Distributed consistent database implementation within an object store

摘要

A storage system is described. The storage system includes an object store having a plurality of objects to implement a quota policy or a part of a relational database. The plurality of objects include a first plurality of immutable objects, a second plurality of immutable objects and a head object. The first plurality of immutable objects contain information kept within the database. The second plurality of immutable objects contain a hierarchy of pages for accessing the first plurality of immutable objects. The plurality of objects also include a mutable object that contains a mapping table that maps identifiers of the pages to identifiers of objects of the second plurality of immutable objects. The information is one of: quota policy information; information within a table of a relational database; a stripe of a thin provisioned file.

主权项

1. A storage system, comprising:
an object store comprising distributed hardware storage devices, said object store having a plurality of objects to implement a database, said plurality of objects comprising:
a first plurality of immutable objects containing information kept within said database, said information being one of:
quota policy information;information within a table of a relational database;a stripe of a thin provisioned file;a second plurality of immutable objects containing a hierarchy of pages for accessing said first plurality of immutable objects;a mutable object containing a mapping table that maps identifiers of said pages to identifiers of objects of said second plurality of immutable objects;wherein, an object of said first plurality of immutable objects having sought for information is identified by fetching objects of said second plurality of immutable objects from said object store through a network, loading their respective pages into hardware memory and processing the respective pages such that a later processed page is identified from an earlier processed page and an object containing the later processed page is identified from said mapping table;wherein, only a small subset of the second plurality of objects are loaded from the object store and their corresponding pages loaded into hardware memory in order to fetch the sought for information;and wherein changed pages keep their page identification to dampen ripple-up of changes upward in the hierarchy of pages;and wherein, to avoid conflicting changes being made to said database, changes to the information of the first plurality of immutable objects and changes to the pages of the second plurality of immutable objects are implemented by deleting objects having old content from the object store, creating new objects having new content and storing the new objects in the object store.