The Zope Object Database (ZODB, ZopeDB) is an object-oriented DBMS library for storing Python objects. ZODB aims to provide a transparent way for Python programmers to make their data persistent. It is completely written by Python, and only provides Python API. It does not support other programming languages. ZODB stores data as objects, and the relationship between objects are automatically maintained by Python object references. As a result, ZODB allows programmers to manage data structures with complicate relationships. The storage layer of ZODB is pluggable, and there are three main implementations: ZEO, NEO, and RelStorage. ZEO is a single-server client-server framework using FileStorage as the backend; NEO extends ZEO to distributed-server client-server storage; RelStorage stores objects in a relational database backend, such as MySQL, PostgreSQL, and Oracle. ZODB uses aggressive caching. As a result, it is not suitable for tasks with heavy writes. Also, ZODB is mainly designed for mapping access through keys and attributes of objects, which means that it does not support queries containing joins and advanced search.
ZODB was a commercial project dated back to the late 90s, which was a part of the Zope web application server. It was made open source as ZODB1 in 2002 and could be used independently. It upgraded to ZODB2 and then to ZODB3 in the same year, 2002. ZODB3 was written in Python and C. It lived until 2012, and then the developers decided to rebuild it into ZODB4 using 100% Python. In 2016, ZODB4 was renamed to ZODB5, which is the current version and is active.
ZODB does not provide ckeckpoint API.
FileStorage, the backend of the default storage layer ZEO, does not compress the data. It simply appends objects to the end of a file. However, there is an optional layer named zlibstorage which uses [gzip](http://www.gzip.org/) to compress the file in FileStorage. Moreover, since ZODB has a pluggable storage layer, it can support compression by using different storage layers such as NEO and relStorage. NEO implements client-side compression, and RelStorage supports compression using MySQL, PostgreSQL, and Oracle backend.
ZODB implements MVCC as an additional layer for those pluggable storage layers without MVCC, such as ZEO and NEO. ZODB does not implement MVCC layer for relStorage, which already has MVCC.
ZODB does not support building indexes on a set of stored objects. However, there is a variable type named Btree functioning as an index in ZODB. If objects are inserted into a Btree, they are maintained in order, and range queries are supported. Btree has variants: IOBTrees, OIBTrees, and IIBTrees, which are optimized for integer keys and integer values.
The relationships between objects are automatically maintained by Python object reference, which is a key internal mechanism of Python.
FileStorage, the backend of the default storage layer ZEO, is itself a log-structured file. So there is no need to implement logging in ZODB. For other alternative storage layers, whether they have logging or not is determined by the implementation of their backend.
ZODB only provides Python APIs. Querying for an object in ZODB is similar to querying for a value in a Python dictionary.
ZODB has pluggable storage layers, including ZEO, NEO, relstorage, Directory Storage, Demo Storage, and so on. Demo Storage is used for development and testing, so it is a in-memory storage for convenient. Others are disk-oriented.
It depends on which pluggable storage layer is used. FileStorage, the backend of the default storage layer ZEO, is a log-structured storage layer, which has only one large file for all data. Demo Storage is a copy-on-write storage layer.
ZODB is used as a Python library. Therefore, Python functions can be viewed as stored procedures for ZODB.
It depends on which pluggable storage layer is used. The default storage layer ZEO is a single-server model, so it shares everything. There are also distributed-server models written for ZODB, such as NEO and ZRS. NEO and ZRS store data on multiple machines across the Internet, and support data duplication, load-balancing read and fault tolerance. They share the data in the disk.