chore(test): wait for mender-connect to exit before restarting in set_limits

The set_limits() helper kills mender-connect and immediately starts a
new instance without waiting for the old process to exit. This creates
a race condition: the old process may still be running (and its WebSocket
still registered as "connected" in deviceconnect) when the new one
starts. wait_for_connect() can then pass on the stale "connected" state
from the old connection, and the subsequent file transfer request hits
an inconsistent state timing out with HTTP 408.

Signed-off-by: pasinskim <marcin.pasinski@northern.tech>

Author: pasinskim

tests/tests/test_filetransfer.py

@@ -107,7 +107,17 @@ def set_limits(docker_env, mender_device, limits, auth, devid):
         )
     finally:
         shutil.rmtree(tmpdir)
-    mender_device.run("kill -TERM `pidof %s`" % connect_service_name)
+    # Stop the respawn loop in the container entrypoint so it doesn't
+    # restart mender-connect automatically after we kill it — otherwise
+    # we'd end up with two daemons running.
+    mender_device.run(
+        "touch /stop-respawn; "
+        "kill -TERM `pidof %s` 2>/dev/null; "
+        "timeout 10 sh -c 'while pidof %s >/dev/null 2>&1; do sleep 0.1; done' || "
+        "kill -9 `pidof %s` 2>/dev/null; "
+        "rm -f /stop-respawn"
+        % (connect_service_name, connect_service_name, connect_service_name)
+    )
     docker_env._docker_compose_cmd(
         "exec -d mender-client %s daemon" % connect_service_name
     )