DISTRIBUTION OF AN OBJECT IN VOLATILE MEMORY ACROSS A MULTI-NODE DATABASE

摘要

Techniques are described herein for distributing distinct portions of a database object across the volatile memories of a plurality of nodes in a clustered database system. The techniques involve establishing a single database server instance located on a node in a multi-node cluster as a load-operation master for a particular data set. The load-operation master determines how the data set may be separated into chunks using a hash function. The load-operation master then broadcasts a small payload of consistency information to other database servers, so each database server may independently execute the hash function and independently load their respectively assigned chunks of data.

主权项

1. A method comprising:
establishing a single database server instance, from a plurality of database server instances, as a load-operation master for a particular set of data; wherein each database server instance of the plurality of database server instances is executing on corresponding node of a plurality of nodes in a multi-node cluster; wherein each database server instance of the plurality of database server instances has shared access to a single on-disk version of the particular set of data; determining, by the single database server instance, how to divide the particular set of data into a plurality of chunks; broadcasting, by the single database server instance, a message to other database server instances in the multi-node cluster; wherein the message includes data that defines the plurality of chunks; based, at least in part, on the data that defines the plurality of chunks, each given database server instance of the plurality of database server instances independently performing the steps of:
creating, in volatile memory that is local to the given database server instance, a mapping that indicates a host node for each chunk of the plurality of chunks; andfor each target chunk, of the plurality of chunks, that is to be hosted by the node that is executing the given database server instance, the given database server instance loading data from the target chunk into volatile memory that is local to the given database server instance; wherein the host node for each chunk is determined by applying a particular hash function to a unique identifier of the chunk.