1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
|
// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
// testwrapper is a wrapper for retrying flaky tests. It is an alternative to
// `go test` and re-runs failed marked flaky tests (using the flakytest pkg). It
// takes different arguments than go test and requires the first positional
// argument to be the pattern to test.
package main
import (
"bufio"
"bytes"
"context"
"encoding/json"
"errors"
"fmt"
"io"
"log"
"os"
"os/exec"
"slices"
"sort"
"strings"
"time"
"tailscale.com/cmd/testwrapper/flakytest"
"tailscale.com/util/slicesx"
)
const (
maxAttempts = 3
)
type testAttempt struct {
pkg string // "tailscale.com/types/key"
testName string // "TestFoo"
outcome string // "pass", "fail", "skip"
logs bytes.Buffer
start, end time.Time
isMarkedFlaky bool // set if the test is marked as flaky
issueURL string // set if the test is marked as flaky
pkgFinished bool
}
// packageTests describes what to run.
// It's also JSON-marshalled to output for analysys tools to parse
// so the fields are all exported.
// TODO(bradfitz): move this type to its own types package?
type packageTests struct {
// Pattern is the package Pattern to run.
// Must be a single Pattern, not a list of patterns.
Pattern string // "./...", "./types/key"
// Tests is a list of Tests to run. If empty, all Tests in the package are
// run.
Tests []string // ["TestFoo", "TestBar"]
// IssueURLs maps from a test name to a URL tracking its flake.
IssueURLs map[string]string // "TestFoo" => "https://github.com/foo/bar/issue/123"
}
type goTestOutput struct {
Time time.Time
Action string
Package string
Test string
Output string
}
var debug = os.Getenv("TS_TESTWRAPPER_DEBUG") != ""
// runTests runs the tests in pt and sends the results on ch. It sends a
// testAttempt for each test and a final testAttempt per pkg with pkgFinished
// set to true. Package build errors will not emit a testAttempt (as no valid
// JSON is produced) but the [os/exec.ExitError] will be returned.
// It calls close(ch) when it's done.
func runTests(ctx context.Context, attempt int, pt *packageTests, goTestArgs, testArgs []string, ch chan<- *testAttempt) error {
defer close(ch)
args := []string{"test"}
args = append(args, goTestArgs...)
args = append(args, pt.Pattern)
if len(pt.Tests) > 0 {
runArg := strings.Join(pt.Tests, "|")
args = append(args, "--run", runArg)
}
args = append(args, testArgs...)
args = append(args, "-json")
if debug {
fmt.Println("running", strings.Join(args, " "))
}
cmd := exec.CommandContext(ctx, "go", args...)
if len(pt.Tests) > 0 {
cmd.Env = append(os.Environ(), "TS_TEST_SHARD=") // clear test shard; run all tests we say to run
}
r, err := cmd.StdoutPipe()
if err != nil {
log.Printf("error creating stdout pipe: %v", err)
}
defer r.Close()
cmd.Stderr = os.Stderr
cmd.Env = os.Environ()
cmd.Env = append(cmd.Env, fmt.Sprintf("%s=%d", flakytest.FlakeAttemptEnv, attempt))
if err := cmd.Start(); err != nil {
log.Printf("error starting test: %v", err)
os.Exit(1)
}
s := bufio.NewScanner(r)
resultMap := make(map[string]map[string]*testAttempt) // pkg -> test -> testAttempt
for s.Scan() {
var goOutput goTestOutput
if err := json.Unmarshal(s.Bytes(), &goOutput); err != nil {
if errors.Is(err, io.EOF) || errors.Is(err, os.ErrClosed) {
break
}
// `go test -json` outputs invalid JSON when a build fails.
// In that case, discard the the output and start reading again.
// The build error will be printed to stderr.
// See: https://github.com/golang/go/issues/35169
if _, ok := err.(*json.SyntaxError); ok {
fmt.Println(s.Text())
continue
}
panic(err)
}
pkg := goOutput.Package
pkgTests := resultMap[pkg]
if pkgTests == nil {
pkgTests = make(map[string]*testAttempt)
resultMap[pkg] = pkgTests
}
if goOutput.Test == "" {
switch goOutput.Action {
case "start":
pkgTests[""] = &testAttempt{start: goOutput.Time}
case "fail", "pass", "skip":
for _, test := range pkgTests {
if test.testName != "" && test.outcome == "" {
test.outcome = "fail"
ch <- test
}
}
ch <- &testAttempt{
pkg: goOutput.Package,
outcome: goOutput.Action,
start: pkgTests[""].start,
end: goOutput.Time,
pkgFinished: true,
}
}
continue
}
testName := goOutput.Test
if test, _, isSubtest := strings.Cut(goOutput.Test, "/"); isSubtest {
testName = test
if goOutput.Action == "output" {
resultMap[pkg][testName].logs.WriteString(goOutput.Output)
}
continue
}
switch goOutput.Action {
case "start":
// ignore
case "run":
pkgTests[testName] = &testAttempt{
pkg: pkg,
testName: testName,
start: goOutput.Time,
}
case "skip", "pass", "fail":
pkgTests[testName].end = goOutput.Time
pkgTests[testName].outcome = goOutput.Action
ch <- pkgTests[testName]
case "output":
if suffix, ok := strings.CutPrefix(strings.TrimSpace(goOutput.Output), flakytest.FlakyTestLogMessage); ok {
pkgTests[testName].isMarkedFlaky = true
pkgTests[testName].issueURL = strings.TrimPrefix(suffix, ": ")
} else {
pkgTests[testName].logs.WriteString(goOutput.Output)
}
}
}
if err := cmd.Wait(); err != nil {
return err
}
if err := s.Err(); err != nil {
return fmt.Errorf("reading go test stdout: %w", err)
}
return nil
}
func _main() {
goTestArgs, packages, testArgs, err := splitArgs(os.Args[1:])
if err != nil {
log.Fatal(err)
return
}
if len(packages) == 0 {
fmt.Println("testwrapper: no packages specified")
return
}
ctx := context.Background()
type nextRun struct {
tests []*packageTests
attempt int // starting at 1
}
firstRun := &nextRun{
attempt: 1,
}
for _, pkg := range packages {
firstRun.tests = append(firstRun.tests, &packageTests{Pattern: pkg})
}
toRun := []*nextRun{firstRun}
printPkgOutcome := func(pkg, outcome string, attempt int, runtime time.Duration) {
if outcome == "skip" {
fmt.Printf("?\t%s [skipped/no tests] \n", pkg)
return
}
if outcome == "pass" {
outcome = "ok"
}
if outcome == "fail" {
outcome = "FAIL"
}
if attempt > 1 {
fmt.Printf("%s\t%s\t%.3fs\t[attempt=%d]\n", outcome, pkg, runtime.Seconds(), attempt)
return
}
fmt.Printf("%s\t%s\t%.3fs\n", outcome, pkg, runtime.Seconds())
}
for len(toRun) > 0 {
var thisRun *nextRun
thisRun, toRun = toRun[0], toRun[1:]
if thisRun.attempt > maxAttempts {
fmt.Println("max attempts reached")
os.Exit(1)
}
if thisRun.attempt > 1 {
j, _ := json.Marshal(thisRun.tests)
fmt.Printf("\n\nAttempt #%d: Retrying flaky tests:\n\nflakytest failures JSON: %s\n\n", thisRun.attempt, j)
}
toRetry := make(map[string][]*testAttempt) // pkg -> tests to retry
for _, pt := range thisRun.tests {
ch := make(chan *testAttempt)
runErr := make(chan error, 1)
go func() {
defer close(runErr)
runErr <- runTests(ctx, thisRun.attempt, pt, goTestArgs, testArgs, ch)
}()
var failed bool
for tr := range ch {
// Go assigns the package name "command-line-arguments" when you
// `go test FILE` rather than `go test PKG`. It's more
// convenient for us to to specify files in tests, so fix tr.pkg
// so that subsequent testwrapper attempts run correctly.
if tr.pkg == "command-line-arguments" {
tr.pkg = packages[0]
}
if tr.pkgFinished {
if tr.outcome == "fail" && len(toRetry[tr.pkg]) == 0 {
// If a package fails and we don't have any tests to
// retry, then we should fail. This typically happens
// when a package times out.
failed = true
}
printPkgOutcome(tr.pkg, tr.outcome, thisRun.attempt, tr.end.Sub(tr.start))
continue
}
if testingVerbose || tr.outcome == "fail" {
io.Copy(os.Stdout, &tr.logs)
}
if tr.outcome != "fail" {
continue
}
if tr.isMarkedFlaky {
toRetry[tr.pkg] = append(toRetry[tr.pkg], tr)
} else {
failed = true
}
}
if failed {
fmt.Println("\n\nNot retrying flaky tests because non-flaky tests failed.")
os.Exit(1)
}
// If there's nothing to retry and no non-retryable tests have
// failed then we've probably hit a build error.
if err := <-runErr; len(toRetry) == 0 && err != nil {
var exit *exec.ExitError
if errors.As(err, &exit) {
if code := exit.ExitCode(); code > -1 {
os.Exit(exit.ExitCode())
}
}
log.Printf("testwrapper: %s", err)
os.Exit(1)
}
}
if len(toRetry) == 0 {
continue
}
pkgs := slicesx.MapKeys(toRetry)
sort.Strings(pkgs)
nextRun := &nextRun{
attempt: thisRun.attempt + 1,
}
for _, pkg := range pkgs {
tests := toRetry[pkg]
slices.SortFunc(tests, func(a, b *testAttempt) int { return strings.Compare(a.testName, b.testName) })
issueURLs := map[string]string{} // test name => URL
var testNames []string
for _, ta := range tests {
issueURLs[ta.testName] = ta.issueURL
testNames = append(testNames, ta.testName)
}
nextRun.tests = append(nextRun.tests, &packageTests{
Pattern: pkg,
Tests: testNames,
IssueURLs: issueURLs,
})
}
toRun = append(toRun, nextRun)
}
}
|