. It smartly determines the optimal time to flush these messages to the underlying logger. In its default configuration, the log package flushes logs upon the logging of an error or when a request is traced. This design choice minimizes logging overhead for untraced requests, ensuring efficient logging operations. -
The
healthpackage provides a simple way to expose a health check endpoint that can be used by orchestration systems such as Kubernetes to determine whether a service is ready to receive traffic. The package implements the concept of checkers that can be used to implement custom health checks with a default implementation that relies on the ability to ping downstream dependencies. -
The
mockpackage provides a way to mock downstream dependencies for testing. The package provides a tool that generates mock implementations of interfaces and a way to configure the generated mocks to validate incoming payloads and return canned responses. -
The
debugpackage provides a way to dynamically control the log level of a running service. The package also provides a way to expose the pprof Go profiling endpoints and a way to expose the log level control endpoint.
Clue requires Go 1.21 or later. Install the packages required for your
application using go get, for example:
go get goa.design/clue/clue
go get goa.design/clue/log
go get goa.design/clue/health
go get goa.design/clue/mock
go get goa.design/clue/debugThe following snippet illustrates how to use clue to instrument a HTTP server:
ctx := log.Context(context.Background(), // Create a clue logger context.
log.WithFunc(log.Span)) // Log trace and span IDs.
metricExporter := stdoutmetric.New() // Create metric and span exporters.
spanExporter := stdouttrace.New()
cfg := clue.NewConfig(ctx, "service", "1.0.0", metricExporter, spanExporter)
clue.ConfigureOpenTelemetry(ctx, cfg) // Configure OpenTelemetry.
handler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("Hello, World!"))
}) // Create HTTP handler.
handler = otelhttp.NewHandler(handler, "service" ) // Instrument handler.
handler = debug.HTTP()(handler) // Setup debug log level control.
handler = log.HTTP(ctx)(handler) // Add logger to request context and log requests.
mux := http.NewServeMux() // Create HTTP mux.
mux.HandleFunc("/", handler) // Mount handler.
debug.MountDebugLogEnabler(mux) // Mount debug log level control handler.
debug.MountDebugPprof(mux) // Mount pprof handlers.
mux.HandleFunc("/health",
health.NewHandler(health.NewChecker())) // Mount health check handler.
http.ListenAndServe(":8080", mux) // Start HTTP server.Similarly, the following snippet illustrates how to instrument a gRPC server:
ctx := log.Context(context.Background(), // Create a clue logger context.
log.WithFunc(log.Span)) // Log trace and span IDs.
metricExporter := stdoutmetric.New()
spanExporter := stdouttrace.New()
cfg := clue.NewConfig(ctx, "service", "1.0.0", metricExporter, spanExporter)
clue.ConfigureOpenTelemetry(ctx, cfg) // Configure OpenTelemetry.
svr := grpc.NewServer(
grpc.ChainUnaryInterceptor(
log.UnaryServerInterceptor(ctx), // Add logger to request context and log requests.
debug.UnaryServerInterceptor()), // Enable debug log level control
grpc.StatsHandler(otelgrpc.NewServerHandler()), // Instrument server.
)Note that in the case of gRPC, a separate HTTP server is required to expose the debug log level control, pprof and health check endpoints:
mux := http.NewServeMux() // Create HTTP mux.
debug.MountDebugLogEnabler(mux) // Mount debug log level control handler.
debug.MountDebugPprof(mux) // Mount pprof handlers.
mux.HandleFunc("/health",
health.NewHandler(health.NewChecker())) // Mount health check handler.
go http.ListenAndServe(":8081", mux) // Start HTTP server.Exporter are responsible for exporting telemetry data to a backend. The OpenTelemetry Exporters documentation provides a list of available exporters.
For example, configuring an OTLP compliant span exporter can be done as follows:
import "go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracegrpc"
// ...
spanExporter, err := otlptracegrpc.New(
context.Background(),
otlptracegrpc.WithEndpoint("localhost:4317"),
otlptracegrpc.WithTLSCredentials(insecure.NewCredentials()))While configuring an OTLP compliant metric exporters can be done as follows:
import "go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc"
// ...
metricExporter, err := otlpmetricgrpc.New(
context.Background(),
otlpmetricgrpc.WithEndpoint("localhost:4317"),
otlpmetricgrpc.WithTLSCredentials(insecure.NewCredentials()))These exporters can then be used to configure Clue:
// Configure OpenTelemetry.
cfg := clue.NewConfig(ctx, "service", "1.0.0", metricExporter, spanExporter)
clue.ConfigureOpenTelemetry(ctx, cfg)HTTP clients can be instrumented using the Clue log and OpenTelemetry otelhttptrace packages. The log.Client function wraps a HTTP transport and logs the request and response. The otelhttptrace.Client function wraps a HTTP transport and adds OpenTelemetry tracing to the request.
import (
"context"
"net/http"
"net/http/httptrace"
"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp"
"go.opentelemetry.io/contrib/instrumentation/net/http/otelhttptrace"
"goa.design/clue/log"
)
// ...
httpc := &http.Client{
Transport: log.Client(
otelhttp.NewTransport(
http.DefaultTransport,
otelhttp.WithClientTrace(func(ctx context.Context) *httptrace.ClientTrace {
return otelhttptrace.NewClientTrace(ctx)
}),
),
),
}Similarly, gRPC clients can be instrumented using the Clue log and OpenTelemtry otelgrpc packages.
import (
"context"
"go.opentelemetry.io/contrib/instrumentation/google.golang.org/grpc/otelgrpc"
"goa.design/clue/log"
)
// ...
grpcconn, err := grpc.DialContext(ctx,
"localhost:8080",
grpc.WithUnaryInterceptor(log.UnaryClientInterceptor()),
grpc.WithStatsHandler(otelgrpc.NewClientHandler()))
)Clue relies on the OpenTelemetry API for creating custom instrumentation. The following snippet illustrates how to create a custom counter and span:
// ... configure OpenTelemetry like in example above
clue.ConfigureOpenTelemetry(ctx, cfg)
// Create a meter and tracer
meter := otel.Meter("mymeter")
tracer := otel.Tracer("mytracer")
// Create a counter
counter, err := meter.Int64Counter("my.counter",
metric.WithDescription("The number of times the service has been called"),
metric.WithUnit("{call}"))
if err != nil {
log.Fatalf("failed to create counter: %s", err)
}
handler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
// Create a span
ctx, span := tracer.Start(r.Context(), "myhandler")
defer span.End()
// ... do something
// Add custom attributes to span and counter
attr := attribute.Int("myattr", 42)
span.SetAttributes(attr)
counter.Add(ctx, 1, metric.WithAttributes(attr))
// ... do something else
if _, err := w.Write([]byte("Hello, World!")); err != nil {
log.Errorf(ctx, err, "failed to write response")
}
})The log package provides a Goa endpoint middleware that adds the service and
method names to the logger context. The debug package provides a Goa endpoint
middleware that logs the request and response payloads when debug logging is
enabled. The example below is a snippet extracted from the main function of
the
genforecaster
service:
svc := forecaster.New(wc)
endpoints := genforecaster.NewEndpoints(svc)
endpoints.Use(debug.LogPayloads())
endpoints.Use(log.Endpoint)The weather example illustrates how to use Clue to instrument a system of Goa microservices. The example comes with a set of scripts that can be used to install all necessary dependencies including the SigNoz instrumentation backend. See the README for more information.
The v1.0.0 release of clue is a major release that introduces breaking
changes. The following sections describe the changes and how to migrate.
The clue package provides a cohesive API for both metrics and tracing,
effectively replacing the previous metrics and trace packages. The
traditional Context function, utilized in the metrics and trace packages
for setting up telemetry, has been deprecated. In its place, the clue package
introduces the NewConfig function, which generates a Config object used in
conjunction with the ConfigureOpenTelemetry function to facilitate telemetry
setup.
v0.x:
ctx := log.Context(context.Background())
traceExporter := tracestdout.New()
metricsExporter := metricstdout.New()
ctx = trace.Context(ctx, "service", traceExporter)
ctx = metrics.Context(ctx, "service", metricsExporter)v1.x:
ctx := log.Context(context.Background())
traceExporter := tracestdout.New()
metricsExporter := metricstdout.New()
cfg := clue.NewConfig(ctx, "service", "1.0.0", metricsExporter, traceExporter)
clue.ConfigureOpenTelemetry(ctx, cfg)Instrumenting a HTTP service is now done using the standard otelhttp package:
v0.x:
handler = trace.HTTP(ctx)(handler)
handler = metrics.HTTP(ctx)(handler)
http.ListenAndServe(":8080", handler)v1.x:
http.ListenAndServe(":8080", otelhttp.NewHandler("service", handler))Similarly, instrumenting a gRPC service is now done using the standard
otelgrpc package. v1.x also switches to using a gRPC stats handler instead of
interceptors:
v0.x:
grpcsvr := grpc.NewServer(
grpcmiddleware.WithUnaryServerChain(
trace.UnaryServerInterceptor(ctx),
metrics.UnaryServerInterceptor(ctx),
),
grpcmiddleware.WithStreamServerChain(
trace.StreamServerInterceptor(ctx),
metrics.StreamServerInterceptor(ctx),
),
)v1.x:
grpcsvr := grpc.NewServer(grpc.StatsHandler(otelgrpc.NewServerHandler()))The otel Goa plugin leverages the OpenTelemetry API to annotate spans and
metrics with HTTP routes.
v0.x:
mux := goahttp.NewMuxer()
ctx = metrics.Context(ctx, genfront.ServiceName,
metrics.WithRouteResolver(func(r *http.Request) string {
return mux.ResolvePattern(r)
}),
)v1.x:
package design
import (
. "goa.design/goa/v3/dsl"
_ "goa.design/plugins/v3/clue"
)See Contributing