1
0
mirror of https://github.com/golang/go synced 2024-11-25 14:37:56 -07:00
go/test/heapsampling.go
Michael Anthony Knyszek 70d499cd89 test/heapsampling.go: slow down allocation rate and reduce iterations
As far as I can tell, this test suffers from #52433. For some reason,
this seems to become more of a problem on the windows/386 than anywhere
else. This CL is an attempt at a mitigation by slowing down the
allocation rate by inserting runtime.Gosched call in the inner loop. It
also cuts the iteration count which should help too (as less memory is
allocated in total), but the main motivation is to make sure the test
doesn't take too long to run.

Fixes #49564.

Change-Id: I8cc622b06a69cdfa66f680a30e1ccf334eea2164
Reviewed-on: https://go-review.googlesource.com/c/go/+/408825
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
2022-05-27 21:36:06 +00:00

315 lines
9.7 KiB
Go

// run
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Test heap sampling logic.
package main
import (
"fmt"
"math"
"runtime"
)
var a16 *[16]byte
var a512 *[512]byte
var a256 *[256]byte
var a1k *[1024]byte
var a16k *[16 * 1024]byte
var a17k *[17 * 1024]byte
var a18k *[18 * 1024]byte
// This test checks that heap sampling produces reasonable results.
// Note that heap sampling uses randomization, so the results vary for
// run to run. To avoid flakes, this test performs multiple
// experiments and only complains if all of them consistently fail.
func main() {
// Sample at 16K instead of default 512K to exercise sampling more heavily.
runtime.MemProfileRate = 16 * 1024
if err := testInterleavedAllocations(); err != nil {
panic(err.Error())
}
if err := testSmallAllocations(); err != nil {
panic(err.Error())
}
}
// Repeatedly exercise a set of allocations and check that the heap
// profile collected by the runtime unsamples to a reasonable
// value. Because sampling is based on randomization, there can be
// significant variability on the unsampled data. To account for that,
// the testcase allows for a 10% margin of error, but only fails if it
// consistently fails across three experiments, avoiding flakes.
func testInterleavedAllocations() error {
const iters = 50000
// Sizes of the allocations performed by each experiment.
frames := []string{"main.allocInterleaved1", "main.allocInterleaved2", "main.allocInterleaved3"}
// Pass if at least one of three experiments has no errors. Use a separate
// function for each experiment to identify each experiment in the profile.
allocInterleaved1(iters)
if checkAllocations(getMemProfileRecords(), frames[0:1], iters, allocInterleavedSizes) == nil {
// Passed on first try, report no error.
return nil
}
allocInterleaved2(iters)
if checkAllocations(getMemProfileRecords(), frames[0:2], iters, allocInterleavedSizes) == nil {
// Passed on second try, report no error.
return nil
}
allocInterleaved3(iters)
// If it fails a third time, we may be onto something.
return checkAllocations(getMemProfileRecords(), frames[0:3], iters, allocInterleavedSizes)
}
var allocInterleavedSizes = []int64{17 * 1024, 1024, 18 * 1024, 512, 16 * 1024, 256}
// allocInterleaved stress-tests the heap sampling logic by interleaving large and small allocations.
func allocInterleaved(n int) {
for i := 0; i < n; i++ {
// Test verification depends on these lines being contiguous.
a17k = new([17 * 1024]byte)
a1k = new([1024]byte)
a18k = new([18 * 1024]byte)
a512 = new([512]byte)
a16k = new([16 * 1024]byte)
a256 = new([256]byte)
// Test verification depends on these lines being contiguous.
// Slow down the allocation rate to avoid #52433.
runtime.Gosched()
}
}
func allocInterleaved1(n int) {
allocInterleaved(n)
}
func allocInterleaved2(n int) {
allocInterleaved(n)
}
func allocInterleaved3(n int) {
allocInterleaved(n)
}
// Repeatedly exercise a set of allocations and check that the heap
// profile collected by the runtime unsamples to a reasonable
// value. Because sampling is based on randomization, there can be
// significant variability on the unsampled data. To account for that,
// the testcase allows for a 10% margin of error, but only fails if it
// consistently fails across three experiments, avoiding flakes.
func testSmallAllocations() error {
const iters = 50000
// Sizes of the allocations performed by each experiment.
sizes := []int64{1024, 512, 256}
frames := []string{"main.allocSmall1", "main.allocSmall2", "main.allocSmall3"}
// Pass if at least one of three experiments has no errors. Use a separate
// function for each experiment to identify each experiment in the profile.
allocSmall1(iters)
if checkAllocations(getMemProfileRecords(), frames[0:1], iters, sizes) == nil {
// Passed on first try, report no error.
return nil
}
allocSmall2(iters)
if checkAllocations(getMemProfileRecords(), frames[0:2], iters, sizes) == nil {
// Passed on second try, report no error.
return nil
}
allocSmall3(iters)
// If it fails a third time, we may be onto something.
return checkAllocations(getMemProfileRecords(), frames[0:3], iters, sizes)
}
// allocSmall performs only small allocations for sanity testing.
func allocSmall(n int) {
for i := 0; i < n; i++ {
// Test verification depends on these lines being contiguous.
a1k = new([1024]byte)
a512 = new([512]byte)
a256 = new([256]byte)
// Slow down the allocation rate to avoid #52433.
runtime.Gosched()
}
}
// Three separate instances of testing to avoid flakes. Will report an error
// only if they all consistently report failures.
func allocSmall1(n int) {
allocSmall(n)
}
func allocSmall2(n int) {
allocSmall(n)
}
func allocSmall3(n int) {
allocSmall(n)
}
// checkAllocations validates that the profile records collected for
// the named function are consistent with count contiguous allocations
// of the specified sizes.
// Check multiple functions and only report consistent failures across
// multiple tests.
// Look only at samples that include the named frames, and group the
// allocations by their line number. All these allocations are done from
// the same leaf function, so their line numbers are the same.
func checkAllocations(records []runtime.MemProfileRecord, frames []string, count int64, size []int64) error {
objectsPerLine := map[int][]int64{}
bytesPerLine := map[int][]int64{}
totalCount := []int64{}
// Compute the line number of the first allocation. All the
// allocations are from the same leaf, so pick the first one.
var firstLine int
for ln := range allocObjects(records, frames[0]) {
if firstLine == 0 || firstLine > ln {
firstLine = ln
}
}
for _, frame := range frames {
var objectCount int64
a := allocObjects(records, frame)
for s := range size {
// Allocations of size size[s] should be on line firstLine + s.
ln := firstLine + s
objectsPerLine[ln] = append(objectsPerLine[ln], a[ln].objects)
bytesPerLine[ln] = append(bytesPerLine[ln], a[ln].bytes)
objectCount += a[ln].objects
}
totalCount = append(totalCount, objectCount)
}
for i, w := range size {
ln := firstLine + i
if err := checkValue(frames[0], ln, "objects", count, objectsPerLine[ln]); err != nil {
return err
}
if err := checkValue(frames[0], ln, "bytes", count*w, bytesPerLine[ln]); err != nil {
return err
}
}
return checkValue(frames[0], 0, "total", count*int64(len(size)), totalCount)
}
// checkValue checks an unsampled value against its expected value.
// Given that this is a sampled value, it will be unexact and will change
// from run to run. Only report it as a failure if all the values land
// consistently far from the expected value.
func checkValue(fname string, ln int, testName string, want int64, got []int64) error {
if got == nil {
return fmt.Errorf("Unexpected empty result")
}
min, max := got[0], got[0]
for _, g := range got[1:] {
if g < min {
min = g
}
if g > max {
max = g
}
}
margin := want / 10 // 10% margin.
if min > want+margin || max < want-margin {
return fmt.Errorf("%s:%d want %s in [%d: %d], got %v", fname, ln, testName, want-margin, want+margin, got)
}
return nil
}
func getMemProfileRecords() []runtime.MemProfileRecord {
// Force the runtime to update the object and byte counts.
// This can take up to two GC cycles to get a complete
// snapshot of the current point in time.
runtime.GC()
runtime.GC()
// Find out how many records there are (MemProfile(nil, true)),
// allocate that many records, and get the data.
// There's a race—more records might be added between
// the two calls—so allocate a few extra records for safety
// and also try again if we're very unlucky.
// The loop should only execute one iteration in the common case.
var p []runtime.MemProfileRecord
n, ok := runtime.MemProfile(nil, true)
for {
// Allocate room for a slightly bigger profile,
// in case a few more entries have been added
// since the call to MemProfile.
p = make([]runtime.MemProfileRecord, n+50)
n, ok = runtime.MemProfile(p, true)
if ok {
p = p[0:n]
break
}
// Profile grew; try again.
}
return p
}
type allocStat struct {
bytes, objects int64
}
// allocObjects examines the profile records for samples including the
// named function and returns the allocation stats aggregated by
// source line number of the allocation (at the leaf frame).
func allocObjects(records []runtime.MemProfileRecord, function string) map[int]allocStat {
a := make(map[int]allocStat)
for _, r := range records {
var pcs []uintptr
for _, s := range r.Stack0 {
if s == 0 {
break
}
pcs = append(pcs, s)
}
frames := runtime.CallersFrames(pcs)
line := 0
for {
frame, more := frames.Next()
name := frame.Function
if line == 0 {
line = frame.Line
}
if name == function {
allocStat := a[line]
allocStat.bytes += r.AllocBytes
allocStat.objects += r.AllocObjects
a[line] = allocStat
}
if !more {
break
}
}
}
for line, stats := range a {
objects, bytes := scaleHeapSample(stats.objects, stats.bytes, int64(runtime.MemProfileRate))
a[line] = allocStat{bytes, objects}
}
return a
}
// scaleHeapSample unsamples heap allocations.
// Taken from src/cmd/pprof/internal/profile/legacy_profile.go
func scaleHeapSample(count, size, rate int64) (int64, int64) {
if count == 0 || size == 0 {
return 0, 0
}
if rate <= 1 {
// if rate==1 all samples were collected so no adjustment is needed.
// if rate<1 treat as unknown and skip scaling.
return count, size
}
avgSize := float64(size) / float64(count)
scale := 1 / (1 - math.Exp(-avgSize/float64(rate)))
return int64(float64(count) * scale), int64(float64(size) * scale)
}