logtracing

logtracing

This package is a toolkit for log-based tracing. It provides several APIs for adding traces for the application and logging them in a standard format.

Usage

Import the package

import 'github.com/theplant/appkit/logtracing'

Inside a function, you can use these three APIs to track it:

• StartSpan(context.Context, string) to start a span
• EndSpan(context.Context, error) to end the span, also log it in an agreed format
• RecordPanic(context.Context) to record the panic into the span

func DoWork(ctx context.Context) err error {
	ctx, _ := logtracing.StartSpan(ctx, "<span.context>")
	defer func() { logtracing.EndSpan(ctx, err) }()
	defer RecordPanic(ctx)
}

It will create a new span, record the error, and log the span with the logger in context.

You can append key-values to an active span with AppendSpanKvs:

func DoWork(ctx context.Context) err error {
	ctx, _ := logtracing.StartSpan(ctx, "<span.context>")
	defer func() { logtracing.EndSpan(ctx, err) }()
	defer RecordPanic(ctx)

	logtracing.AppendSpanKVs(ctx,
		"service", "greeter",
	)
}

If you want to append key-values to all spans, you can append the key-values to the context with ContextWithKVs:

func DoWork(ctx context.Context) err error {
	// all the spans within this context will contain `"key": "value"`
	ctx = logtracing.ContextWithKVs(ctx, "key", "value")

	ctx, _ = logtracing.StartSpan(ctx, "test")
	defer func() { logtracing.EndSpan(ctx, err) }()
	defer RecordPanic(ctx)
}

For flexibility, the package provides the following APIs to manipulate the span directly:

• (*span).AppendKVs(...interface{})
• (*span).RecordError(err error)
• (*span).RecordPanic(panic interface{}})
• (*span).End()
• LogSpan(ctx context, s *span)

Key-values

Common

With logtracing.StartSpan and logtracing.EndSpan, these are automatically added to the span:

• ts
• msg
• trace.id
• span.id
• span.context
• span.dur_ms
• span.parent_id

If the span records an error, these key-values will be added:

• span.err
• span.err_type
• span.with_err

Reference

• span.type: This usually describes "how" the span is sent/received, what kind of underlying API or network transport is used, eg http, sql, grpc, dns, aws.<service>
• span.role: This describes what role this span plays in an interaction with another service, eg for HTTP a client makes a request to a server. Or in a queue, a producer adds work to the queue, that is consumed by a consumer.

Data that is for specific types should use the type as a prefix when adding supplementary keys-values, For example:

• With span.type=http, the HTTP method would be logged as http.method, the response status would be logged as http.status=200.

• AWS S3 API calls (span.type=aws.s3), use s3 or aws.s3 as a prefix, an object used in the API call could be logged as s3.object=<object key> (or aws.s3.object=<object key>)

XMLRPC

• span.type: xmlrpc
• span.role: client
• xmlrpc.method: the service method
• http.method: post
• http.url: the server URL
• xmlrpc.fault_string: only exists when getting a fault error

GRPC

• span.type: grpc
• span.role: client or server
• grpc.service: gRPC service name
• grpc.method: gRPC method name
• grpc.code: gRPC response status code

The package provides server and client interceptors to log these key-values automatically.

HTTP

For the client requests:

• span.type: http
• span.role: client
• http.url: the request full URL
• http.method: the request method
• http.status: the response status, only exists when getting the response successfully

For the server requests:

• span.type: http
• span.role: server
• http.path: the request path
• http.method: the request method
• http.user_agent: the request user agent
• http.client_ip: the request client IP
• http.status: the response status

Queue

• span.type: queue
• span.role: consumer or producer

Function

For internal functions:

• span.role: internal

Export span data

You can export span data to an expected destination such as Honeycomb by registering an exporter.

1. Implement the Exporter interface:

   type Exporter interface {
   	ExportSpan(s *SpanData)
   }
   

2. Initialize and register your exporter:

   exporter := <your exporter>
   logtracing.RegisterExporter(exporter)
   

When you register exporters, logtracing will pass sampled spans to the exporters after logging.

Honeycomb exporter

You can use this exporter to send spans to Honeycomb. The sent events are in the same format as the logged events.

1. Import the package:

   import github.com/theplant/appkit/logtracing/exporters/honeycomb
   

2. Initialize the exporter:

   exporter, err := honeycomb.NewExporter(libhoney.Config{
   	WriteKey:     "mock",
   	Dataset:      "mock",
   	Transmission: mockSender,
   })
   ```
   

3. Register the exporter:

   logtracing.RegisterExporter(expoter)
   

How to migrate from util/trace.go

1. Use logtracing.TraceFunc to replace util.Lt
2. Use logtracing.AppendSpanKVs to replace util.AppendKVs
3. Use logtracing.TraceHTTPRequest to replace util.LtRequest
4. Use logtracing.HTTPTransport to replace util.LtTransport
5. Use logtracing.XMLRPCClientKVs to replace util.XMLRpcKVs