Francesco Mari created OAK-4649:
-----------------------------------
Summary: Decouple segments and indexes from TAR files
Key: OAK-4649
URL: https://issues.apache.org/jira/browse/OAK-4649
Project: Jackrabbit Oak
Issue Type: Improvement
Components: segment-tar
Reporter: Francesco Mari
Assignee: Francesco Mari
TAR files currently embed three indexes: an index of the segments contained in
the TAR files, a graph index and an index of external binary references.
Index files are checked for consistency purposes at the startup of the system.
Normally, if an index file is corrupted it is recreated. Since the index file
is contained inside the TAR file, recreating them implies rewriting the whole
TAR file and appending the new index.
This process creates unnecessary backups, since the biggest part of the TAR
file is effectively immutable. Moreover, because index files are stored in the
TAR files, we can't treat TAR files as true read-only files. There is always
the possibility that they have to be opened again in write mode for the
recovery of the index file.
I propose to move those index files outside of the TAR files. TAR files will
end up being truly read-only files containing immutable data, and index files
will be granted their own physical files on the file system. Being index files
derived data, they now can be recreated at will without impacting the read-only
part of the segment store.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
