This implementation makes sure that the ringbuffers time stays aligned with running_time of the pipeline. In cases of skews, it will realign the ringbuffer_time to the running_time, without doing any dropping of buffers, only retimestamping. It will always align the last buffer written into the ringbuffer (from the device) to running_time, so any buffers already in the ringbuffer will then be "late", but I think this is correct behavior, since elements downstream might choose to drop these buffers, but it is not the job of the ringbuffer to do so.