Hi team, first PR to StarRocks here, would love a review when someone has time. In shared-data mode, TABLESAMPLE / SAMPLE('percent'='N') on lake tables doesn't actually take effect. From what I can see, FE parses it fine but TLakeScanNode is missing sample_options, so the info never reaches BE. End result: ANALYZE SAMPLE TABLE does a full scan on each selected tablet, which I think is what's behind the heavy IO some people see after large INSERTs. What the PR does: wires sample_options through the lake path, the same way the shared-nothing path already does it (Thrift → FE → lake_connector → tablet_reader → rowset). On a big table in my env, ANALYZE SAMPLE went from ~2 min to ~8 sec. https://github.com/StarRocks/starrocks/pull/71874 Fixes #71873. Please correct me if I'm missing context anywhere!