v4 回填 (M3): 每部分分块不会占用内存 — full_sorting_merge 会对输入/输出进行排序

SELECT project_id, trace_id FROM dataset_run_items_rmt ) SETTINGS

• 复制

     join_algorithm = 'full_sorting_merge',

• 复制

     -- 'hash', not a sorting algorithm, because the two sides of this join

• 复制

     -- are asymmetric. The right side is one partition of traces, deduped to

• 复制

     -- one row per (project_id, id) and carrying

…