[CI] not ok 135 runc run with tmpfs perms

[lifubang]

ci / test (ubuntu-20.04, 1.21.x, rootless, -race) (pull_request) Failing after 3m

not ok 135 runc run with tmpfs perms
# (in test file tests/integration/run.bats, line 119)
#   `[ "$output" = "710" ]' failed
# runc spec --rootless (status=0):
#
# runc run test_tmpfs (status=0):
# 444
# runc run test_tmpfs (status=0):
# 444
# runc run test_tmpfs (status=0):
# 710

[lifubang]

https://github.com/opencontainers/runc/actions/runs/6639817702/job/18038954312
It happended again. I have seen it several times.

ok 139 runc run with tmpfs # skip test requires root
not ok 140 runc run with tmpfs perms
# (in test file tests/integration/run.bats, line 104)
#   `[ "$output" = "444" ]' failed
# runc spec --rootless (status=0):
#
# runc run test_tmpfs (status=0):
# 444

[kolyshkin]

I think this is a manifestation of the same issue as in #2425 (a race in runc tty driver). I tried to solve it a few years ago (see #2723), albeit unsuccessfully (see #2749).

Maybe someone else can untangle this knot; I really tried.

As for the test case, a fix is simple, see #4095.

[kolyshkin]

I think this is a manifestation of the same issue as in #2425 (a race in runc tty driver). I tried to solve it a few years ago (see #2723), albeit unsuccessfully (see #2749).

Maybe someone else can untangle this knot; I really tried.

@lifubang if you want to give it a try, the repro is really simple:

1. Remove the t.Skip from the TestExecInTTY:

diff --git a/libcontainer/integration/execin_test.go b/libcontainer/integration/execin_test.go
index c5c32413..43eebeaf 100644
--- a/libcontainer/integration/execin_test.go
+++ b/libcontainer/integration/execin_test.go
@@ -234,7 +234,7 @@ func TestExecInTTY(t *testing.T) {
        if testing.Short() {
                return
        }
-       t.Skip("racy; see https://github.com/opencontainers/runc/issues/2425")
+       //      t.Skip("racy; see https://github.com/opencontainers/runc/issues/2425")
        config := newTemplateConfig(t, nil)
        container, err := newContainer(t, config)
        ok(t, err)

2. Run the test many times:

$ cd libcontainer/integration
$ go test -v -count 1000 -exec sudo -run ExecInTTY

Eventually you will see it failing.

I guess another way is to do runc run with a short-lived process (same as in "runc run with tmpfs perms"), as this is what we are seeing in CI.

[kolyshkin]

Another manifestation of the same bug: #4074