ODB (a.k.a. "spugdb" due to a name clash on code.google.com) is a pure-python key-value style database system. It supports nested transactions, object schemas, cursors and a simple, X-Path like query language.
ODB is a good fit for situations where you have relatively small storage requirements, low throughput and need a lightweight pure-python database. In simple benchmarking scenarios on modern CPUs and local filesystems, it can handle on the order of 1000 general operations per second, or 300 insertions per seecond. Tables larger than 100 megabytes have not been tested.
ODB started out as a front-end to BDB (Berkeley DB). After the author encountered problems with BDB, two additional pure-python storage engines were written. At this time, the primary interface has been rewritten and the older BDB-style interface (while still available) is deprecated.
At this point, ODB is fairly mature. It is being used in a number of
projects (several personal projects by the author and two systems at Google).
Links
Added a new interface for the system. The BDB, Mem and Full interfaces are now deprecated.
Added the "model" system, which allows you to define object schemas mapping python objects to a set of database tables and indeces.
Abstracted out the low-level storage system (allowing storage to non-filesystems and backups after checkpointing)
Fixed bug in free node lists.
Added support for full recovery from logfiles.
Switched license to LGPLv3.
Re-wrote the b-tree code from the ground up.
Implemented "blists" - btree implementation of a random-access sequence.
Documentation improvements.
Fixed several cases where a b-tree wasn't demand loading a node before performing some action on it.
Differentiated between a key of None and an empty key (""), allowing empty keys to be used.
Fixed the use of the key's reference count (it wasn't getting incremented when the key was reused)
When replacing a node, the original node wasn't getting marked deleted. Fixed this.
The full back-end wasn't building the dictionary of all of the tables during initialization, causing exceptions while processing the transaction log if the last log didn't also create the table. Fixed this.
Added numerous test cases to exercise all of these fixes, including an "update" action in the test of 1000 random actions.
Initial release.