Section offsets are a subpropery of the data-parsoid but we're storing it both in data-parsoid and separately, in sections.offsets table in Cassandra. The original idea was to optimize the reading of the sections so that the full data-parsoid structure shouldn't be read from storage and parsed to extract the offsets. However, I believe that was a very premature optimization. The sections endpoints are by far not the most popular, fetching and parsing the data-parsoid will probably take some negligible amount of time compared to everything else that should be done. On the other hand, removing it will simplify the code and given that we're short on IO in Cassandra, saving 1/5th of the writes for Parsoid content could be a decent win.
What do you think?