fix: avoid tasks redistribution on the errored worker node