The migration_with_exec test is failing sporadically for all
architectures due to a race when the destination socat process takes
too long to start listening while the source process is already
issuing connect().
The race is inherent because the exec: migration spawns the
to-be-exec'ed command asynchronously and returns from the
migrate-incoming command. The localhost-only testcase is not
representative of the majority of migrations. In a real scenario
between two different hosts that race wouldn't happen.
Fix the testcase by configuring the source socat command to wait
indefinitely while trying to connect.
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Link: https://lore.kernel.org/qemu-devel/20260422230001.3168-1-farosas@suse.de
Signed-off-by: Fabiano Rosas <farosas@suse.de>
with Ports() as ports:
free_port = self._get_free_port(ports)
dst_uri = 'exec:socat TCP-LISTEN:%u -' % free_port
- src_uri = 'exec:socat - TCP:localhost:%u' % free_port
+ src_uri = 'exec:socat - TCP:localhost:%u,forever' % free_port
self.migrate(dst_uri, src_uri)