feat: Add initial profiling support (#626)

Co-authored-by: Anton Ovchinnikov <anton@tonyo.info>
Co-authored-by: Michi Hoffmann <cleptric@users.noreply.github.com>

CHANGELOG.md

@@ -1,6 +1,10 @@
 # Changelog
 
-## Unreleased
+## Unrelesed
+
+### Features
+
+- Initial alpha support for profiling [#626](https://github.com/getsentry/sentry-go/pull/626)
 
 ### Bug fixes
 

_examples/profiling/main.go

@@ -0,0 +1,80 @@
+//	go run main.go
+//
+// To actually report events to Sentry, set the DSN either by editing the
+// appropriate line below or setting the environment variable SENTRY_DSN to
+// match the DSN of your Sentry project.
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"runtime"
+	"sync"
+	"time"
+
+	"github.com/getsentry/sentry-go"
+)
+
+func main() {
+	err := sentry.Init(sentry.ClientOptions{
+		// Either set your DSN here or set the SENTRY_DSN environment variable.
+		Dsn: "",
+		// Enable printing of SDK debug messages.
+		// Useful when getting started or trying to figure something out.
+		Debug:              true,
+		EnableTracing:      true,
+		TracesSampleRate:   1.0,
+		ProfilesSampleRate: 1.0,
+	})
+
+	// Flush buffered events before the program terminates.
+	// Set the timeout to the maximum duration the program can afford to wait.
+	defer sentry.Flush(2 * time.Second)
+
+	if err != nil {
+		log.Fatalf("sentry.Init: %s", err)
+	}
+	ctx := context.Background()
+	tx := sentry.StartTransaction(ctx, "top")
+
+	fmt.Println("Finding prime numbers")
+	var wg sync.WaitGroup
+	wg.Add(10)
+	for i := 0; i < 10; i++ {
+		go func(num int) {
+			span := tx.StartChild(fmt.Sprintf("Goroutine %d", num))
+			defer span.Finish()
+			for i := 0; i < num; i++ {
+				_ = findPrimeNumber(50000)
+				runtime.Gosched() // we need to manually yield this busy loop
+			}
+			fmt.Printf("routine %d done\n", num)
+			wg.Done()
+		}(i)
+	}
+	wg.Wait()
+	fmt.Println("all")
+	tx.Finish()
+}
+
+func findPrimeNumber(n int) int {
+	count := 0
+	a := 2
+	for count < n {
+		b := 2
+		prime := true // to check if found a prime
+		for b*b <= a {
+			if a%b == 0 {
+				prime = false
+				break
+			}
+			b++
+		}
+		if prime {
+			count++
+		}
+		a++
+	}
+	return a - 1
+}

client.go

@@ -130,6 +130,9 @@ type ClientOptions struct {
 	TracesSampleRate float64
 	// Used to customize the sampling of traces, overrides TracesSampleRate.
 	TracesSampler TracesSampler
+	// The sample rate for profiling traces in the range [0.0, 1.0].
+	// This is relative to TracesSampleRate - it is a ratio of profiled traces out of all sampled traces.
+	ProfilesSampleRate float64
 	// List of regexp strings that will be used to match against event's message
 	// and if applicable, caught errors type and value.
 	// If the match is found, then a whole event will be dropped.
@@ -371,6 +374,7 @@ func (client *Client) AddEventProcessor(processor EventProcessor) {
 }
 
 // Options return ClientOptions for the current Client.
+// TODO don't access this internally to avoid creating a copy each time.
 func (client Client) Options() ClientOptions {
 	return client.options
 }
@@ -573,6 +577,7 @@ func (client *Client) processEvent(event *Event, hint *EventHint, scope EventMod
 
 func (client *Client) prepareEvent(event *Event, hint *EventHint, scope EventModifier) *Event {
 	if event.EventID == "" {
+		// TODO set EventID when the event is created, same as in other SDKs. It's necessary for profileTransaction.ID.
 		event.EventID = EventID(uuid())
 	}
 
@@ -640,6 +645,10 @@ func (client *Client) prepareEvent(event *Event, hint *EventHint, scope EventMod
 		}
 	}
 
+	if event.sdkMetaData.transactionProfile != nil {
+		event.sdkMetaData.transactionProfile.UpdateFromEvent(event)
+	}
+
 	return event
 }
 

example_transportwithhooks_test.go

@@ -60,6 +60,4 @@ func Example_transportWithHooks() {
 	defer sentry.Flush(2 * time.Second)
 
 	sentry.CaptureMessage("test")
-
-	// Output:
 }

interfaces.go

@@ -20,6 +20,8 @@ const transactionType = "transaction"
 // eventType is the type of an error event.
 const eventType = "event"
 
+const profileType = "profile"
+
 // Level marks the severity of the event.
 type Level string
 
@@ -237,7 +239,8 @@ type Exception struct {
 // SDKMetaData is a struct to stash data which is needed at some point in the SDK's event processing pipeline
 // but which shouldn't get send to Sentry.
 type SDKMetaData struct {
-	dsc DynamicSamplingContext
+	dsc                DynamicSamplingContext
+	transactionProfile *profileInfo
 }
 
 // Contains information about how the name of the transaction was determined.

internal/traceparser/README.md

@@ -0,0 +1,15 @@
+## Benchmark results
+
+```
+goos: windows
+goarch: amd64
+pkg: github.com/getsentry/sentry-go/internal/trace
+cpu: 12th Gen Intel(R) Core(TM) i7-12700K
+BenchmarkEqualBytes-20                  44323621                26.08 ns/op
+BenchmarkStringEqual-20                 60980257                18.27 ns/op
+BenchmarkEqualPrefix-20                 41369181                31.12 ns/op
+BenchmarkFullParse-20                     702012              1507 ns/op        1353.42 MB/s        1024 B/op          6 allocs/op
+BenchmarkFramesIterator-20               1229971               969.3 ns/op           896 B/op          5 allocs/op
+BenchmarkFramesReversedIterator-20       1271061               944.5 ns/op           896 B/op          5 allocs/op
+BenchmarkSplitOnly-20                    2250800               534.0 ns/op      3818.23 MB/s         128 B/op          1 allocs/op
+```

internal/traceparser/parser.go

@@ -0,0 +1,210 @@
+package traceparser
+
+import (
+	"bytes"
+	"strconv"
+)
+
+var blockSeparator = []byte("\n\n")
+var lineSeparator = []byte("\n")
+
+// Parses multi-stacktrace text dump produced by runtime.Stack([]byte, all=true).
+// The parser prioritizes performance but requires the input to be well-formed in order to return correct data.
+// See https://github.com/golang/go/blob/go1.20.4/src/runtime/mprof.go#L1191
+func Parse(data []byte) TraceCollection {
+	var it = TraceCollection{}
+	if len(data) > 0 {
+		it.blocks = bytes.Split(data, blockSeparator)
+	}
+	return it
+}
+
+type TraceCollection struct {
+	blocks [][]byte
+}
+
+func (it TraceCollection) Length() int {
+	return len(it.blocks)
+}
+
+// Returns the stacktrace item at the given index.
+func (it *TraceCollection) Item(i int) Trace {
+	// The first item may have a leading data separator and the last one may have a trailing one.
+	// Note: Trim() doesn't make a copy for single-character cutset under 0x80. It will just slice the original.
+	var data []byte
+	switch {
+	case i == 0:
+		data = bytes.TrimLeft(it.blocks[i], "\n")
+	case i == len(it.blocks)-1:
+		data = bytes.TrimRight(it.blocks[i], "\n")
+	default:
+		data = it.blocks[i]
+	}
+
+	var splitAt = bytes.IndexByte(data, '\n')
+	if splitAt < 0 {
+		return Trace{header: data}
+	}
+
+	return Trace{
+		header: data[:splitAt],
+		data:   data[splitAt+1:],
+	}
+}
+
+// Trace represents a single stacktrace block, identified by a Goroutine ID and a sequence of Frames.
+type Trace struct {
+	header []byte
+	data   []byte
+}
+
+var goroutinePrefix = []byte("goroutine ")
+
+// GoID parses the Goroutine ID from the header.
+func (t *Trace) GoID() (id uint64) {
+	if bytes.HasPrefix(t.header, goroutinePrefix) {
+		var line = t.header[len(goroutinePrefix):]
+		var splitAt = bytes.IndexByte(line, ' ')
+		if splitAt >= 0 {
+			id, _ = strconv.ParseUint(string(line[:splitAt]), 10, 64)
+		}
+	}
+	return id
+}
+
+// UniqueIdentifier can be used as a map key to identify the trace.
+func (t *Trace) UniqueIdentifier() []byte {
+	return t.data
+}
+
+func (t *Trace) Frames() FrameIterator {
+	var lines = bytes.Split(t.data, lineSeparator)
+	return FrameIterator{lines: lines, i: 0, len: len(lines)}
+}
+
+func (t *Trace) FramesReversed() ReverseFrameIterator {
+	var lines = bytes.Split(t.data, lineSeparator)
+	return ReverseFrameIterator{lines: lines, i: len(lines)}
+}
+
+const framesElided = "...additional frames elided..."
+
+// FrameIterator iterates over stack frames.
+type FrameIterator struct {
+	lines [][]byte
+	i     int
+	len   int
+}
+
+// Next returns the next frame, or nil if there are none.
+func (it *FrameIterator) Next() Frame {
+	return Frame{it.popLine(), it.popLine()}
+}
+
+func (it *FrameIterator) popLine() []byte {
+	switch {
+	case it.i >= it.len:
+		return nil
+	case string(it.lines[it.i]) == framesElided:
+		it.i++
+		return it.popLine()
+	default:
+		it.i++
+		return it.lines[it.i-1]
+	}
+}
+
+// HasNext return true if there are values to be read.
+func (it *FrameIterator) HasNext() bool {
+	return it.i < it.len
+}
+
+// LengthUpperBound returns the maximum number of elements this stacks may contain.
+// The actual number may be lower because of elided frames. As such, the returned value
+// cannot be used to iterate over the frames but may be used to reserve capacity.
+func (it *FrameIterator) LengthUpperBound() int {
+	return it.len / 2
+}
+
+// ReverseFrameIterator iterates over stack frames in reverse order.
+type ReverseFrameIterator struct {
+	lines [][]byte
+	i     int
+}
+
+// Next returns the next frame, or nil if there are none.
+func (it *ReverseFrameIterator) Next() Frame {
+	var line2 = it.popLine()
+	return Frame{it.popLine(), line2}
+}
+
+func (it *ReverseFrameIterator) popLine() []byte {
+	it.i--
+	switch {
+	case it.i < 0:
+		return nil
+	case string(it.lines[it.i]) == framesElided:
+		return it.popLine()
+	default:
+		return it.lines[it.i]
+	}
+}
+
+// HasNext return true if there are values to be read.
+func (it *ReverseFrameIterator) HasNext() bool {
+	return it.i > 1
+}
+
+// LengthUpperBound returns the maximum number of elements this stacks may contain.
+// The actual number may be lower because of elided frames. As such, the returned value
+// cannot be used to iterate over the frames but may be used to reserve capacity.
+func (it *ReverseFrameIterator) LengthUpperBound() int {
+	return len(it.lines) / 2
+}
+
+type Frame struct {
+	line1 []byte
+	line2 []byte
+}
+
+// UniqueIdentifier can be used as a map key to identify the frame.
+func (f *Frame) UniqueIdentifier() []byte {
+	// line2 contains file path, line number and program-counter offset from the beginning of a function
+	// e.g. C:/Users/name/scoop/apps/go/current/src/testing/testing.go:1906 +0x63a
+	return f.line2
+}
+
+var createdByPrefix = []byte("created by ")
+
+func (f *Frame) Func() []byte {
+	if bytes.HasPrefix(f.line1, createdByPrefix) {
+		return f.line1[len(createdByPrefix):]
+	}
+
+	var end = bytes.LastIndexByte(f.line1, '(')
+	if end >= 0 {
+		return f.line1[:end]
+	}
+
+	return f.line1
+}
+
+func (f *Frame) File() (path []byte, lineNumber int) {
+	var line = f.line2
+	if len(line) > 0 && line[0] == '\t' {
+		line = line[1:]
+	}
+
+	var splitAt = bytes.IndexByte(line, ' ')
+	if splitAt >= 0 {
+		line = line[:splitAt]
+	}
+
+	splitAt = bytes.LastIndexByte(line, ':')
+	if splitAt < 0 {
+		return line, 0
+	}
+
+	lineNumber, _ = strconv.Atoi(string(line[splitAt+1:]))
+	return line[:splitAt], lineNumber
+}