Hash Join Runtime Filter Selection Using Build-Phase Table Statistics

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Hash Join queries in SQL databases face performance degradation due to inaccurate determination of when to use Runtime filters, leading to increased resource consumption and decreased query performance, especially in distributed systems where global estimation inaccuracies are amplified.

Innovation Solution

A method to determine the suitability of Runtime filters by using actual table statistical information from the hash data table constructed during the build phase, allowing for accurate assessment of performance improvement and consumption, thereby optimizing Hash Join queries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If Runtime filter is applied to all Hash Join queries, then query performance is improved for suitable cases, but resource consumption increases and performance deteriorates for unsuitable cases

Engineering Contradiction:
Improvequery performanceVSAvoidresource consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent changes the parameter of filter applicability from a fixed blanket application to a dynamic decision based on estimated data distribution parameters (selectivity, correlation coefficient). By evaluating these parameters and comparing against thresholds, the system adapts the Runtime filter application to match the actual data characteristics, applying the filter only when parameters indicate it will be beneficial.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces a feedback mechanism where the system estimates data distribution parameters, evaluates whether Runtime filter applicability conditions are met, and adjusts its behavior accordingly. The cost model provides feedback about the expected performance impact, allowing the system to make informed decisions about filter application rather than blindly applying it to all queries.

Inventive Principle:
Principle #23Feedback

2Reliability

If Runtime filter is applied blindly without evaluation, then optimization effect is achieved for suitable queries, but additional performance consumption and system burden occur for unsuitable queries

Engineering Contradiction:
Improveoptimization effectVSAvoidquery performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent performs preliminary evaluation of data distribution parameters (selectivity, correlation coefficient) before applying the Runtime filter. By estimating these parameters in advance and comparing them against predefined thresholds, the system determines in advance whether the filter is likely to be beneficial, preventing performance degradation from premature or inappropriate filter application.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent transforms the decision-making process by introducing parameter-based evaluation (selectivity, correlation coefficient) that changes the system's behavior from blind application to conditional application. The parameters serve as indicators that trigger different optimization strategies based on the actual data characteristics.

Inventive Principle:
Principle #35Parameter changes

3Ease of operation

If global estimation is used in distributed systems, then Runtime filter determination is simplified, but inaccuracy is amplified across multiple computing nodes

Engineering Contradiction:
Improvefilter determination simplicityVSAvoidestimation accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent segments the estimation process by computing data distribution parameters independently at each computing node based on local data characteristics. Instead of relying on a single global estimation that gets amplified across nodes, each node performs its own parameter evaluation (selectivity, correlation coefficient) on its local data, ensuring accurate local decisions without propagation of global estimation errors.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4488842B1Query method for data table in database, apparatus, computer device and storage medium
Publication Date: 2026.03.04 BEIJING VOLCANO ENGINE TECH CO LTD
  • EP4488842B1 patent drawingFigure 1~2
  • EP4488842B1 patent drawingFigure 3a~4
  • EP4488842B1 patent drawingFigure 5

AI summary

The present disclosure provides a query method for a data table in a database, an apparatus, a computer device and a storage medium. The method includes: acquiring a data table query instruction, the data table query instruction is used to perform a Hash Join query operation on target join columns of a plurality of data tables in the database; constructing, in a build phase of the Hash Join, a hash data table according to tabular data corresponding to a target join column of a first data table among the plurality of data tables; determining, based on table statistical information of the hash data table, whether to construct a runtime filter for filtering a second data table in a probe phase of the Hash Join; and if so, determining a data query result with the constructed runtime filter based on the hash data table and the second data table.