Multihost Database Host Removal Without Replica Redistribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Elastic cloud database systems face inefficiencies in scale-in operations due to the time-consuming redistribution of data when removing hosts that store replica tables, offsetting the performance gains from scale-out operations.
Innovation Solution
A host removal shortcut technique that determines whether a host stores only replica tables without obtaining landscape information, allowing for immediate removal without redistribution, thereby reducing processing and time spent on host removal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data redistribution is performed during host removal, then data integrity is maintained, but removal time increases significantly
Solution Approach 1:
The patent segments host removal operations into two distinct paths: (1) hosts storing source tables require landscape information gathering and data redistribution to maintain integrity, while (2) hosts storing only replica tables can be removed directly without redistribution. This segmentation allows the system to apply different removal procedures based on the host's role, eliminating unnecessary redistribution operations for replica-only hosts and significantly reducing their removal time while preserving data integrity for hosts that actually hold source data.
Solution Approach 2:
The patent introduces a host-type identifier parameter that changes the removal procedure. By checking this parameter, the system determines whether a host stores source tables or only replica tables, and accordingly selects the appropriate removal path. This parameter-driven approach enables conditional execution of landscape information gathering and data redistribution, optimizing the removal process by avoiding these operations when they are not necessary for data integrity.
2Measurement precision
If landscape information is obtained for each host system, then accurate data distribution is ensured, but processing overhead increases
Solution Approach 1:
The patent segments the host population into two categories based on their data storage role: hosts with source tables and hosts with only replica tables. This segmentation allows the system to apply landscape information gathering selectively - only to hosts that actually require it for accurate data distribution. Replica-only hosts are identified through their host-type identifier and excluded from the landscape information gathering process, eliminating unnecessary processing overhead while maintaining data distribution accuracy for the relevant subset of hosts.
Solution Approach 2:
Instead of obtaining landscape information for all hosts (excessive action), the patent applies this operation only to the subset of hosts that store source tables (partial action). This selective approach reduces the overall processing overhead significantly, especially in systems with many replica-only hosts, while still ensuring accurate data distribution where it is actually needed.
3Reliability
If data redistribution is performed during scale-in operations, then data consistency is maintained, but the time offset gains from scale-out operations
Solution Approach 1:
The patent applies segmentation to scale-in operations by categorizing hosts into those requiring data redistribution (source table hosts) and those that do not (replica-only hosts). This allows the system to perform rapid removal of replica-only hosts during scale-in operations, immediately realizing the performance gains from scaling out, while maintaining data consistency only for the subset of hosts where source tables are actually redistributed. This selective approach eliminates the time penalty that would otherwise offset the benefits of scale-out operations.
Solution Approach 2:
The host-type identifier parameter drives the conditional execution of data redistribution during scale-in operations. By evaluating this parameter, the system determines whether to perform the time-consuming redistribution operation or proceed directly with host removal. This parameter-based decision-making enables the system to maintain data consistency where necessary while achieving rapid scaling efficiency when replica-only hosts are involved, thereby preserving the productivity benefits of elastic scaling operations.
Data Source
AI summary
The present disclosure provides more efficient techniques for removing a host from a multi-host database system. An instruction to remove a host system may be received. In response, a determination of whether the first host system does or does not store any source tables is made based on a host-type identifier for the host system. This determination may not require obtaining landscape information for each of the hosts in the database system. If the host system stores replica tables and does not store source tables, those replica tables may be dropped based on the determination that the first host system does not store any source tables. As such, in cases where table redistribution is not needed the landscape information is not obtained, thereby making the host removal process more efficient.