Execution Tuning
Execution tuning addresses performance bottlenecks exposed at runtime: improper parallelism, RuntimeFilters that do not take effect in time, Shuffle data skew, and scenarios where the optimizer's choice requires manual intervention. First locate the symptom in the Profile, then choose the corresponding tuning method.
Runtime Optimization
Parallelism Tuning
Provides recommendations for setting `parallel_pipeline_task_num` across single-table scan, JOIN, aggregation, and high-concurrency scenarios, balancing query latency with scheduling overhead.
Adjustment of RuntimeFilter Wait Time
Identify the RuntimeFilter NOT_READY state from the Profile, and adjust `runtime_filter_wait_time_ms` so that the Probe side actually benefits from the filtering effect.
Data Skew Handling
Identify single-point execution bottlenecks caused by Join Key or bucket distribution, and rebalance Shuffle with techniques such as Broadcast Join and Leading Hint.
