Re: [PATCH v1] tests/qtest/migration: Fix slow test dirty_limit

[Peter Xu <peterx@redhat.com>]

On Thu, Mar 26, 2026 at 06:13:14PM -0300, Fabiano Rosas wrote:
> Peter Xu <peterx@redhat.com> writes:
> 
> > On Thu, Mar 26, 2026 at 01:44:05PM -0300, Fabiano Rosas wrote:
> >> After the referenced commit, the incoming side doesn't exit
> >> automatically after a failure. Tests that expect the destination to
> >> fail should use -incoming defer, issue QMP migrate-incoming, wait for
> >> the failure event and issue QMP quit.
> >> 
> >> Fix the dirty_limit test which wasn't updated properly.
> >> 
> >> Fixes: 4e8c4dda97 ("tests/qtest/migration: Force exit-on-error=false")
> >> Reported-by: Thomas Huth <thuth@redhat.com>
> >> Signed-off-by: Fabiano Rosas <farosas@suse.de>
> >
> > Highly likely I missed something, but could you remind me what is the
> > problem before the change, and why "defer" fixes it?
> >
> 
> The original issue was that on failure the incoming side exits abruptly
> and leaves resources in the io/channel.c and io/task.c without freeing
> and that triggers ASAN (when in use). The fix was to set
> exit-on-error=false always. -incoming defer is needed because -incoming
> uri has no means of setting the exit-on-error flag, we need an explicit
> call to migrate-incoming to be able to set the flag.
> 
> The dirty_limit test was not updated properly because it's under g_slow
> and I forgot to test it. I removed the qtest_set_expected_status(to,
> EXIT_FAILURE) line, but didn't make sure exit-on-error is set to false
> for this test. So it's now aborting with "expected 0, got 1".

Mistery resolved after I git pull looking at the new code..

Reviewed-by: Peter Xu <peterx@redhat.com>

-- 
Peter Xu