Dynamically indentifying and preventing skewed partitions in a shared-nothing database

摘要

Dynamically identifying and preventing skewed partitions in a shared-nothing database is provided. The database management system software receives a parameter for identifying a threshold value associated with at least one distribution key value. Optimizer statistics are gathered on a first table that is distributed across one or more partitions in the shared-nothing database, wherein the first table includes a first table name. Distribution key skew is identified based on the gathered optimizer statistics indicating the threshold value being exceeded. A second table with an alternate distribution key, is created having a second table name for receiving overflow data rows associated with the at least one distribution key value based on the identified distribution key skew. A union all view is created based on the first and the second table.

主权项

1. A computer program product for dynamically identifying and preventing skewed partitions in a shared-nothing database comprising:
a computer readable storage medium readable by a processing circuit and storing instructions for execution by the processing circuit for performing a method comprising: receiving a parameter for identifying a threshold value associated with at least one distribution key value; gathering optimizer statistics on an original table distributed across one or more partitions in the shared-nothing database, wherein the original table includes an original table name; identifying distribution key skew based on the gathered optimizer statistics indicating the threshold value being exceeded; creating an overflow table for receiving overflow data rows associated with the at least one distribution key value based on the identified distribution key skew, the created overflow table having an alternate distribution key and an overflow table name; creating a union all view based on the original table and the overflow table, wherein the creating of the union all view comprises:
renaming the original to an other table name;assigning the original table name to the union all view; andapplying a check constraint to the distribution key column and the overflow indicator column in the original table and in the overflow table.