Repeatedly run flaky Go tests with stress

It runs your test hundreds of times, which can be useful for finding flaky failures.

I was trying to debug a flaky Go test, and Paul pointed me at a Go utility called stress, which is specifically designed to catch sporadic failures. It was able to reproduce my flaky test failures, and showed that my fix was incomplete.

Installing stress

I installed stress with the following command:

$ go install golang.org/x/tools/cmd/stress@latest

This created a binary stress in ~/go/bin, which I had to add to my PATH.

Using stress to run a flaky test repeatedly

Here’s a simple Go program which includes two tests: one which always passes, and one which fails on ~0.03% of runs:

package flaky

import (
	"math/rand/v2"
	"testing"
)

func TestReliable(t *testing.T) {}

func TestFlaky(t *testing.T) {
	r := rand.IntN(10000)

	if r > 9996 {
		t.Errorf("r = %d, which is too big", r)
	}
}

If you run this with go test flaky_test.go, it will almost always pass.

We can compile this test into a binary, and then run that binary using stress. It reports any failures, and every so often prints a progress report on what it’s found so far:

$ go test -o flaky.test flaky_test.go
$ stress ./flaky.test
ok  	command-line-arguments	0.003s

./T/go-stress-20250819T191229-3552286396
--- FAIL: TestFlaky (0.00s)
    flaky_test.go:14: r = 9999, which is too big
FAIL


ERROR: exit status 1


./T/go-stress-20250819T191229-2915137162
--- FAIL: TestFlaky (0.00s)
    flaky_test.go:14: r = 9998, which is too big
FAIL


ERROR: exit status 1

5s: 8809 runs so far, 2 failures (0.02%), 16 active

If your test suite contains many tests and you only want to stress one of them, you can add the -test.run flag to select a subset of tests. Here’s an example where we skip the known-reliable test:

$ stress ./flaky.test -test.run=TestFlaky