| Message ID | 1490299158-12933-4-git-send-email-adrian@lisas.de |
|---|---|
| State | Accepted |
| Series | "zdtm: also dump and restore tests with --check-only" |
| Commit | f915df11ca5e502836d9a4ee4df30b604cc5e180 |
| Headers | show |
diff --git a/test/zdtm.py b/test/zdtm.py index 57e2ab0..8d38e05 100755 --- a/test/zdtm.py +++ b/test/zdtm.py @@ -329,6 +329,13 @@ def wait_pid_die(pid, who, tmo = 30): def test_flag(tdesc, flag): return flag in tdesc.get('flags', '').split() + +def reset_pid(pid = 1): + fd = open('/proc/sys/kernel/ns_last_pid', 'w') + fd.write('%s' % pid) + fd.close() + + # # Exception thrown when something inside the test goes wrong, # e.g. test doesn't start, criu returns with non zero code or
On Thu, Mar 23, 2017 at 07:59:16PM +0000, Adrian Reber wrote: > From: Adrian Reber <areber@redhat.com> > > In preparation for the zdtm option '--check-only' a new helper function > reset_pid() is added which writes to ns_last_pid to avoid PID collisions > during check-only restore and the real restore. > > Signed-off-by: Adrian Reber <areber@redhat.com> > --- > test/zdtm.py | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/test/zdtm.py b/test/zdtm.py > index 57e2ab0..8d38e05 100755 > --- a/test/zdtm.py > +++ b/test/zdtm.py > @@ -329,6 +329,13 @@ def wait_pid_die(pid, who, tmo = 30): > def test_flag(tdesc, flag): > return flag in tdesc.get('flags', '').split() > > + > +def reset_pid(pid = 1): I don't understand how pid = 1 helps to avoid collisions. We create a new pid namespace for each test (test/zdtm_ct) and we know that pids are allocated sequentialy, so test process uses "small" pids. In this case setting 1 into ns_last_pid has to increate probability of collisions... > + fd = open('/proc/sys/kernel/ns_last_pid', 'w') > + fd.write('%s' % pid) > + fd.close() > + > + > # > # Exception thrown when something inside the test goes wrong, > # e.g. test doesn't start, criu returns with non zero code or > -- > 1.8.3.1 > > _______________________________________________ > CRIU mailing list > CRIU@openvz.org > https://lists.openvz.org/mailman/listinfo/criu
On Thu, Mar 23, 2017 at 05:10:12PM -0700, Andrei Vagin wrote: > On Thu, Mar 23, 2017 at 07:59:16PM +0000, Adrian Reber wrote: > > From: Adrian Reber <areber@redhat.com> > > > > In preparation for the zdtm option '--check-only' a new helper function > > reset_pid() is added which writes to ns_last_pid to avoid PID collisions > > during check-only restore and the real restore. > > > > Signed-off-by: Adrian Reber <areber@redhat.com> > > --- > > test/zdtm.py | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/test/zdtm.py b/test/zdtm.py > > index 57e2ab0..8d38e05 100755 > > --- a/test/zdtm.py > > +++ b/test/zdtm.py > > @@ -329,6 +329,13 @@ def wait_pid_die(pid, who, tmo = 30): > > def test_flag(tdesc, flag): > > return flag in tdesc.get('flags', '').split() > > > > + > > +def reset_pid(pid = 1): > > I don't understand how pid = 1 helps to avoid collisions. We create a > new pid namespace for each test (test/zdtm_ct) and we know that pids are > allocated sequentialy, so test process uses > "small" pids. In this case setting 1 into ns_last_pid has to increate > probability of collisions... The following tests fail in --check-only mode if I do not reset the PID back to 1: * zdtm/static/zombie00(h) * zdtm/static/sched_prio00(h) * zdtm/static/pthread00(h) * zdtm/static/sigpending(h) * zdtm/static/fdt_shared(h) * zdtm/transition/epoll(h) * zdtm/transition/fork(h) * zdtm/transition/fork2(h) * zdtm/transition/maps008(h) * zdtm/transition/pipe_loop00(h) * zdtm/transition/socket_loop00(h) * zdtm/transition/unix_sock(h) * zdtm/transition/fifo_loop(h) * zdtm/transition/pipe_shared00(h) * zdtm/transition/file_read(h) The test restore and the real restore are running in the same PID namespace. For those testcases the real restore fails with a PID collision: (00.004419) Error (criu/cr-restore.c:1353): Pid 25 do not match expected 24 Resetting it to '1' helps. Adrian