mirror of https://github.com/apache/activemq.git
4bace216b9
Instead of locking on the PList iterator which forces a lock on every access to hasNext we should lock down at the PageFile level in readPage making the seek and read an atomic operation. Since many reads on a single already loaded page don't need to be synchronized we pay a higher price for locking in the specific PList case than we do buy creating this atomic read operation in PageFile. git-svn-id: https://svn.apache.org/repos/asf/activemq/trunk@1375852 13f79535-47bb-0310-9956-ffa450edef68 |
||
|---|---|---|
| .. | ||
| src | ||
| README | ||
| pom.xml | ||
README
=======================================================================
KahaDB : An Embedded Lightweight Non-Relational Database
=======================================================================
Goals:
* Provide a simple collections based interface for storing persistent
data.
* Provide a lower level key value interface for storing data. Secondary
keys should be supported.
* Fast recovery
Status:
* Moved the org.apache.activemq.kaha package into this module
and renamed it as org.apache.kahadb
* Moved a refactored version of the low level indexing logic
used by Apache xindice to the org.apache.kahadb.xindice page.
TODO:
* None of our current index support fast recovery. They suffer
from partial page write failures so on a crash, they have
to be fully rebuilt from the data logs.
* Indexes should know what the last transaction id that updated
it, so that missing operations can be replayed.
* Investigate implementing data log replication for HA.