TwinDB

2010年8月14日 (六) 11:57的版本

简介

TwinDB is an embedded key/value database based on BerkeleyDB. It basically adds security from database corruption on top of the non-transactional version of BerkeleyDB. I wrote it because the transactional version of Berkeley DB is expensive ($100,000 for a commercial license).

It accomplishes this by maintaining twin BerkeleyDB databases and using "lock" files to indicate when a set of writes are in progress.

Put and delete entries are cached in memory until TwinDB_sync() is called or a read occurs (by calling TwinDB_get()). Then, it:

  1. locks the first database
  2. writes to it
  3. unlocks it 

and repeats this process for the backup database. Upon startup, it checks to see if a lock is on either file. If so, it assumes that database file is corrupt and replaces it with the other.

Cons:

   * writes are 1/2 as fast
   * database is 2x as big 

Pros:

   * the database cannot be corrupted
   * it's still *very* fast
   * it's free for commercial use
   * the size difference might not matter (current hard disks are typically over 100G, so a 500M database will still only take up less than 1% of the drive if doubled) 

相关链接

• 官方网站 http://www.dekorte.com/projects/opensource/TwinDB/