Black Lives Matter. Support the Equal Justice Initiative.

Source file src/runtime/trace/annotation.go

Documentation: runtime/trace

     1  // Copyright 2018 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package trace
     6  
     7  import (
     8  	"context"
     9  	"fmt"
    10  	"sync/atomic"
    11  	_ "unsafe"
    12  )
    13  
    14  type traceContextKey struct{}
    15  
    16  // NewTask creates a task instance with the type taskType and returns
    17  // it along with a Context that carries the task.
    18  // If the input context contains a task, the new task is its subtask.
    19  //
    20  // The taskType is used to classify task instances. Analysis tools
    21  // like the Go execution tracer may assume there are only a bounded
    22  // number of unique task types in the system.
    23  //
    24  // The returned end function is used to mark the task's end.
    25  // The trace tool measures task latency as the time between task creation
    26  // and when the end function is called, and provides the latency
    27  // distribution per task type.
    28  // If the end function is called multiple times, only the first
    29  // call is used in the latency measurement.
    30  //
    31  //   ctx, task := trace.NewTask(ctx, "awesomeTask")
    32  //   trace.WithRegion(ctx, "preparation", prepWork)
    33  //   // preparation of the task
    34  //   go func() {  // continue processing the task in a separate goroutine.
    35  //       defer task.End()
    36  //       trace.WithRegion(ctx, "remainingWork", remainingWork)
    37  //   }()
    38  func NewTask(pctx context.Context, taskType string) (ctx context.Context, task *Task) {
    39  	pid := fromContext(pctx).id
    40  	id := newID()
    41  	userTaskCreate(id, pid, taskType)
    42  	s := &Task{id: id}
    43  	return context.WithValue(pctx, traceContextKey{}, s), s
    44  
    45  	// We allocate a new task and the end function even when
    46  	// the tracing is disabled because the context and the detach
    47  	// function can be used across trace enable/disable boundaries,
    48  	// which complicates the problem.
    49  	//
    50  	// For example, consider the following scenario:
    51  	//   - trace is enabled.
    52  	//   - trace.WithRegion is called, so a new context ctx
    53  	//     with a new region is created.
    54  	//   - trace is disabled.
    55  	//   - trace is enabled again.
    56  	//   - trace APIs with the ctx is called. Is the ID in the task
    57  	//   a valid one to use?
    58  	//
    59  	// TODO(hyangah): reduce the overhead at least when
    60  	// tracing is disabled. Maybe the id can embed a tracing
    61  	// round number and ignore ids generated from previous
    62  	// tracing round.
    63  }
    64  
    65  func fromContext(ctx context.Context) *Task {
    66  	if s, ok := ctx.Value(traceContextKey{}).(*Task); ok {
    67  		return s
    68  	}
    69  	return &bgTask
    70  }
    71  
    72  // Task is a data type for tracing a user-defined, logical operation.
    73  type Task struct {
    74  	id uint64
    75  	// TODO(hyangah): record parent id?
    76  }
    77  
    78  // End marks the end of the operation represented by the Task.
    79  func (t *Task) End() {
    80  	userTaskEnd(t.id)
    81  }
    82  
    83  var lastTaskID uint64 = 0 // task id issued last time
    84  
    85  func newID() uint64 {
    86  	// TODO(hyangah): use per-P cache
    87  	return atomic.AddUint64(&lastTaskID, 1)
    88  }
    89  
    90  var bgTask = Task{id: uint64(0)}
    91  
    92  // Log emits a one-off event with the given category and message.
    93  // Category can be empty and the API assumes there are only a handful of
    94  // unique categories in the system.
    95  func Log(ctx context.Context, category, message string) {
    96  	id := fromContext(ctx).id
    97  	userLog(id, category, message)
    98  }
    99  
   100  // Logf is like Log, but the value is formatted using the specified format spec.
   101  func Logf(ctx context.Context, category, format string, args ...interface{}) {
   102  	if IsEnabled() {
   103  		// Ideally this should be just Log, but that will
   104  		// add one more frame in the stack trace.
   105  		id := fromContext(ctx).id
   106  		userLog(id, category, fmt.Sprintf(format, args...))
   107  	}
   108  }
   109  
   110  const (
   111  	regionStartCode = uint64(0)
   112  	regionEndCode   = uint64(1)
   113  )
   114  
   115  // WithRegion starts a region associated with its calling goroutine, runs fn,
   116  // and then ends the region. If the context carries a task, the region is
   117  // associated with the task. Otherwise, the region is attached to the background
   118  // task.
   119  //
   120  // The regionType is used to classify regions, so there should be only a
   121  // handful of unique region types.
   122  func WithRegion(ctx context.Context, regionType string, fn func()) {
   123  	// NOTE:
   124  	// WithRegion helps avoiding misuse of the API but in practice,
   125  	// this is very restrictive:
   126  	// - Use of WithRegion makes the stack traces captured from
   127  	//   region start and end are identical.
   128  	// - Refactoring the existing code to use WithRegion is sometimes
   129  	//   hard and makes the code less readable.
   130  	//     e.g. code block nested deep in the loop with various
   131  	//          exit point with return values
   132  	// - Refactoring the code to use this API with closure can
   133  	//   cause different GC behavior such as retaining some parameters
   134  	//   longer.
   135  	// This causes more churns in code than I hoped, and sometimes
   136  	// makes the code less readable.
   137  
   138  	id := fromContext(ctx).id
   139  	userRegion(id, regionStartCode, regionType)
   140  	defer userRegion(id, regionEndCode, regionType)
   141  	fn()
   142  }
   143  
   144  // StartRegion starts a region and returns a function for marking the
   145  // end of the region. The returned Region's End function must be called
   146  // from the same goroutine where the region was started.
   147  // Within each goroutine, regions must nest. That is, regions started
   148  // after this region must be ended before this region can be ended.
   149  // Recommended usage is
   150  //
   151  //     defer trace.StartRegion(ctx, "myTracedRegion").End()
   152  //
   153  func StartRegion(ctx context.Context, regionType string) *Region {
   154  	if !IsEnabled() {
   155  		return noopRegion
   156  	}
   157  	id := fromContext(ctx).id
   158  	userRegion(id, regionStartCode, regionType)
   159  	return &Region{id, regionType}
   160  }
   161  
   162  // Region is a region of code whose execution time interval is traced.
   163  type Region struct {
   164  	id         uint64
   165  	regionType string
   166  }
   167  
   168  var noopRegion = &Region{}
   169  
   170  // End marks the end of the traced code region.
   171  func (r *Region) End() {
   172  	if r == noopRegion {
   173  		return
   174  	}
   175  	userRegion(r.id, regionEndCode, r.regionType)
   176  }
   177  
   178  // IsEnabled reports whether tracing is enabled.
   179  // The information is advisory only. The tracing status
   180  // may have changed by the time this function returns.
   181  func IsEnabled() bool {
   182  	enabled := atomic.LoadInt32(&tracing.enabled)
   183  	return enabled == 1
   184  }
   185  
   186  //
   187  // Function bodies are defined in runtime/trace.go
   188  //
   189  
   190  // emits UserTaskCreate event.
   191  func userTaskCreate(id, parentID uint64, taskType string)
   192  
   193  // emits UserTaskEnd event.
   194  func userTaskEnd(id uint64)
   195  
   196  // emits UserRegion event.
   197  func userRegion(id, mode uint64, regionType string)
   198  
   199  // emits UserLog event.
   200  func userLog(id uint64, category, message string)
   201  

View as plain text